import zstd 1.3.7vendor/zstd/1.3.7

52 files changed, 5663 insertions, 2474 deletions

diff --git a/lib/BUCK b/lib/BUCK
index dbe8885fc4cf..bd93b082a347 100644
--- a/lib/BUCK
+++ b/lib/BUCK
@@ -69,6 +69,7 @@ cxx_library(
     ]),
     headers=subdir_glob([
         ('dictBuilder', 'divsufsort.h'),
+        ('dictBuilder', 'cover.h'),
     ]),
     srcs=glob(['dictBuilder/*.c']),
     deps=[':common'],
diff --git a/lib/Makefile b/lib/Makefile
index cdfdc5cdfd46..9711f75eea87 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -19,23 +19,62 @@ LIBVER := $(shell echo $(LIBVER_SCRIPT))
 VERSION?= $(LIBVER)
 
 CPPFLAGS+= -I. -I./common -DXXH_NAMESPACE=ZSTD_
+ifeq ($(OS),Windows_NT)   # MinGW assumed
+CPPFLAGS   += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
+endif
 CFLAGS  ?= -O3
-DEBUGFLAGS = -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
+DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
             -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
             -Wstrict-prototypes -Wundef -Wpointer-arith -Wformat-security \
             -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
-            -Wredundant-decls
+            -Wredundant-decls -Wmissing-prototypes
 CFLAGS  += $(DEBUGFLAGS) $(MOREFLAGS)
 FLAGS    = $(CPPFLAGS) $(CFLAGS)
 
+GREP = grep --color=never
+
+ZSTDCOMMON_FILES := $(sort $(wildcard common/*.c))
+ZSTDCOMP_FILES := $(sort $(wildcard compress/*.c))
+ZSTDDECOMP_FILES := $(sort $(wildcard decompress/*.c))
+ZDICT_FILES := $(sort $(wildcard dictBuilder/*.c))
+ZDEPR_FILES := $(sort $(wildcard deprecated/*.c))
+ZSTD_FILES := $(ZSTDCOMMON_FILES)
 
-ZSTD_FILES := $(sort $(wildcard common/*.c compress/*.c decompress/*.c dictBuilder/*.c deprecated/*.c))
+ZSTD_LEGACY_SUPPORT ?= 5
+ZSTD_LIB_COMPRESSION ?= 1
+ZSTD_LIB_DECOMPRESSION ?= 1
+ZSTD_LIB_DICTBUILDER ?= 1
+ZSTD_LIB_DEPRECATED ?= 1
 
-ZSTD_LEGACY_SUPPORT ?= 4
+ifeq ($(ZSTD_LIB_COMPRESSION), 0)
+	ZSTD_LIB_DICTBUILDER = 0
+	ZSTD_LIB_DEPRECATED = 0
+endif
+
+ifeq ($(ZSTD_LIB_DECOMPRESSION), 0)
+	ZSTD_LEGACY_SUPPORT = 0
+	ZSTD_LIB_DEPRECATED = 0
+endif
+
+ifneq ($(ZSTD_LIB_COMPRESSION), 0)
+	ZSTD_FILES += $(ZSTDCOMP_FILES)
+endif
+
+ifneq ($(ZSTD_LIB_DECOMPRESSION), 0)
+	ZSTD_FILES += $(ZSTDDECOMP_FILES)
+endif
+
+ifneq ($(ZSTD_LIB_DEPRECATED), 0)
+	ZSTD_FILES += $(ZDEPR_FILES)
+endif
+
+ifneq ($(ZSTD_LIB_DICTBUILDER), 0)
+	ZSTD_FILES += $(ZDICT_FILES)
+endif
 
 ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
 ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-	ZSTD_FILES += $(shell ls legacy/*.c | grep 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
+	ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
 endif
 	CPPFLAGS += -I./legacy
 endif
@@ -43,7 +82,7 @@ CPPFLAGS  += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
 
 ZSTD_OBJ   := $(patsubst %.c,%.o,$(ZSTD_FILES))
 
-# OS X linker doesn't support -soname, and use different extension
+# macOS linker doesn't support -soname, and use different extension
 # see : https://developer.apple.com/library/mac/documentation/DeveloperTools/Conceptual/DynamicLibraries/100-Articles/DynamicLibraryDesignGuidelines.html
 ifeq ($(shell uname), Darwin)
 	SHARED_EXT = dylib
@@ -57,8 +96,6 @@ else
 	SHARED_EXT_VER = $(SHARED_EXT).$(LIBVER)
 endif
 
-LIBZSTD = libzstd.$(SHARED_EXT_VER)
-
 
 .PHONY: default all clean install uninstall
 
@@ -74,19 +111,28 @@ libzstd.a: $(ZSTD_OBJ)
 libzstd.a-mt: CPPFLAGS += -DZSTD_MULTITHREAD
 libzstd.a-mt: libzstd.a
 
-$(LIBZSTD): LDFLAGS += -shared -fPIC -fvisibility=hidden
+ifneq (,$(filter Windows%,$(OS)))
+
+LIBZSTD = dll\libzstd.dll
 $(LIBZSTD): $(ZSTD_FILES)
 	@echo compiling dynamic library $(LIBVER)
-ifneq (,$(filter Windows%,$(OS)))
-	@$(CC) $(FLAGS) -DZSTD_DLL_EXPORT=1 -shared $^ -o dll\libzstd.dll
-	dlltool -D dll\libzstd.dll -d dll\libzstd.def -l dll\libzstd.lib
+	@$(CC) $(FLAGS) -DZSTD_DLL_EXPORT=1 -shared $^ -o $@
+	dlltool -D $@ -d dll\libzstd.def -l dll\libzstd.lib
+
 else
+
+LIBZSTD = libzstd.$(SHARED_EXT_VER)
+$(LIBZSTD): LDFLAGS += -shared -fPIC -fvisibility=hidden
+$(LIBZSTD): $(ZSTD_FILES)
+	@echo compiling dynamic library $(LIBVER)
 	@$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
 	@echo creating versioned links
 	@ln -sf $@ libzstd.$(SHARED_EXT_MAJOR)
 	@ln -sf $@ libzstd.$(SHARED_EXT)
+
 endif
 
+
 libzstd : $(LIBZSTD)
 
 libzstd-mt : CPPFLAGS += -DZSTD_MULTITHREAD
@@ -111,16 +157,16 @@ libzstd-nomt: $(ZSTD_NOMT_FILES)
 	@$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
 
 clean:
-	@$(RM) -r *.dSYM   # Mac OS-X specific
+	@$(RM) -r *.dSYM   # macOS-specific
 	@$(RM) core *.o *.a *.gcda *.$(SHARED_EXT) *.$(SHARED_EXT).* libzstd.pc
 	@$(RM) dll/libzstd.dll dll/libzstd.lib libzstd-nomt*
 	@$(RM) common/*.o compress/*.o decompress/*.o dictBuilder/*.o legacy/*.o deprecated/*.o
 	@echo Cleaning library completed
 
 #-----------------------------------------------------------------------------
-# make install is validated only for Linux, OSX, BSD, Hurd and Solaris targets
+# make install is validated only for Linux, macOS, BSD, Hurd and Solaris targets
 #-----------------------------------------------------------------------------
-ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS))
+ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
 
 DESTDIR     ?=
 # directory variables : GNU conventions prefer lowercase
@@ -134,7 +180,7 @@ LIBDIR      ?= $(libdir)
 includedir  ?= $(PREFIX)/include
 INCLUDEDIR  ?= $(includedir)
 
-ifneq (,$(filter $(shell uname),OpenBSD FreeBSD NetBSD DragonFly))
+ifneq (,$(filter $(shell uname),FreeBSD NetBSD DragonFly))
 PKGCONFIGDIR ?= $(PREFIX)/libdata/pkgconfig
 else
 PKGCONFIGDIR ?= $(LIBDIR)/pkgconfig
@@ -159,20 +205,32 @@ libzstd.pc: libzstd.pc.in
              -e 's|@VERSION@|$(VERSION)|' \
              $< >$@
 
-install: libzstd.a libzstd libzstd.pc
-	@$(INSTALL) -d -m 755 $(DESTDIR)$(PKGCONFIGDIR)/ $(DESTDIR)$(INCLUDEDIR)/
+install: install-pc install-static install-shared install-includes
+	@echo zstd static and shared library installed
+
+install-pc: libzstd.pc
+	@$(INSTALL) -d -m 755 $(DESTDIR)$(PKGCONFIGDIR)/
 	@$(INSTALL_DATA) libzstd.pc $(DESTDIR)$(PKGCONFIGDIR)/
-	@echo Installing libraries
+
+install-static: libzstd.a
+	@echo Installing static library
+	@$(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
 	@$(INSTALL_DATA) libzstd.a $(DESTDIR)$(LIBDIR)
+
+install-shared: libzstd
+	@echo Installing shared library
+	@$(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
 	@$(INSTALL_PROGRAM) $(LIBZSTD) $(DESTDIR)$(LIBDIR)
 	@ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR)
 	@ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT)
+
+install-includes:
 	@echo Installing includes
+	@$(INSTALL) -d -m 755 $(DESTDIR)$(INCLUDEDIR)/
 	@$(INSTALL_DATA) zstd.h $(DESTDIR)$(INCLUDEDIR)
 	@$(INSTALL_DATA) common/zstd_errors.h $(DESTDIR)$(INCLUDEDIR)
 	@$(INSTALL_DATA) deprecated/zbuff.h $(DESTDIR)$(INCLUDEDIR)     # prototypes generate deprecation warnings
 	@$(INSTALL_DATA) dictBuilder/zdict.h $(DESTDIR)$(INCLUDEDIR)
-	@echo zstd static and shared library installed
 
 uninstall:
 	@$(RM) $(DESTDIR)$(LIBDIR)/libzstd.a
diff --git a/lib/README.md b/lib/README.md
index 95196e46784c..0966c7aef497 100644
--- a/lib/README.md
+++ b/lib/README.md
@@ -13,7 +13,7 @@ including commands variables, staged install, directory variables and standard t
 - `make install` : install libraries in default system directories
 
 `libzstd` default scope includes compression, decompression, dictionary building,
-and decoding support for legacy formats >= v0.4.0.
+and decoding support for legacy formats >= v0.5.0.
 
 
 #### API
@@ -48,19 +48,24 @@ It's possible to compile only a limited set of features.
         This module depends on both `lib/common` and `lib/compress` .
 - `lib/legacy` : source code to decompress legacy zstd formats, starting from `v0.1.0`.
         This module depends on `lib/common` and `lib/decompress`.
-        To enable this feature, it's required to define `ZSTD_LEGACY_SUPPORT` during compilation.
-        Typically, with `gcc`, add argument `-DZSTD_LEGACY_SUPPORT=1`.
-        Using higher number limits versions supported.
+        To enable this feature, define `ZSTD_LEGACY_SUPPORT` during compilation.
+        Specifying a number limits versions supported to that version onward.
         For example, `ZSTD_LEGACY_SUPPORT=2` means : "support legacy formats >= v0.2.0".
         `ZSTD_LEGACY_SUPPORT=3` means : "support legacy formats >= v0.3.0", and so on.
-        Starting v0.8.0, all versions of `zstd` produce frames compliant with specification.
-        As a consequence, `ZSTD_LEGACY_SUPPORT=8` (or more) doesn't trigger legacy support.
-        Also, `ZSTD_LEGACY_SUPPORT=0` means "do __not__ support legacy formats".
+        Currently, the default library setting is `ZST_LEGACY_SUPPORT=5`.
+        It can be changed at build by any other value.
+        Note that any number >= 8 translates into "do __not__ support legacy formats",
+        since all versions of `zstd` >= v0.8 are compatible with v1+ specification.
+        `ZSTD_LEGACY_SUPPORT=0` also means "do __not__ support legacy formats".
         Once enabled, this capability is transparently triggered within decompression functions.
         It's also possible to invoke directly legacy API, as exposed in `lib/legacy/zstd_legacy.h`.
         Each version also provides an additional dedicated set of advanced API.
         For example, advanced API for version `v0.4` is exposed in `lib/legacy/zstd_v04.h` .
         Note : `lib/legacy` only supports _decoding_ legacy formats.
+- Similarly, you can define `ZSTD_LIB_COMPRESSION, ZSTD_LIB_DECOMPRESSION`, `ZSTD_LIB_DICTBUILDER`,
+        and `ZSTD_LIB_DEPRECATED` as 0 to forgo compilation of the corresponding features. This will
+        also disable compilation of all dependencies (eg. `ZSTD_LIB_COMPRESSION=0` will also disable
+        dictBuilder).
 
 
 #### Multithreading support
diff --git a/lib/common/bitstream.h b/lib/common/bitstream.h
index f7f389fe0fa8..ef89b9878e22 100644
--- a/lib/common/bitstream.h
+++ b/lib/common/bitstream.h
@@ -1,8 +1,7 @@
 /* ******************************************************************
    bitstream
    Part of FSE library
-   header file (to include)
-   Copyright (C) 2013-2017, Yann Collet.
+   Copyright (C) 2013-present, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -49,21 +48,10 @@ extern "C" {
 *  Dependencies
 ******************************************/
 #include "mem.h"            /* unaligned access routines */
+#include "debug.h"          /* assert(), DEBUGLOG(), RAWLOG() */
 #include "error_private.h"  /* error codes and messages */
 
 
-/*-*************************************
-*  Debug
-***************************************/
-#if defined(BIT_DEBUG) && (BIT_DEBUG>=1)
-#  include <assert.h>
-#else
-#  ifndef assert
-#    define assert(condition) ((void)0)
-#  endif
-#endif
-
-
 /*=========================================
 *  Target specific
 =========================================*/
@@ -83,8 +71,7 @@ extern "C" {
  * A critical property of these streams is that they encode and decode in **reverse** direction.
  * So the first bit sequence you add will be the last to be read, like a LIFO stack.
  */
-typedef struct
-{
+typedef struct {
     size_t bitContainer;
     unsigned bitPos;
     char*  startPtr;
@@ -118,8 +105,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
 /*-********************************************
 *  bitStream decoding API (read backward)
 **********************************************/
-typedef struct
-{
+typedef struct {
     size_t   bitContainer;
     unsigned bitsConsumed;
     const char* ptr;
@@ -236,7 +222,8 @@ MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
 }
 
 /*! BIT_addBitsFast() :
- *  works only if `value` is _clean_, meaning all high bits above nbBits are 0 */
+ *  works only if `value` is _clean_,
+ *  meaning all high bits above nbBits are 0 */
 MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
                                 size_t value, unsigned nbBits)
 {
@@ -352,17 +339,10 @@ MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
 
 MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
 {
-#if defined(__BMI__) && defined(__GNUC__) && __GNUC__*1000+__GNUC_MINOR__ >= 4008  /* experimental */
-#  if defined(__x86_64__)
-    if (sizeof(bitContainer)==8)
-        return _bextr_u64(bitContainer, start, nbBits);
-    else
-#  endif
-        return _bextr_u32(bitContainer, start, nbBits);
-#else
+    U32 const regMask = sizeof(bitContainer)*8 - 1;
+    /* if start > regMask, bitstream is corrupted, and result is undefined */
     assert(nbBits < BIT_MASK_SIZE);
-    return (bitContainer >> start) & BIT_mask[nbBits];
-#endif
+    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
 }
 
 MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
@@ -379,9 +359,13 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
  * @return : value extracted */
 MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
 {
-#if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
+    /* arbitrate between double-shift and shift+mask */
+#if 1
+    /* if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8,
+     * bitstream is likely corrupted, and result is undefined */
     return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
 #else
+    /* this code path is slower on my os-x laptop */
     U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
     return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
 #endif
diff --git a/lib/common/compiler.h b/lib/common/compiler.h
index e90a3bcde36c..07f875e4d38e 100644
--- a/lib/common/compiler.h
+++ b/lib/common/compiler.h
@@ -77,9 +77,9 @@
  * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
  */
 #ifndef DYNAMIC_BMI2
-  #if (defined(__clang__) && __has_attribute(__target__)) \
+  #if ((defined(__clang__) && __has_attribute(__target__)) \
       || (defined(__GNUC__) \
-          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
       && (defined(__x86_64__) || defined(_M_X86)) \
       && !defined(__BMI2__)
   #  define DYNAMIC_BMI2 1
@@ -88,15 +88,37 @@
   #endif
 #endif
 
-/* prefetch */
-#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
-#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-#  define PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
-#elif defined(__GNUC__)
-#  define PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
+/* prefetch
+ * can be disabled, by declaring NO_PREFETCH macro
+ * All prefetch invocations use a single default locality 2,
+ * generating instruction prefetcht1,
+ * which, according to Intel, means "load data into L2 cache".
+ * This is a good enough "middle ground" for the time being,
+ * though in theory, it would be better to specialize locality depending on data being prefetched.
+ * Tests could not determine any sensible difference based on locality value. */
+#if defined(NO_PREFETCH)
+#  define PREFETCH(ptr)     (void)(ptr)  /* disabled */
 #else
-#  define PREFETCH(ptr)   /* disabled */
-#endif
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define PREFETCH(ptr)   _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define PREFETCH(ptr)   __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
+#  else
+#    define PREFETCH(ptr)   (void)(ptr)  /* disabled */
+#  endif
+#endif  /* NO_PREFETCH */
+
+#define CACHELINE_SIZE 64
+
+#define PREFETCH_AREA(p, s)  {            \
+    const char* const _ptr = (const char*)(p);  \
+    size_t const _size = (size_t)(s);     \
+    size_t _pos;                          \
+    for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) {  \
+        PREFETCH(_ptr + _pos);            \
+    }                                     \
+}
 
 /* disable warnings */
 #ifdef _MSC_VER    /* Visual Studio */
diff --git a/lib/common/cpu.h b/lib/common/cpu.h
index 4eb48e39e10e..eeb428ad5f6a 100644
--- a/lib/common/cpu.h
+++ b/lib/common/cpu.h
@@ -36,7 +36,7 @@ MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
     U32 f1d = 0;
     U32 f7b = 0;
     U32 f7c = 0;
-#ifdef _MSC_VER
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
     int reg[4];
     __cpuid((int*)reg, 0);
     {
@@ -72,8 +72,7 @@ MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
           "cpuid\n\t"
           "popl %%ebx\n\t"
           : "=a"(f1a), "=c"(f1c), "=d"(f1d)
-          : "a"(1)
-          :);
+          : "a"(1));
     }
     if (n >= 7) {
       __asm__(
diff --git a/lib/common/debug.c b/lib/common/debug.c
new file mode 100644
index 000000000000..3ebdd1cb15a6
--- /dev/null
+++ b/lib/common/debug.c
@@ -0,0 +1,44 @@
+/* ******************************************************************
+   debug
+   Part of FSE library
+   Copyright (C) 2013-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+
+/*
+ * This module only hosts one global variable
+ * which can be used to dynamically influence the verbosity of traces,
+ * such as DEBUGLOG and RAWLOG
+ */
+
+#include "debug.h"
+
+int g_debuglevel = DEBUGLEVEL;
diff --git a/lib/common/debug.h b/lib/common/debug.h
new file mode 100644
index 000000000000..0c04ad2cc98c
--- /dev/null
+++ b/lib/common/debug.h
@@ -0,0 +1,123 @@
+/* ******************************************************************
+   debug
+   Part of FSE library
+   Copyright (C) 2013-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+
+/*
+ * The purpose of this header is to enable debug functions.
+ * They regroup assert(), DEBUGLOG() and RAWLOG() for run-time,
+ * and DEBUG_STATIC_ASSERT() for compile-time.
+ *
+ * By default, DEBUGLEVEL==0, which means run-time debug is disabled.
+ *
+ * Level 1 enables assert() only.
+ * Starting level 2, traces can be generated and pushed to stderr.
+ * The higher the level, the more verbose the traces.
+ *
+ * It's possible to dynamically adjust level using variable g_debug_level,
+ * which is only declared if DEBUGLEVEL>=2,
+ * and is a global variable, not multi-thread protected (use with care)
+ */
+
+#ifndef DEBUG_H_12987983217
+#define DEBUG_H_12987983217
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* static assert is triggered at compile time, leaving no runtime artefact,
+ * but can only work with compile-time constants.
+ * This variant can only be used inside a function. */
+#define DEBUG_STATIC_ASSERT(c) (void)sizeof(char[(c) ? 1 : -1])
+
+
+/* DEBUGLEVEL is expected to be defined externally,
+ * typically through compiler command line.
+ * Value must be a number. */
+#ifndef DEBUGLEVEL
+#  define DEBUGLEVEL 0
+#endif
+
+/* recommended values for DEBUGLEVEL :
+ * 0 : no debug, all run-time functions disabled
+ * 1 : no display, enables assert() only
+ * 2 : reserved, for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (verbose)
+ * 7+: events at every position (*very* verbose)
+ *
+ * It's generally inconvenient to output traces > 5.
+ * In which case, it's possible to selectively enable higher verbosity levels
+ * by modifying g_debug_level.
+ */
+
+#if (DEBUGLEVEL>=1)
+#  include <assert.h>
+#else
+#  ifndef assert   /* assert may be already defined, due to prior #include <assert.h> */
+#    define assert(condition) ((void)0)   /* disable assert (default) */
+#  endif
+#endif
+
+#if (DEBUGLEVEL>=2)
+#  include <stdio.h>
+extern int g_debuglevel; /* here, this variable is only declared,
+                           it actually lives in debug.c,
+                           and is shared by the whole process.
+                           It's typically used to enable very verbose levels
+                           on selective conditions (such as position in src) */
+
+#  define RAWLOG(l, ...) {                                      \
+                if (l<=g_debuglevel) {                          \
+                    fprintf(stderr, __VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                    \
+                if (l<=g_debuglevel) {                          \
+                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
+                    fprintf(stderr, " \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* DEBUG_H_12987983217 */
diff --git a/lib/common/entropy_common.c b/lib/common/entropy_common.c
index b37a082fee2c..b12944e1de93 100644
--- a/lib/common/entropy_common.c
+++ b/lib/common/entropy_common.c
@@ -72,7 +72,21 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     unsigned charnum = 0;
     int previous0 = 0;
 
-    if (hbSize < 4) return ERROR(srcSize_wrong);
+    if (hbSize < 4) {
+        /* This function only works when hbSize >= 4 */
+        char buffer[4];
+        memset(buffer, 0, sizeof(buffer));
+        memcpy(buffer, headerBuffer, hbSize);
+        {   size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
+                                                    buffer, sizeof(buffer));
+            if (FSE_isError(countSize)) return countSize;
+            if (countSize > hbSize) return ERROR(corruption_detected);
+            return countSize;
+    }   }
+    assert(hbSize >= 4);
+
+    /* init */
+    memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
     bitStream = MEM_readLE32(ip);
     nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
     if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
@@ -105,6 +119,7 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
             while (charnum < n0) normalizedCounter[charnum++] = 0;
             if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                assert((bitCount >> 3) <= 3); /* For first condition to work */
                 ip += bitCount>>3;
                 bitCount &= 7;
                 bitStream = MEM_readLE32(ip) >> bitCount;
diff --git a/lib/common/fse.h b/lib/common/fse.h
index 6a1d272be5cb..a5a6b6d4db70 100644
--- a/lib/common/fse.h
+++ b/lib/common/fse.h
@@ -72,6 +72,7 @@ extern "C" {
 #define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
 FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
 
+
 /*-****************************************
 *  FSE simple functions
 ******************************************/
@@ -129,7 +130,7 @@ FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src,
 ******************************************/
 /*!
 FSE_compress() does the following:
-1. count symbol occurrence from source[] into table count[]
+1. count symbol occurrence from source[] into table count[] (see hist.h)
 2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
 3. save normalized counters to memory buffer using writeNCount()
 4. build encoding table 'CTable' from normalized counters
@@ -147,15 +148,6 @@ or to save and provide normalized distribution using external method.
 
 /* *** COMPRESSION *** */
 
-/*! FSE_count():
-    Provides the precise count of each byte within a table 'count'.
-    'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
-    *maxSymbolValuePtr will be updated if detected smaller than initial value.
-    @return : the count of the most frequent symbol (which is not identified).
-              if return == srcSize, there is only one symbol.
-              Can also return an error code, which can be tested with FSE_isError(). */
-FSE_PUBLIC_API size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
 /*! FSE_optimalTableLog():
     dynamically downsize 'tableLog' when conditions are met.
     It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
@@ -167,7 +159,8 @@ FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize
     'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
     @return : tableLog,
               or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
+                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_NCountWriteBound():
     Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
@@ -178,8 +171,9 @@ FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tab
     Compactly save 'normalizedCounter' into 'buffer'.
     @return : size of the compressed table,
               or an errorCode, which can be tested using FSE_isError(). */
-FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
-
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                                 const short* normalizedCounter,
+                                 unsigned maxSymbolValue, unsigned tableLog);
 
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
@@ -250,7 +244,9 @@ If there is an error, the function will return an ErrorCode (which can be tested
     @return : size read from 'rBuffer',
               or an errorCode, which can be tested using FSE_isError().
               maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
-FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize);
 
 /*! Constructor and Destructor of FSE_DTable.
     Note that its size depends on 'tableLog' */
@@ -325,33 +321,8 @@ If there is an error, the function will return an error code, which can be teste
 
 
 /* *****************************************
-*  FSE advanced API
-*******************************************/
-/* FSE_count_wksp() :
- * Same as FSE_count(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned
- */
-size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* source, size_t sourceSize, unsigned* workSpace);
-
-/** FSE_countFast() :
- *  same as FSE_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr
- */
-size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
-/* FSE_countFast_wksp() :
- * Same as FSE_countFast(), but using an externally provided scratch buffer.
- * `workSpace` must be a table of minimum `1024` unsigned
- */
-size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* workSpace);
-
-/*! FSE_count_simple() :
- * Same as FSE_countFast(), but does not use any additional memory (not even on stack).
- * This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr` (presuming it's also the size of `count`).
-*/
-size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
-
+ *  FSE advanced API
+ ***************************************** */
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
 /**< same as FSE_optimalTableLog(), which used `minus==2` */
@@ -576,6 +547,39 @@ MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePt
 }
 
 
+/* FSE_getMaxNbBits() :
+ * Approximate maximum cost of a symbol, in bits.
+ * Fractional get rounded up (i.e : a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
+}
+
+/* FSE_bitCost() :
+ * Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
+    U32 const threshold = (minNbBits+1) << 16;
+    assert(tableLog < 16);
+    assert(accuracyLog < 31-tableLog);  /* ensure enough room for renormalization double shift */
+    {   U32 const tableSize = 1 << tableLog;
+        U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);
+        U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog;   /* linear interpolation (very approximate) */
+        U32 const bitMultiplier = 1 << accuracyLog;
+        assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);
+        assert(normalizedDeltaFromThreshold <= bitMultiplier);
+        return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;
+    }
+}
+
+
 /* ======    Decompression    ====== */
 
 typedef struct {
diff --git a/lib/common/fse_decompress.c b/lib/common/fse_decompress.c
index 4c66c3b77464..72bbead5beea 100644
--- a/lib/common/fse_decompress.c
+++ b/lib/common/fse_decompress.c
@@ -49,7 +49,7 @@
 *  Error Management
 ****************************************************************/
 #define FSE_isError ERR_isError
-#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
 
 /* check and forward error code */
 #define CHECK_F(f) { size_t const e = f; if (FSE_isError(e)) return e; }
diff --git a/lib/common/huf.h b/lib/common/huf.h
index b4645b4e5197..de9464111064 100644
--- a/lib/common/huf.h
+++ b/lib/common/huf.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
-   Huffman coder, part of New Generation Entropy library
-   header file
-   Copyright (C) 2013-2016, Yann Collet.
+   huff0 huffman codec,
+   part of Finite State Entropy library
+   Copyright (C) 2013-present, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -163,25 +163,25 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
 /* static allocation of HUF's DTable */
 typedef U32 HUF_DTable;
 #define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
-#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+#define HUF_CREATE_STATIC_DTABLEX1(DTable, maxTableLog) \
         HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
-#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
         HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
 
 
 /* ****************************************
 *  Advanced decompression functions
 ******************************************/
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
 
 size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
 size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
 size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /**< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
 
 
 /* ****************************************
@@ -208,7 +208,7 @@ size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, si
 typedef enum {
    HUF_repeat_none,  /**< Cannot use the previous table */
    HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
-   HUF_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+   HUF_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
  } HUF_repeat;
 /** HUF_compress4X_repeat() :
  *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
@@ -227,7 +227,9 @@ size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
  */
 #define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
 #define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
-size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize);
+size_t HUF_buildCTable_wksp (HUF_CElt* tree,
+                       const U32* count, U32 maxSymbolValue, U32 maxNbBits,
+                             void* workSpace, size_t wkspSize);
 
 /*! HUF_readStats() :
  *  Read compact Huffman tree, saved by HUF_writeCTable().
@@ -242,10 +244,15 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
  *  Loading a CTable saved with HUF_writeCTable() */
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
 
+/** HUF_getNbBits() :
+ *  Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
+ *  Note 1 : is not inlined, as HUF_CElt definition is private
+ *  Note 2 : const void* used, so that it can provide a statically allocated table as argument (which uses type U32) */
+U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue);
 
 /*
  * HUF_decompress() does the following:
- * 1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+ * 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
  * 2. build Huffman table from save, using HUF_readDTableX?()
  * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
  */
@@ -253,13 +260,13 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
 /** HUF_selectDecoder() :
  *  Tells which decoder is likely to decode faster,
  *  based on a set of pre-computed metrics.
- * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
  *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 
 /**
  *  The minimum workspace size for the `workSpace` used in
- *  HUF_readDTableX2_wksp() and HUF_readDTableX4_wksp().
+ *  HUF_readDTableX1_wksp() and HUF_readDTableX2_wksp().
  *
  *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
  *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
@@ -270,14 +277,14 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 #define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
 #define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
 
+size_t HUF_readDTableX1 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX1_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
 size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
 size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX4_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress4X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 
 
 /* ====================== */
@@ -298,25 +305,25 @@ size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
                        void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
                        HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
 
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
-size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
 
 size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
 size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
-size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+size_t HUF_decompress1X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
 
 size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
+size_t HUF_decompress1X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 
 /* BMI2 variants.
  * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
  */
 size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
-size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
 size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
 size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
 
diff --git a/lib/common/mem.h b/lib/common/mem.h
index 47d2300177c0..5da248756ffd 100644
--- a/lib/common/mem.h
+++ b/lib/common/mem.h
@@ -39,6 +39,10 @@ extern "C" {
 #  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
 #endif
 
+#ifndef __has_builtin
+#  define __has_builtin(x) 0  /* compat. with non-clang compilers */
+#endif
+
 /* code only tested on 32 and 64 bits systems */
 #define MEM_STATIC_ASSERT(c)   { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
 MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
@@ -57,11 +61,23 @@ MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (size
   typedef  uint64_t U64;
   typedef   int64_t S64;
 #else
+# include <limits.h>
+#if CHAR_BIT != 8
+#  error "this implementation requires char to be exactly 8-bit type"
+#endif
   typedef unsigned char      BYTE;
+#if USHRT_MAX != 65535
+#  error "this implementation requires short to be exactly 16-bit type"
+#endif
   typedef unsigned short      U16;
   typedef   signed short      S16;
+#if UINT_MAX != 4294967295
+#  error "this implementation requires int to be exactly 32-bit type"
+#endif
   typedef unsigned int        U32;
   typedef   signed int        S32;
+/* note : there are no limits defined for long long type in C90.
+ * limits exist in C99, however, in such case, <stdint.h> is preferred */
   typedef unsigned long long  U64;
   typedef   signed long long  S64;
 #endif
@@ -186,7 +202,8 @@ MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_ulong(in);
-#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap32))
     return __builtin_bswap32(in);
 #else
     return  ((in << 24) & 0xff000000 ) |
@@ -200,7 +217,8 @@ MEM_STATIC U64 MEM_swap64(U64 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_uint64(in);
-#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap64))
     return __builtin_bswap64(in);
 #else
     return  ((in << 56) & 0xff00000000000000ULL) |
diff --git a/lib/common/pool.c b/lib/common/pool.c
index 773488b07255..281b3824ac4d 100644
--- a/lib/common/pool.c
+++ b/lib/common/pool.c
@@ -10,9 +10,10 @@
 
 
 /* ======   Dependencies   ======= */
-#include <stddef.h>  /* size_t */
-#include "pool.h"
+#include <stddef.h>    /* size_t */
+#include "debug.h"     /* assert */
 #include "zstd_internal.h"  /* ZSTD_malloc, ZSTD_free */
+#include "pool.h"
 
 /* ======   Compiler specifics   ====== */
 #if defined(_MSC_VER)
@@ -33,8 +34,9 @@ typedef struct POOL_job_s {
 struct POOL_ctx_s {
     ZSTD_customMem customMem;
     /* Keep track of the threads */
-    ZSTD_pthread_t *threads;
-    size_t numThreads;
+    ZSTD_pthread_t* threads;
+    size_t threadCapacity;
+    size_t threadLimit;
 
     /* The queue is a circular buffer */
     POOL_job *queue;
@@ -58,10 +60,10 @@ struct POOL_ctx_s {
 };
 
 /* POOL_thread() :
-   Work thread for the thread pool.
-   Waits for jobs and executes them.
-   @returns : NULL on failure else non-null.
-*/
+ * Work thread for the thread pool.
+ * Waits for jobs and executes them.
+ * @returns : NULL on failure else non-null.
+ */
 static void* POOL_thread(void* opaque) {
     POOL_ctx* const ctx = (POOL_ctx*)opaque;
     if (!ctx) { return NULL; }
@@ -69,14 +71,17 @@ static void* POOL_thread(void* opaque) {
         /* Lock the mutex and wait for a non-empty queue or until shutdown */
         ZSTD_pthread_mutex_lock(&ctx->queueMutex);
 
-        while (ctx->queueEmpty && !ctx->shutdown) {
+        while ( ctx->queueEmpty
+            || (ctx->numThreadsBusy >= ctx->threadLimit) ) {
+            if (ctx->shutdown) {
+                /* even if !queueEmpty, (possible if numThreadsBusy >= threadLimit),
+                 * a few threads will be shutdown while !queueEmpty,
+                 * but enough threads will remain active to finish the queue */
+                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+                return opaque;
+            }
             ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
         }
-        /* empty => shutting down: so stop */
-        if (ctx->queueEmpty) {
-            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
-            return opaque;
-        }
         /* Pop a job off the queue */
         {   POOL_job const job = ctx->queue[ctx->queueHead];
             ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
@@ -89,30 +94,32 @@ static void* POOL_thread(void* opaque) {
             job.function(job.opaque);
 
             /* If the intended queue size was 0, signal after finishing job */
+            ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+            ctx->numThreadsBusy--;
             if (ctx->queueSize == 1) {
-                ZSTD_pthread_mutex_lock(&ctx->queueMutex);
-                ctx->numThreadsBusy--;
-                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
                 ZSTD_pthread_cond_signal(&ctx->queuePushCond);
-        }   }
+            }
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        }
     }  /* for (;;) */
-    /* Unreachable */
+    assert(0);  /* Unreachable */
 }
 
 POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
     return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
 }
 
-POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem) {
     POOL_ctx* ctx;
-    /* Check the parameters */
+    /* Check parameters */
     if (!numThreads) { return NULL; }
     /* Allocate the context and zero initialize */
     ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
     if (!ctx) { return NULL; }
     /* Initialize the job queue.
-     * It needs one extra space since one space is wasted to differentiate empty
-     * and full queues.
+     * It needs one extra space since one space is wasted to differentiate
+     * empty and full queues.
      */
     ctx->queueSize = queueSize + 1;
     ctx->queue = (POOL_job*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
@@ -126,7 +133,7 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customM
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
     ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
-    ctx->numThreads = 0;
+    ctx->threadCapacity = 0;
     ctx->customMem = customMem;
     /* Check for errors */
     if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
@@ -134,11 +141,12 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customM
     {   size_t i;
         for (i = 0; i < numThreads; ++i) {
             if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
-                ctx->numThreads = i;
+                ctx->threadCapacity = i;
                 POOL_free(ctx);
                 return NULL;
         }   }
-        ctx->numThreads = numThreads;
+        ctx->threadCapacity = numThreads;
+        ctx->threadLimit = numThreads;
     }
     return ctx;
 }
@@ -156,8 +164,8 @@ static void POOL_join(POOL_ctx* ctx) {
     ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
     /* Join all of the threads */
     {   size_t i;
-        for (i = 0; i < ctx->numThreads; ++i) {
-            ZSTD_pthread_join(ctx->threads[i], NULL);
+        for (i = 0; i < ctx->threadCapacity; ++i) {
+            ZSTD_pthread_join(ctx->threads[i], NULL);  /* note : could fail */
     }   }
 }
 
@@ -172,24 +180,68 @@ void POOL_free(POOL_ctx *ctx) {
     ZSTD_free(ctx, ctx->customMem);
 }
 
+
+
 size_t POOL_sizeof(POOL_ctx *ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
     return sizeof(*ctx)
         + ctx->queueSize * sizeof(POOL_job)
-        + ctx->numThreads * sizeof(ZSTD_pthread_t);
+        + ctx->threadCapacity * sizeof(ZSTD_pthread_t);
+}
+
+
+/* @return : 0 on success, 1 on error */
+static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
+{
+    if (numThreads <= ctx->threadCapacity) {
+        if (!numThreads) return 1;
+        ctx->threadLimit = numThreads;
+        return 0;
+    }
+    /* numThreads > threadCapacity */
+    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
+        if (!threadPool) return 1;
+        /* replace existing thread pool */
+        memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
+        ZSTD_free(ctx->threads, ctx->customMem);
+        ctx->threads = threadPool;
+        /* Initialize additional threads */
+        {   size_t threadId;
+            for (threadId = ctx->threadCapacity; threadId < numThreads; ++threadId) {
+                if (ZSTD_pthread_create(&threadPool[threadId], NULL, &POOL_thread, ctx)) {
+                    ctx->threadCapacity = threadId;
+                    return 1;
+            }   }
+    }   }
+    /* successfully expanded */
+    ctx->threadCapacity = numThreads;
+    ctx->threadLimit = numThreads;
+    return 0;
+}
+
+/* @return : 0 on success, 1 on error */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads)
+{
+    int result;
+    if (ctx==NULL) return 1;
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    result = POOL_resize_internal(ctx, numThreads);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return result;
 }
 
 /**
  * Returns 1 if the queue is full and 0 otherwise.
  *
- * If the queueSize is 1 (the pool was created with an intended queueSize of 0),
- * then a queue is empty if there is a thread free and no job is waiting.
+ * When queueSize is 1 (pool was created with an intended queueSize of 0),
+ * then a queue is empty if there is a thread free _and_ no job is waiting.
  */
 static int isQueueFull(POOL_ctx const* ctx) {
     if (ctx->queueSize > 1) {
         return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
     } else {
-        return ctx->numThreadsBusy == ctx->numThreads ||
+        return (ctx->numThreadsBusy == ctx->threadLimit) ||
                !ctx->queueEmpty;
     }
 }
@@ -263,6 +315,11 @@ void POOL_free(POOL_ctx* ctx) {
     (void)ctx;
 }
 
+int POOL_resize(POOL_ctx* ctx, size_t numThreads) {
+    (void)ctx; (void)numThreads;
+    return 0;
+}
+
 void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
     (void)ctx;
     function(opaque);
diff --git a/lib/common/pool.h b/lib/common/pool.h
index a57e9b4fabc2..458d37f13c3e 100644
--- a/lib/common/pool.h
+++ b/lib/common/pool.h
@@ -30,40 +30,50 @@ typedef struct POOL_ctx_s POOL_ctx;
 */
 POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
 
-POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem);
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem);
 
 /*! POOL_free() :
-    Free a thread pool returned by POOL_create().
-*/
+ *  Free a thread pool returned by POOL_create().
+ */
 void POOL_free(POOL_ctx* ctx);
 
+/*! POOL_resize() :
+ *  Expands or shrinks pool's number of threads.
+ *  This is more efficient than releasing + creating a new context,
+ *  since it tries to preserve and re-use existing threads.
+ * `numThreads` must be at least 1.
+ * @return : 0 when resize was successful,
+ *           !0 (typically 1) if there is an error.
+ *    note : only numThreads can be resized, queueSize remains unchanged.
+ */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads);
+
 /*! POOL_sizeof() :
-    return memory usage of pool returned by POOL_create().
-*/
+ * @return threadpool memory usage
+ *  note : compatible with NULL (returns 0 in this case)
+ */
 size_t POOL_sizeof(POOL_ctx* ctx);
 
 /*! POOL_function :
-    The function type that can be added to a thread pool.
-*/
+ *  The function type that can be added to a thread pool.
+ */
 typedef void (*POOL_function)(void*);
-/*! POOL_add_function :
-    The function type for a generic thread pool add function.
-*/
-typedef void (*POOL_add_function)(void*, POOL_function, void*);
 
 /*! POOL_add() :
-    Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
-    Possibly blocks until there is room in the queue.
-    Note : The function may be executed asynchronously, so `opaque` must live until the function has been completed.
-*/
+ *  Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
+ *  Possibly blocks until there is room in the queue.
+ *  Note : The function may be executed asynchronously,
+ *         therefore, `opaque` must live until function has been completed.
+ */
 void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
 
 
 /*! POOL_tryAdd() :
-    Add the job `function(opaque)` to the thread pool if a worker is available.
-    return immediately otherwise.
-   @return : 1 if successful, 0 if not.
-*/
+ *  Add the job `function(opaque)` to thread pool _if_ a worker is available.
+ *  Returns immediately even if not (does not block).
+ * @return : 1 if successful, 0 if not.
+ */
 int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
 
 
diff --git a/lib/common/xxhash.c b/lib/common/xxhash.c
index 9d9c0e963cbf..532b8161929d 100644
--- a/lib/common/xxhash.c
+++ b/lib/common/xxhash.c
@@ -98,6 +98,7 @@
 /* Modify the local functions below should you wish to use some other memory routines */
 /* for malloc(), free() */
 #include <stdlib.h>
+#include <stddef.h>     /* size_t */
 static void* XXH_malloc(size_t s) { return malloc(s); }
 static void  XXH_free  (void* p)  { free(p); }
 /* for memcpy() */
diff --git a/lib/common/zstd_common.c b/lib/common/zstd_common.c
index bccc948892d8..6f05d240e43c 100644
--- a/lib/common/zstd_common.c
+++ b/lib/common/zstd_common.c
@@ -46,11 +46,6 @@ ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
  *  provides error code string from enum */
 const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
 
-/*! g_debuglog_enable :
- *  turn on/off debug traces (global switch) */
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 2)
-int g_debuglog_enable = 1;
-#endif
 
 
 /*=**************************************************************
diff --git a/lib/common/zstd_internal.h b/lib/common/zstd_internal.h
index 65c08a825706..e75adfa61323 100644
--- a/lib/common/zstd_internal.h
+++ b/lib/common/zstd_internal.h
@@ -21,6 +21,7 @@
 ***************************************/
 #include "compiler.h"
 #include "mem.h"
+#include "debug.h"                 /* assert, DEBUGLOG, RAWLOG, g_debuglevel */
 #include "error_private.h"
 #define ZSTD_STATIC_LINKING_ONLY
 #include "zstd.h"
@@ -38,43 +39,8 @@
 extern "C" {
 #endif
 
-
-/*-*************************************
-*  Debug
-***************************************/
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-#  include <assert.h>
-#else
-#  ifndef assert
-#    define assert(condition) ((void)0)
-#  endif
-#endif
-
-#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
-
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
-#  include <stdio.h>
-extern int g_debuglog_enable;
-/* recommended values for ZSTD_DEBUG display levels :
- * 1 : no display, enables assert() only
- * 2 : reserved for currently active debug path
- * 3 : events once per object lifetime (CCtx, CDict, etc.)
- * 4 : events once per frame
- * 5 : events once per block
- * 6 : events once per sequence (*very* verbose) */
-#  define RAWLOG(l, ...) {                                      \
-                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
-                    fprintf(stderr, __VA_ARGS__);               \
-            }   }
-#  define DEBUGLOG(l, ...) {                                    \
-                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
-                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
-                    fprintf(stderr, " \n");                     \
-            }   }
-#else
-#  define RAWLOG(l, ...)      {}    /* disabled */
-#  define DEBUGLOG(l, ...)    {}    /* disabled */
-#endif
+/* ---- static assert (debug) --- */
+#define ZSTD_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)
 
 
 /*-*************************************
@@ -113,8 +79,7 @@ static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
 static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
 
-#define ZSTD_FRAMEIDSIZE 4
-static const size_t ZSTD_frameIdSize = ZSTD_FRAMEIDSIZE;  /* magic number size */
+#define ZSTD_FRAMEIDSIZE 4   /* magic number size */
 
 #define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
 static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
@@ -227,6 +192,8 @@ typedef struct {
     BYTE* llCode;
     BYTE* mlCode;
     BYTE* ofCode;
+    size_t maxNbSeq;
+    size_t maxNbLit;
     U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
     U32   longLengthPos;
 } seqStore_t;
diff --git a/lib/compress/fse_compress.c b/lib/compress/fse_compress.c
index cb8f1fa3233e..4408f0ed5b50 100644
--- a/lib/compress/fse_compress.c
+++ b/lib/compress/fse_compress.c
@@ -1,6 +1,6 @@
 /* ******************************************************************
    FSE : Finite State Entropy encoder
-   Copyright (C) 2013-2015, Yann Collet.
+   Copyright (C) 2013-present, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -37,9 +37,11 @@
 ****************************************************************/
 #include <stdlib.h>     /* malloc, free, qsort */
 #include <string.h>     /* memcpy, memset */
-#include <stdio.h>      /* printf (debug) */
-#include "bitstream.h"
 #include "compiler.h"
+#include "mem.h"        /* U32, U16, etc. */
+#include "debug.h"      /* assert, DEBUGLOG */
+#include "hist.h"       /* HIST_count_wksp */
+#include "bitstream.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #include "error_private.h"
@@ -49,7 +51,6 @@
 *  Error Management
 ****************************************************************/
 #define FSE_isError ERR_isError
-#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
 
 
 /* **************************************************************
@@ -82,7 +83,9 @@
  * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
  * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
  */
-size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+size_t FSE_buildCTable_wksp(FSE_CTable* ct,
+                      const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                            void* workSpace, size_t wkspSize)
 {
     U32 const tableSize = 1 << tableLog;
     U32 const tableMask = tableSize - 1;
@@ -100,9 +103,14 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
     if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
     tableU16[-2] = (U16) tableLog;
     tableU16[-1] = (U16) maxSymbolValue;
+    assert(tableLog < 16);   /* required for threshold strategy to work */
 
     /* For explanations on how to distribute symbol values over the table :
-    *  http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+     * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+     #ifdef __clang_analyzer__
+     memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     #endif
 
     /* symbol start positions */
     {   U32 u;
@@ -122,13 +130,15 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
         U32 symbol;
         for (symbol=0; symbol<=maxSymbolValue; symbol++) {
             int nbOccurences;
-            for (nbOccurences=0; nbOccurences<normalizedCounter[symbol]; nbOccurences++) {
+            int const freq = normalizedCounter[symbol];
+            for (nbOccurences=0; nbOccurences<freq; nbOccurences++) {
                 tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
                 position = (position + step) & tableMask;
-                while (position > highThreshold) position = (position + step) & tableMask;   /* Low proba area */
+                while (position > highThreshold)
+                    position = (position + step) & tableMask;   /* Low proba area */
         }   }
 
-        if (position!=0) return ERROR(GENERIC);   /* Must have gone through all positions */
+        assert(position==0);  /* Must have initialized all positions */
     }
 
     /* Build table */
@@ -143,7 +153,10 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
         for (s=0; s<=maxSymbolValue; s++) {
             switch (normalizedCounter[s])
             {
-            case  0: break;
+            case  0:
+                /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
+                symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
+                break;
 
             case -1:
             case  1:
@@ -160,6 +173,18 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
                     total +=  normalizedCounter[s];
     }   }   }   }
 
+#if 0  /* debug : symbol costs */
+    DEBUGLOG(5, "\n --- table statistics : ");
+    {   U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
+                symbol, normalizedCounter[symbol],
+                FSE_getMaxNbBits(symbolTT, symbol),
+                (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
+        }
+    }
+#endif
+
     return 0;
 }
 
@@ -174,8 +199,9 @@ size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned
 
 #ifndef FSE_COMMONDEFS_ONLY
 
+
 /*-**************************************************************
-*  FSE NCount encoding-decoding
+*  FSE NCount encoding
 ****************************************************************/
 size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
 {
@@ -183,9 +209,10 @@ size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
-static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
-                                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
-                                       unsigned writeIsSafe)
+static size_t
+FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                         unsigned writeIsSafe)
 {
     BYTE* const ostart = (BYTE*) header;
     BYTE* out = ostart;
@@ -194,13 +221,12 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     const int tableSize = 1 << tableLog;
     int remaining;
     int threshold;
-    U32 bitStream;
-    int bitCount;
-    unsigned charnum = 0;
-    int previous0 = 0;
+    U32 bitStream = 0;
+    int bitCount = 0;
+    unsigned symbol = 0;
+    unsigned const alphabetSize = maxSymbolValue + 1;
+    int previousIs0 = 0;
 
-    bitStream = 0;
-    bitCount  = 0;
     /* Table Size */
     bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
     bitCount  += 4;
@@ -210,48 +236,53 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     threshold = tableSize;
     nbBits = tableLog+1;
 
-    while (remaining>1) {  /* stops at 1 */
-        if (previous0) {
-            unsigned start = charnum;
-            while (!normalizedCounter[charnum]) charnum++;
-            while (charnum >= start+24) {
+    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
+        if (previousIs0) {
+            unsigned start = symbol;
+            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
+            if (symbol == alphabetSize) break;   /* incorrect distribution */
+            while (symbol >= start+24) {
                 start+=24;
                 bitStream += 0xFFFFU << bitCount;
-                if ((!writeIsSafe) && (out > oend-2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend-2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE) bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out+=2;
                 bitStream>>=16;
             }
-            while (charnum >= start+3) {
+            while (symbol >= start+3) {
                 start+=3;
                 bitStream += 3 << bitCount;
                 bitCount += 2;
             }
-            bitStream += (charnum-start) << bitCount;
+            bitStream += (symbol-start) << bitCount;
             bitCount += 2;
             if (bitCount>16) {
-                if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend - 2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE)bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out += 2;
                 bitStream >>= 16;
                 bitCount -= 16;
         }   }
-        {   int count = normalizedCounter[charnum++];
-            int const max = (2*threshold-1)-remaining;
+        {   int count = normalizedCounter[symbol++];
+            int const max = (2*threshold-1) - remaining;
             remaining -= count < 0 ? -count : count;
             count++;   /* +1 for extra accuracy */
-            if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            if (count>=threshold)
+                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
             bitStream += count << bitCount;
             bitCount  += nbBits;
             bitCount  -= (count<max);
-            previous0  = (count==1);
+            previousIs0  = (count==1);
             if (remaining<1) return ERROR(GENERIC);
             while (remaining<threshold) { nbBits--; threshold>>=1; }
         }
         if (bitCount>16) {
-            if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            if ((!writeIsSafe) && (out > oend - 2))
+                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
             out[0] = (BYTE)bitStream;
             out[1] = (BYTE)(bitStream>>8);
             out += 2;
@@ -259,19 +290,23 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
             bitCount -= 16;
     }   }
 
+    if (remaining != 1)
+        return ERROR(GENERIC);  /* incorrect normalized distribution */
+    assert(symbol <= alphabetSize);
+
     /* flush remaining bitStream */
-    if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    if ((!writeIsSafe) && (out > oend - 2))
+        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
     out[0] = (BYTE)bitStream;
     out[1] = (BYTE)(bitStream>>8);
     out+= (bitCount+7) /8;
 
-    if (charnum > maxSymbolValue + 1) return ERROR(GENERIC);
-
     return (out-ostart);
 }
 
 
-size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
@@ -279,179 +314,13 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalized
     if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
         return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
 
-    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1);
-}
-
-
-
-/*-**************************************************************
-*  Counting histogram
-****************************************************************/
-/*! FSE_count_simple
-    This function counts byte values within `src`, and store the histogram into table `count`.
-    It doesn't use any additional memory.
-    But this function is unsafe : it doesn't check that all values within `src` can fit into `count`.
-    For this reason, prefer using a table `count` with 256 elements.
-    @return : count of most numerous element.
-*/
-size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
-                        const void* src, size_t srcSize)
-{
-    const BYTE* ip = (const BYTE*)src;
-    const BYTE* const end = ip + srcSize;
-    unsigned maxSymbolValue = *maxSymbolValuePtr;
-    unsigned max=0;
-
-    memset(count, 0, (maxSymbolValue+1)*sizeof(*count));
-    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
-
-    while (ip<end) {
-        assert(*ip <= maxSymbolValue);
-        count[*ip++]++;
-    }
-
-    while (!count[maxSymbolValue]) maxSymbolValue--;
-    *maxSymbolValuePtr = maxSymbolValue;
-
-    { U32 s; for (s=0; s<=maxSymbolValue; s++) if (count[s] > max) max = count[s]; }
-
-    return (size_t)max;
-}
-
-
-/* FSE_count_parallel_wksp() :
- * Same as FSE_count_parallel(), but using an externally provided scratch buffer.
- * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`.
- * @return : largest histogram frequency, or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
-static size_t FSE_count_parallel_wksp(
-                                unsigned* count, unsigned* maxSymbolValuePtr,
-                                const void* source, size_t sourceSize,
-                                unsigned checkMax, unsigned* const workSpace)
-{
-    const BYTE* ip = (const BYTE*)source;
-    const BYTE* const iend = ip+sourceSize;
-    unsigned maxSymbolValue = *maxSymbolValuePtr;
-    unsigned max=0;
-    U32* const Counting1 = workSpace;
-    U32* const Counting2 = Counting1 + 256;
-    U32* const Counting3 = Counting2 + 256;
-    U32* const Counting4 = Counting3 + 256;
-
-    memset(workSpace, 0, 4*256*sizeof(unsigned));
-
-    /* safety checks */
-    if (!sourceSize) {
-        memset(count, 0, maxSymbolValue + 1);
-        *maxSymbolValuePtr = 0;
-        return 0;
-    }
-    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
-
-    /* by stripes of 16 bytes */
-    {   U32 cached = MEM_read32(ip); ip += 4;
-        while (ip < iend-15) {
-            U32 c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-        }
-        ip-=4;
-    }
-
-    /* finish last symbols */
-    while (ip<iend) Counting1[*ip++]++;
-
-    if (checkMax) {   /* verify stats will fit into destination table */
-        U32 s; for (s=255; s>maxSymbolValue; s--) {
-            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
-            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
-    }   }
-
-    {   U32 s;
-        if (maxSymbolValue > 255) maxSymbolValue = 255;
-        for (s=0; s<=maxSymbolValue; s++) {
-            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
-            if (count[s] > max) max = count[s];
-    }   }
-
-    while (!count[maxSymbolValue]) maxSymbolValue--;
-    *maxSymbolValuePtr = maxSymbolValue;
-    return (size_t)max;
-}
-
-/* FSE_countFast_wksp() :
- * Same as FSE_countFast(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned */
-size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                          const void* source, size_t sourceSize,
-                          unsigned* workSpace)
-{
-    if (sourceSize < 1500) /* heuristic threshold */
-        return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
-    return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
-}
-
-/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
-size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
-                     const void* source, size_t sourceSize)
-{
-    unsigned tmpCounters[1024];
-    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
-}
-
-/* FSE_count_wksp() :
- * Same as FSE_count(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned */
-size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* source, size_t sourceSize, unsigned* workSpace)
-{
-    if (*maxSymbolValuePtr < 255)
-        return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
-    *maxSymbolValuePtr = 255;
-    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace);
-}
-
-size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* src, size_t srcSize)
-{
-    unsigned tmpCounters[1024];
-    return FSE_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters);
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
 }
 
 
-
 /*-**************************************************************
 *  FSE Compression Code
 ****************************************************************/
-/*! FSE_sizeof_CTable() :
-    FSE_CTable is a variable size structure which contains :
-    `U16 tableLog;`
-    `U16 maxSymbolValue;`
-    `U16 nextStateNumber[1 << tableLog];`                         // This size is variable
-    `FSE_symbolCompressionTransform symbolTT[maxSymbolValue+1];`  // This size is variable
-Allocation is manual (C standard does not support variable-size structures).
-*/
-size_t FSE_sizeof_CTable (unsigned maxSymbolValue, unsigned tableLog)
-{
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
-    return FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-}
 
 FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
 {
@@ -466,7 +335,7 @@ void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 {
-    U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
+    U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
     U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
     U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
@@ -529,6 +398,9 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
     }
     ToDistribute = (1 << tableLog) - distributed;
 
+    if (ToDistribute == 0)
+        return 0;
+
     if ((total / ToDistribute) > lowOne) {
         /* risk of rounding to zero */
         lowOne = (U32)((total * 3) / (ToDistribute * 2));
@@ -629,11 +501,11 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
         U32 s;
         U32 nTotal = 0;
         for (s=0; s<=maxSymbolValue; s++)
-            printf("%3i: %4i \n", s, normalizedCounter[s]);
+            RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
         for (s=0; s<=maxSymbolValue; s++)
             nTotal += abs(normalizedCounter[s]);
         if (nTotal != (1U<<tableLog))
-            printf("Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+            RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
         getchar();
     }
 #endif
@@ -800,7 +672,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(maxCount, FSE_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
+    {   CHECK_V_F(maxCount, HIST_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
         if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
         if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
         if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
@@ -835,7 +707,7 @@ typedef struct {
 size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
 {
     fseWkspMax_t scratchBuffer;
-    FSE_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
     return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
 }
diff --git a/lib/compress/hist.c b/lib/compress/hist.c
new file mode 100644
index 000000000000..16524756b8dc
--- /dev/null
+++ b/lib/compress/hist.c
@@ -0,0 +1,195 @@
+/* ******************************************************************
+   hist : Histogram functions
+   part of Finite State Entropy project
+   Copyright (C) 2013-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* --- dependencies --- */
+#include "mem.h"             /* U32, BYTE, etc. */
+#include "debug.h"           /* assert, DEBUGLOG */
+#include "error_private.h"   /* ERROR */
+#include "hist.h"
+
+
+/* --- Error management --- */
+unsigned HIST_isError(size_t code) { return ERR_isError(code); }
+
+/*-**************************************************************
+ *  Histogram functions
+ ****************************************************************/
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned largestCount=0;
+
+    memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    {   U32 s;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > largestCount) largestCount = count[s];
+    }
+
+    return largestCount;
+}
+
+
+/* HIST_count_parallel_wksp() :
+ * store histogram into 4 intermediate tables, recombined at the end.
+ * this design makes better use of OoO cpus,
+ * and is noticeably faster when some values are heavily repeated.
+ * But it needs some additional workspace for intermediate tables.
+ * `workSpace` size must be a table of size >= HIST_WKSP_SIZE_U32.
+ * @return : largest histogram frequency,
+ *           or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
+static size_t HIST_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                unsigned checkMax,
+                                unsigned* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    memset(workSpace, 0, 4*256*sizeof(unsigned));
+
+    /* safety checks */
+    if (!sourceSize) {
+        memset(count, 0, maxSymbolValue + 1);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    if (checkMax) {   /* verify stats will fit into destination table */
+        U32 s; for (s=255; s>maxSymbolValue; s--) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
+    }   }
+
+    {   U32 s;
+        if (maxSymbolValue > 255) maxSymbolValue = 255;
+        for (s=0; s<=maxSymbolValue; s++) {
+            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
+            if (count[s] > max) max = count[s];
+    }   }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+    return (size_t)max;
+}
+
+/* HIST_countFast_wksp() :
+ * Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                          const void* source, size_t sourceSize,
+                          unsigned* workSpace)
+{
+    if (sourceSize < 1500) /* heuristic threshold */
+        return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
+}
+
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
+}
+
+/* HIST_count_wksp() :
+ * Same as HIST_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* source, size_t sourceSize, unsigned* workSpace)
+{
+    if (*maxSymbolValuePtr < 255)
+        return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
+    *maxSymbolValuePtr = 255;
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace);
+}
+
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters);
+}
diff --git a/lib/compress/hist.h b/lib/compress/hist.h
new file mode 100644
index 000000000000..8b1991a90bd3
--- /dev/null
+++ b/lib/compress/hist.h
@@ -0,0 +1,92 @@
+/* ******************************************************************
+   hist : Histogram functions
+   part of Finite State Entropy project
+   Copyright (C) 2013-present, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* --- dependencies --- */
+#include <stddef.h>   /* size_t */
+
+
+/* --- simple histogram functions --- */
+
+/*! HIST_count():
+ *  Provides the precise count of each byte within a table 'count'.
+ *  'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
+ *  Updates *maxSymbolValuePtr with actual largest symbol value detected.
+ *  @return : count of the most frequent symbol (which isn't identified).
+ *            or an error code, which can be tested using HIST_isError().
+ *            note : if return == srcSize, there is only one symbol.
+ */
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                  const void* src, size_t srcSize);
+
+unsigned HIST_isError(size_t code);  /**< tells if a return value is an error code */
+
+
+/* --- advanced histogram functions --- */
+
+#define HIST_WKSP_SIZE_U32 1024
+/** HIST_count_wksp() :
+ *  Same as HIST_count(), but using an externally provided scratch buffer.
+ *  Benefit is this function will use very little stack space.
+ * `workSpace` must be a table of unsigned of size >= HIST_WKSP_SIZE_U32
+ */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* src, size_t srcSize,
+                       unsigned* workSpace);
+
+/** HIST_countFast() :
+ *  same as HIST_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr.
+ *  This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr`
+ */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                      const void* src, size_t srcSize);
+
+/** HIST_countFast_wksp() :
+ *  Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` must be a table of unsigned of size >= HIST_WKSP_SIZE_U32
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize,
+                           unsigned* workSpace);
+
+/*! HIST_count_simple() :
+ *  Same as HIST_countFast(), this function is unsafe,
+ *  and will segfault if any value within `src` is `> *maxSymbolValuePtr`.
+ *  It is also a bit slower for large inputs.
+ *  However, it does not need any additional memory (not even on stack).
+ * @return : count of the most frequent symbol.
+ *  Note this function doesn't produce any error (i.e. it must succeed).
+ */
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize);
diff --git a/lib/compress/huf_compress.c b/lib/compress/huf_compress.c
index 83230b415f9c..4c40572f2284 100644
--- a/lib/compress/huf_compress.c
+++ b/lib/compress/huf_compress.c
@@ -45,8 +45,9 @@
 ****************************************************************/
 #include <string.h>     /* memcpy, memset */
 #include <stdio.h>      /* printf (debug) */
-#include "bitstream.h"
 #include "compiler.h"
+#include "bitstream.h"
+#include "hist.h"
 #define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
 #include "fse.h"        /* header compression */
 #define HUF_STATIC_LINKING_ONLY
@@ -58,7 +59,7 @@
 *  Error Management
 ****************************************************************/
 #define HUF_isError ERR_isError
-#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
 #define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
 #define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
 
@@ -81,7 +82,7 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
  * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
  */
 #define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
-size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
+static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
 {
     BYTE* const ostart = (BYTE*) dst;
     BYTE* op = ostart;
@@ -100,9 +101,9 @@ size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable,
     if (wtSize <= 1) return 0;  /* Not compressible */
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(maxCount, FSE_count_simple(count, &maxSymbolValue, weightTable, wtSize) );
+    {   unsigned const maxCount = HIST_count_simple(count, &maxSymbolValue, weightTable, wtSize);   /* never fails */
         if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
-        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+        if (maxCount == 1) return 0;        /* each symbol present maximum once => not compressible */
     }
 
     tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
@@ -216,6 +217,13 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32* maxSymbolValuePtr, const void* src
     return readSize;
 }
 
+U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue)
+{
+    const HUF_CElt* table = (const HUF_CElt*)symbolTable;
+    assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
+    return table[symbolValue].nbBits;
+}
+
 
 typedef struct nodeElt_s {
     U32 count;
@@ -660,9 +668,9 @@ static size_t HUF_compress_internal (
     }
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(largest, FSE_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->count) );
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->count) );
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
-        if (largest <= (srcSize >> 7)+1) return 0;   /* heuristic : probably not compressible enough */
+        if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
 
     /* Check validity of previous table */
diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
index 2aa26da4cd43..5f6280a8f7cc 100644
--- a/lib/compress/zstd_compress.c
+++ b/lib/compress/zstd_compress.c
@@ -8,21 +8,13 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
-
-/*-*************************************
-*  Tuning parameters
-***************************************/
-#ifndef ZSTD_CLEVEL_DEFAULT
-#  define ZSTD_CLEVEL_DEFAULT 3
-#endif
-
-
 /*-*************************************
 *  Dependencies
 ***************************************/
 #include <string.h>         /* memset */
 #include "cpu.h"
 #include "mem.h"
+#include "hist.h"           /* HIST_countFast_wksp */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
@@ -54,7 +46,6 @@ struct ZSTD_CDict_s {
     size_t workspaceSize;
     ZSTD_matchState_t matchState;
     ZSTD_compressedBlockState_t cBlockState;
-    ZSTD_compressionParameters cParams;
     ZSTD_customMem customMem;
     U32 dictID;
 };  /* typedef'd to ZSTD_CDict within "zstd.h" */
@@ -64,17 +55,26 @@ ZSTD_CCtx* ZSTD_createCCtx(void)
     return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
 }
 
+static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
+{
+    assert(cctx != NULL);
+    memset(cctx, 0, sizeof(*cctx));
+    cctx->customMem = memManager;
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    {   size_t const err = ZSTD_CCtx_resetParameters(cctx);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
 ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
 {
     ZSTD_STATIC_ASSERT(zcss_init==0);
     ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
     if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
-    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_calloc(sizeof(ZSTD_CCtx), customMem);
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_malloc(sizeof(ZSTD_CCtx), customMem);
         if (!cctx) return NULL;
-        cctx->customMem = customMem;
-        cctx->requestedParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
-        cctx->requestedParams.fParams.contentSizeFlag = 1;
-        cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+        ZSTD_initCCtx(cctx, customMem);
         return cctx;
     }
 }
@@ -102,17 +102,24 @@ ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
     return cctx;
 }
 
-size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
 {
-    if (cctx==NULL) return 0;   /* support free on NULL */
-    if (cctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static CCtx */
+    assert(cctx != NULL);
+    assert(cctx->staticSize == 0);
     ZSTD_free(cctx->workSpace, cctx->customMem); cctx->workSpace = NULL;
     ZSTD_freeCDict(cctx->cdictLocal); cctx->cdictLocal = NULL;
 #ifdef ZSTD_MULTITHREAD
     ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
 #endif
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    if (cctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static CCtx */
+    ZSTD_freeCCtxContent(cctx);
     ZSTD_free(cctx, cctx->customMem);
-    return 0;   /* reserved as a potential error code in the future */
+    return 0;
 }
 
 
@@ -143,21 +150,6 @@ size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
 /* private API call, for dictBuilder only */
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
 
-ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
-        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
-{
-    ZSTD_compressionParameters cParams = ZSTD_getCParams(CCtxParams->compressionLevel, srcSizeHint, dictSize);
-    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
-    if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
-    if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
-    if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
-    if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
-    if (CCtxParams->cParams.searchLength) cParams.searchLength = CCtxParams->cParams.searchLength;
-    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
-    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
-    return cParams;
-}
-
 static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
         ZSTD_compressionParameters cParams)
 {
@@ -251,7 +243,6 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_p_minMatch:
     case ZSTD_p_targetLength:
     case ZSTD_p_compressionStrategy:
-    case ZSTD_p_compressLiterals:
         return 1;
 
     case ZSTD_p_format:
@@ -268,6 +259,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_p_ldmMinMatch:
     case ZSTD_p_ldmBucketSizeLog:
     case ZSTD_p_ldmHashEveryLog:
+    case ZSTD_p_forceAttachDict:
     default:
         return 0;
     }
@@ -302,7 +294,6 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned v
         if (cctx->cdict) return ERROR(stage_wrong);
         return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
 
-    case ZSTD_p_compressLiterals:
     case ZSTD_p_contentSizeFlag:
     case ZSTD_p_checksumFlag:
     case ZSTD_p_dictIDFlag:
@@ -313,6 +304,9 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned v
                                    * default : 0 when using a CDict, 1 when using a Prefix */
         return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
 
+    case ZSTD_p_forceAttachDict:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
     case ZSTD_p_nbWorkers:
         if ((value>0) && cctx->staticSize) {
             return ERROR(parameter_unsupported);  /* MT not compatible with static alloc */
@@ -351,7 +345,6 @@ size_t ZSTD_CCtxParam_setParameter(
         int cLevel = (int)value;  /* cast expected to restore negative sign */
         if (cLevel > ZSTD_maxCLevel()) cLevel = ZSTD_maxCLevel();
         if (cLevel) {  /* 0 : does not change current level */
-            CCtxParams->disableLiteralCompression = (cLevel<0);  /* negative levels disable huffman */
             CCtxParams->compressionLevel = cLevel;
         }
         if (CCtxParams->compressionLevel >= 0) return CCtxParams->compressionLevel;
@@ -399,10 +392,6 @@ size_t ZSTD_CCtxParam_setParameter(
         CCtxParams->cParams.strategy = (ZSTD_strategy)value;
         return (size_t)CCtxParams->cParams.strategy;
 
-    case ZSTD_p_compressLiterals:
-        CCtxParams->disableLiteralCompression = !value;
-        return !CCtxParams->disableLiteralCompression;
-
     case ZSTD_p_contentSizeFlag :
         /* Content size written in frame header _when known_ (default:1) */
         DEBUGLOG(4, "set content size flag = %u", (value>0));
@@ -423,6 +412,12 @@ size_t ZSTD_CCtxParam_setParameter(
         CCtxParams->forceWindow = (value > 0);
         return CCtxParams->forceWindow;
 
+    case ZSTD_p_forceAttachDict :
+        CCtxParams->attachDictPref = value ?
+                                    (value > 0 ? ZSTD_dictForceAttach : ZSTD_dictForceCopy) :
+                                     ZSTD_dictDefaultAttach;
+        return CCtxParams->attachDictPref;
+
     case ZSTD_p_nbWorkers :
 #ifndef ZSTD_MULTITHREAD
         if (value>0) return ERROR(parameter_unsupported);
@@ -477,6 +472,98 @@ size_t ZSTD_CCtxParam_setParameter(
     }
 }
 
+size_t ZSTD_CCtx_getParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned* value)
+{
+    return ZSTD_CCtxParam_getParameter(&cctx->requestedParams, param, value);
+}
+
+size_t ZSTD_CCtxParam_getParameter(
+        ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, unsigned* value)
+{
+    switch(param)
+    {
+    case ZSTD_p_format :
+        *value = CCtxParams->format;
+        break;
+    case ZSTD_p_compressionLevel :
+        *value = CCtxParams->compressionLevel;
+        break;
+    case ZSTD_p_windowLog :
+        *value = CCtxParams->cParams.windowLog;
+        break;
+    case ZSTD_p_hashLog :
+        *value = CCtxParams->cParams.hashLog;
+        break;
+    case ZSTD_p_chainLog :
+        *value = CCtxParams->cParams.chainLog;
+        break;
+    case ZSTD_p_searchLog :
+        *value = CCtxParams->cParams.searchLog;
+        break;
+    case ZSTD_p_minMatch :
+        *value = CCtxParams->cParams.searchLength;
+        break;
+    case ZSTD_p_targetLength :
+        *value = CCtxParams->cParams.targetLength;
+        break;
+    case ZSTD_p_compressionStrategy :
+        *value = (unsigned)CCtxParams->cParams.strategy;
+        break;
+    case ZSTD_p_contentSizeFlag :
+        *value = CCtxParams->fParams.contentSizeFlag;
+        break;
+    case ZSTD_p_checksumFlag :
+        *value = CCtxParams->fParams.checksumFlag;
+        break;
+    case ZSTD_p_dictIDFlag :
+        *value = !CCtxParams->fParams.noDictIDFlag;
+        break;
+    case ZSTD_p_forceMaxWindow :
+        *value = CCtxParams->forceWindow;
+        break;
+    case ZSTD_p_forceAttachDict :
+        *value = CCtxParams->attachDictPref;
+        break;
+    case ZSTD_p_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        assert(CCtxParams->nbWorkers == 0);
+#endif
+        *value = CCtxParams->nbWorkers;
+        break;
+    case ZSTD_p_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        *value = CCtxParams->jobSize;
+        break;
+#endif
+    case ZSTD_p_overlapSizeLog :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        *value = CCtxParams->overlapSizeLog;
+        break;
+#endif
+    case ZSTD_p_enableLongDistanceMatching :
+        *value = CCtxParams->ldmParams.enableLdm;
+        break;
+    case ZSTD_p_ldmHashLog :
+        *value = CCtxParams->ldmParams.hashLog;
+        break;
+    case ZSTD_p_ldmMinMatch :
+        *value = CCtxParams->ldmParams.minMatchLength;
+        break;
+    case ZSTD_p_ldmBucketSizeLog :
+        *value = CCtxParams->ldmParams.bucketSizeLog;
+        break;
+    case ZSTD_p_ldmHashEveryLog :
+        *value = CCtxParams->ldmParams.hashEveryLog;
+        break;
+    default: return ERROR(parameter_unsupported);
+    }
+    return 0;
+}
+
 /** ZSTD_CCtx_setParametersUsingCCtxParams() :
  *  just applies `params` into `cctx`
  *  no action is performed, parameters are merely stored.
@@ -487,6 +574,7 @@ size_t ZSTD_CCtxParam_setParameter(
 size_t ZSTD_CCtx_setParametersUsingCCtxParams(
         ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
 {
+    DEBUGLOG(4, "ZSTD_CCtx_setParametersUsingCCtxParams");
     if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
     if (cctx->cdict) return ERROR(stage_wrong);
 
@@ -565,18 +653,19 @@ size_t ZSTD_CCtx_refPrefix_advanced(
     return 0;
 }
 
-static void ZSTD_startNewCompression(ZSTD_CCtx* cctx)
+/*! ZSTD_CCtx_reset() :
+ *  Also dumps dictionary */
+void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
 {
     cctx->streamStage = zcss_init;
     cctx->pledgedSrcSizePlusOne = 0;
 }
 
-/*! ZSTD_CCtx_reset() :
- *  Also dumps dictionary */
-void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
+size_t ZSTD_CCtx_resetParameters(ZSTD_CCtx* cctx)
 {
-    ZSTD_startNewCompression(cctx);
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
     cctx->cdict = NULL;
+    return ZSTD_CCtxParams_reset(&cctx->requestedParams);
 }
 
 /** ZSTD_checkCParams() :
@@ -589,8 +678,9 @@ size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
     CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
     CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
     CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
-    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN)
-        return ERROR(parameter_unsupported);
+    ZSTD_STATIC_ASSERT(ZSTD_TARGETLENGTH_MIN == 0);
+    if (cParams.targetLength > ZSTD_TARGETLENGTH_MAX)
+        return ERROR(parameter_outOfBound);
     if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
         return ERROR(parameter_unsupported);
     return 0;
@@ -599,7 +689,8 @@ size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
 /** ZSTD_clampCParams() :
  *  make CParam values within valid range.
  *  @return : valid CParams */
-static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+static ZSTD_compressionParameters
+ZSTD_clampCParams(ZSTD_compressionParameters cParams)
 {
 #   define CLAMP(val,min,max) {      \
         if (val<min) val=min;        \
@@ -610,8 +701,10 @@ static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters c
     CLAMP(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
     CLAMP(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
     CLAMP(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
-    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN) cParams.targetLength = ZSTD_TARGETLENGTH_MIN;
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) cParams.strategy = ZSTD_btultra;
+    ZSTD_STATIC_ASSERT(ZSTD_TARGETLENGTH_MIN == 0);
+    if (cParams.targetLength > ZSTD_TARGETLENGTH_MAX)
+        cParams.targetLength = ZSTD_TARGETLENGTH_MAX;
+    CLAMP(cParams.strategy, ZSTD_fast, ZSTD_btultra);
     return cParams;
 }
 
@@ -627,8 +720,11 @@ static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
     optimize `cPar` for a given input (`srcSize` and `dictSize`).
     mostly downsizing to reduce memory consumption and initialization latency.
     Both `srcSize` and `dictSize` are optional (use 0 if unknown).
-    Note : cPar is considered validated at this stage. Use ZSTD_checkCParams() to ensure that condition. */
-ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+    Note : cPar is assumed validated. Use ZSTD_checkCParams() to ensure this condition. */
+static ZSTD_compressionParameters
+ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
+                            unsigned long long srcSize,
+                            size_t dictSize)
 {
     static const U64 minSrcSize = 513; /* (1<<9) + 1 */
     static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
@@ -648,7 +744,7 @@ ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameter
                             ZSTD_highbit32(tSize-1) + 1;
         if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
     }
-    if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
+    if (cPar.hashLog > cPar.windowLog+1) cPar.hashLog = cPar.windowLog+1;
     {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
         if (cycleLog > cPar.windowLog)
             cPar.chainLog -= (cycleLog - cPar.windowLog);
@@ -660,13 +756,34 @@ ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameter
     return cPar;
 }
 
-ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+ZSTD_compressionParameters
+ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
+                   unsigned long long srcSize,
+                   size_t dictSize)
 {
     cPar = ZSTD_clampCParams(cPar);
     return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
 }
 
-static size_t ZSTD_sizeof_matchState(ZSTD_compressionParameters const* cParams, const U32 forCCtx)
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(CCtxParams->compressionLevel, srcSizeHint, dictSize);
+    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
+    if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
+    if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
+    if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
+    if (CCtxParams->cParams.searchLength) cParams.searchLength = CCtxParams->cParams.searchLength;
+    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
+    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
+    assert(!ZSTD_checkCParams(cParams));
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize);
+}
+
+static size_t
+ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
+                       const U32 forCCtx)
 {
     size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
     size_t const hSize = ((size_t)1) << cParams->hashLog;
@@ -693,7 +810,7 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
         U32    const divider = (cParams.searchLength==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const tokenSpace = WILDCOPY_OVERLENGTH + blockSize + 11*maxNbSeq;
         size_t const entropySpace = HUF_WORKSPACE_SIZE;
         size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
         size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
@@ -752,12 +869,14 @@ size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
     return ZSTD_estimateCStreamSize_usingCCtxParams(&params);
 }
 
-static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel) {
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel)
+{
     ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
     return ZSTD_estimateCStreamSize_usingCParams(cParams);
 }
 
-size_t ZSTD_estimateCStreamSize(int compressionLevel) {
+size_t ZSTD_estimateCStreamSize(int compressionLevel)
+{
     int level;
     size_t memBudget = 0;
     for (level=1; level<=compressionLevel; level++) {
@@ -786,9 +905,27 @@ ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
         fp.ingested = cctx->consumedSrcSize + buffered;
         fp.consumed = cctx->consumedSrcSize;
         fp.produced = cctx->producedCSize;
+        fp.flushed  = cctx->producedCSize;   /* simplified; some data might still be left within streaming output buffer */
+        fp.currentJobID = 0;
+        fp.nbActiveWorkers = 0;
         return fp;
 }   }
 
+/*! ZSTD_toFlushNow()
+ *  Only useful for multithreading scenarios currently (nbWorkers >= 1).
+ */
+size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_toFlushNow(cctx->mtctx);
+    }
+#endif
+    (void)cctx;
+    return 0;   /* over-simplification; could also check if context is currently running in streaming mode, and in which case, report how many bytes are left to be flushed within output buffer */
+}
+
+
 
 static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
                                   ZSTD_compressionParameters cParams2)
@@ -799,6 +936,20 @@ static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
          & ((cParams1.searchLength==3) == (cParams2.searchLength==3));  /* hashlog3 space */
 }
 
+static void ZSTD_assertEqualCParams(ZSTD_compressionParameters cParams1,
+                                    ZSTD_compressionParameters cParams2)
+{
+    (void)cParams1;
+    (void)cParams2;
+    assert(cParams1.windowLog    == cParams2.windowLog);
+    assert(cParams1.chainLog     == cParams2.chainLog);
+    assert(cParams1.hashLog      == cParams2.hashLog);
+    assert(cParams1.searchLog    == cParams2.searchLog);
+    assert(cParams1.searchLength == cParams2.searchLength);
+    assert(cParams1.targetLength == cParams2.targetLength);
+    assert(cParams1.strategy     == cParams2.strategy);
+}
+
 /** The parameters are equivalent if ldm is not enabled in both sets or
  *  all the parameters are equivalent. */
 static U32 ZSTD_equivalentLdmParams(ldmParams_t ldmParams1,
@@ -817,33 +968,51 @@ typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
 /* ZSTD_sufficientBuff() :
  * check internal buffers exist for streaming if buffPol == ZSTDb_buffered .
  * Note : they are assumed to be correctly sized if ZSTD_equivalentCParams()==1 */
-static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t blockSize1,
+static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t maxNbSeq1,
+                            size_t maxNbLit1,
                             ZSTD_buffered_policy_e buffPol2,
                             ZSTD_compressionParameters cParams2,
                             U64 pledgedSrcSize)
 {
     size_t const windowSize2 = MAX(1, (size_t)MIN(((U64)1 << cParams2.windowLog), pledgedSrcSize));
     size_t const blockSize2 = MIN(ZSTD_BLOCKSIZE_MAX, windowSize2);
+    size_t const maxNbSeq2 = blockSize2 / ((cParams2.searchLength == 3) ? 3 : 4);
+    size_t const maxNbLit2 = blockSize2;
     size_t const neededBufferSize2 = (buffPol2==ZSTDb_buffered) ? windowSize2 + blockSize2 : 0;
-    DEBUGLOG(4, "ZSTD_sufficientBuff: is windowSize2=%u <= wlog1=%u",
-                (U32)windowSize2, cParams2.windowLog);
-    DEBUGLOG(4, "ZSTD_sufficientBuff: is blockSize2=%u <= blockSize1=%u",
-                (U32)blockSize2, (U32)blockSize1);
-    return (blockSize2 <= blockSize1) /* seqStore space depends on blockSize */
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is neededBufferSize2=%u <= bufferSize1=%u",
+                (U32)neededBufferSize2, (U32)bufferSize1);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is maxNbSeq2=%u <= maxNbSeq1=%u",
+                (U32)maxNbSeq2, (U32)maxNbSeq1);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is maxNbLit2=%u <= maxNbLit1=%u",
+                (U32)maxNbLit2, (U32)maxNbLit1);
+    return (maxNbLit2 <= maxNbLit1)
+         & (maxNbSeq2 <= maxNbSeq1)
          & (neededBufferSize2 <= bufferSize1);
 }
 
 /** Equivalence for resetCCtx purposes */
 static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
                                  ZSTD_CCtx_params params2,
-                                 size_t buffSize1, size_t blockSize1,
+                                 size_t buffSize1,
+                                 size_t maxNbSeq1, size_t maxNbLit1,
                                  ZSTD_buffered_policy_e buffPol2,
                                  U64 pledgedSrcSize)
 {
     DEBUGLOG(4, "ZSTD_equivalentParams: pledgedSrcSize=%u", (U32)pledgedSrcSize);
-    return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
-           ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams) &&
-           ZSTD_sufficientBuff(buffSize1, blockSize1, buffPol2, params2.cParams, pledgedSrcSize);
+    if (!ZSTD_equivalentCParams(params1.cParams, params2.cParams)) {
+      DEBUGLOG(4, "ZSTD_equivalentCParams() == 0");
+      return 0;
+    }
+    if (!ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams)) {
+      DEBUGLOG(4, "ZSTD_equivalentLdmParams() == 0");
+      return 0;
+    }
+    if (!ZSTD_sufficientBuff(buffSize1, maxNbSeq1, maxNbLit1, buffPol2,
+                             params2.cParams, pledgedSrcSize)) {
+      DEBUGLOG(4, "ZSTD_sufficientBuff() == 0");
+      return 0;
+    }
+    return 1;
 }
 
 static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
@@ -851,10 +1020,10 @@ static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
     int i;
     for (i = 0; i < ZSTD_REP_NUM; ++i)
         bs->rep[i] = repStartValue[i];
-    bs->entropy.hufCTable_repeatMode = HUF_repeat_none;
-    bs->entropy.offcode_repeatMode = FSE_repeat_none;
-    bs->entropy.matchlength_repeatMode = FSE_repeat_none;
-    bs->entropy.litlength_repeatMode = FSE_repeat_none;
+    bs->entropy.huf.repeatMode = HUF_repeat_none;
+    bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.litlength_repeatMode = FSE_repeat_none;
 }
 
 /*! ZSTD_invalidateMatchState()
@@ -866,8 +1035,10 @@ static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
     ZSTD_window_clear(&ms->window);
 
     ms->nextToUpdate = ms->window.dictLimit + 1;
+    ms->nextToUpdate3 = ms->window.dictLimit + 1;
     ms->loadedDictEnd = 0;
     ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
+    ms->dictMatchState = NULL;
 }
 
 /*! ZSTD_continueCCtx() :
@@ -880,6 +1051,7 @@ static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pl
 
     cctx->blockSize = blockSize;   /* previous block size could be different even for same windowLog, due to pledgedSrcSize */
     cctx->appliedParams = params;
+    cctx->blockState.matchState.cParams = params.cParams;
     cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
     cctx->consumedSrcSize = 0;
     cctx->producedCSize = 0;
@@ -900,7 +1072,11 @@ static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pl
 
 typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
 
-static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compressionParameters const* cParams, ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
+static void*
+ZSTD_reset_matchState(ZSTD_matchState_t* ms,
+                      void* ptr,
+                const ZSTD_compressionParameters* cParams,
+                      ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
 {
     size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
     size_t const hSize = ((size_t)1) << cParams->hashLog;
@@ -912,6 +1088,9 @@ static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compre
 
     ms->hashLog3 = hashLog3;
     memset(&ms->window, 0, sizeof(ms->window));
+    ms->window.dictLimit = 1;    /* start from 1, so that 1st position is valid */
+    ms->window.lowLimit = 1;     /* it ensures first and later CCtx usages compress the same */
+    ms->window.nextSrc = ms->window.base + 1;   /* see issue #1241 */
     ZSTD_invalidateMatchState(ms);
 
     /* opt parser space */
@@ -937,14 +1116,24 @@ static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compre
     ms->hashTable3 = ms->chainTable + chainSize;
     ptr = ms->hashTable3 + h3Size;
 
+    ms->cParams = *cParams;
+
     assert(((size_t)ptr & 3) == 0);
     return ptr;
 }
 
+#define ZSTD_WORKSPACETOOLARGE_FACTOR 3 /* define "workspace is too large" as this number of times larger than needed */
+#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128  /* when workspace is continuously too large
+                                         * during at least this number of times,
+                                         * context's memory usage is considered wasteful,
+                                         * because it's sized to handle a worst case scenario which rarely happens.
+                                         * In which case, resize it down to free some memory */
+
 /*! ZSTD_resetCCtx_internal() :
     note : `params` are assumed fully validated at this stage */
 static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
-                                      ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                                      ZSTD_CCtx_params params,
+                                      U64 pledgedSrcSize,
                                       ZSTD_compResetPolicy_e const crp,
                                       ZSTD_buffered_policy_e const zbuff)
 {
@@ -954,34 +1143,35 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
     if (crp == ZSTDcrp_continue) {
         if (ZSTD_equivalentParams(zc->appliedParams, params,
-                                zc->inBuffSize, zc->blockSize,
-                                zbuff, pledgedSrcSize)) {
-            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%u)",
-                        zc->appliedParams.cParams.windowLog, (U32)zc->blockSize);
-            return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
+                                  zc->inBuffSize,
+                                  zc->seqStore.maxNbSeq, zc->seqStore.maxNbLit,
+                                  zbuff, pledgedSrcSize)) {
+            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%zu)",
+                        zc->appliedParams.cParams.windowLog, zc->blockSize);
+            zc->workSpaceOversizedDuration += (zc->workSpaceOversizedDuration > 0);   /* if it was too large, it still is */
+            if (zc->workSpaceOversizedDuration <= ZSTD_WORKSPACETOOLARGE_MAXDURATION)
+                return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
     }   }
     DEBUGLOG(4, "ZSTD_equivalentParams()==0 -> reset CCtx");
 
     if (params.ldmParams.enableLdm) {
         /* Adjust long distance matching parameters */
-        params.ldmParams.windowLog = params.cParams.windowLog;
         ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
         assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
         assert(params.ldmParams.hashEveryLog < 32);
-        zc->ldmState.hashPower =
-                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+        zc->ldmState.hashPower = ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
     }
 
     {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
         U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const tokenSpace = WILDCOPY_OVERLENGTH + blockSize + 11*maxNbSeq;
         size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
         size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
         size_t const matchStateSize = ZSTD_sizeof_matchState(&params.cParams, /* forCCtx */ 1);
         size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
-        void* ptr;
+        void* ptr;   /* used to partition workSpace */
 
         /* Check if workSpace is large enough, alloc a new one if needed */
         {   size_t const entropySpace = HUF_WORKSPACE_SIZE;
@@ -993,14 +1183,20 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             size_t const neededSpace = entropySpace + blockStateSpace + ldmSpace +
                                        ldmSeqSpace + matchStateSize + tokenSpace +
                                        bufferSpace;
-            DEBUGLOG(4, "Need %uKB workspace, including %uKB for match state, and %uKB for buffers",
-                        (U32)(neededSpace>>10), (U32)(matchStateSize>>10), (U32)(bufferSpace>>10));
-            DEBUGLOG(4, "windowSize: %u - blockSize: %u", (U32)windowSize, (U32)blockSize);
-
-            if (zc->workSpaceSize < neededSpace) {  /* too small : resize */
-                DEBUGLOG(4, "Need to update workSpaceSize from %uK to %uK",
-                            (unsigned)(zc->workSpaceSize>>10),
-                            (unsigned)(neededSpace>>10));
+
+            int const workSpaceTooSmall = zc->workSpaceSize < neededSpace;
+            int const workSpaceTooLarge = zc->workSpaceSize > ZSTD_WORKSPACETOOLARGE_FACTOR * neededSpace;
+            int const workSpaceWasteful = workSpaceTooLarge && (zc->workSpaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION);
+            zc->workSpaceOversizedDuration = workSpaceTooLarge ? zc->workSpaceOversizedDuration+1 : 0;
+
+            DEBUGLOG(4, "Need %zuKB workspace, including %zuKB for match state, and %zuKB for buffers",
+                        neededSpace>>10, matchStateSize>>10, bufferSpace>>10);
+            DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
+
+            if (workSpaceTooSmall || workSpaceWasteful) {
+                DEBUGLOG(4, "Need to resize workSpaceSize from %zuKB to %zuKB",
+                            zc->workSpaceSize >> 10,
+                            neededSpace >> 10);
                 /* static cctx : no resize, error out */
                 if (zc->staticSize) return ERROR(memory_allocation);
 
@@ -1009,9 +1205,11 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                 zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
                 if (zc->workSpace == NULL) return ERROR(memory_allocation);
                 zc->workSpaceSize = neededSpace;
-                ptr = zc->workSpace;
+                zc->workSpaceOversizedDuration = 0;
 
-                /* Statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+                /* Statically sized space.
+                 * entropyWorkspace never moves,
+                 * though prev/next block swap places */
                 assert(((size_t)zc->workSpace & 3) == 0);   /* ensure correct alignment */
                 assert(zc->workSpaceSize >= 2 * sizeof(ZSTD_compressedBlockState_t));
                 zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)zc->workSpace;
@@ -1022,6 +1220,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
         /* init params */
         zc->appliedParams = params;
+        zc->blockState.matchState.cParams = params.cParams;
         zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
         zc->consumedSrcSize = 0;
         zc->producedCSize = 0;
@@ -1058,13 +1257,18 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, &params.cParams, crp, /* forCCtx */ 1);
 
         /* sequences storage */
+        zc->seqStore.maxNbSeq = maxNbSeq;
         zc->seqStore.sequencesStart = (seqDef*)ptr;
         ptr = zc->seqStore.sequencesStart + maxNbSeq;
         zc->seqStore.llCode = (BYTE*) ptr;
         zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
         zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
         zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
-        ptr = zc->seqStore.litStart + blockSize;
+        /* ZSTD_wildcopy() is used to copy into the literals buffer,
+         * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
+         */
+        zc->seqStore.maxNbLit = blockSize;
+        ptr = zc->seqStore.litStart + blockSize + WILDCOPY_OVERLENGTH;
 
         /* ldm bucketOffsets table */
         if (params.ldmParams.enableLdm) {
@@ -1098,28 +1302,110 @@ void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
     assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
 }
 
-static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+/* These are the approximate sizes for each strategy past which copying the
+ * dictionary tables into the working context is faster than using them
+ * in-place.
+ */
+static const size_t attachDictSizeCutoffs[(unsigned)ZSTD_btultra+1] = {
+    8 KB, /* unused */
+    8 KB, /* ZSTD_fast */
+    16 KB, /* ZSTD_dfast */
+    32 KB, /* ZSTD_greedy */
+    32 KB, /* ZSTD_lazy */
+    32 KB, /* ZSTD_lazy2 */
+    32 KB, /* ZSTD_btlazy2 */
+    32 KB, /* ZSTD_btopt */
+    8 KB /* ZSTD_btultra */
+};
+
+static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
+                                 ZSTD_CCtx_params params,
+                                 U64 pledgedSrcSize)
+{
+    size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
+    return ( pledgedSrcSize <= cutoff
+          || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+          || params.attachDictPref == ZSTD_dictForceAttach )
+        && params.attachDictPref != ZSTD_dictForceCopy
+        && !params.forceWindow; /* dictMatchState isn't correctly
+                                 * handled in _enforceMaxDist */
+}
+
+static size_t ZSTD_resetCCtx_byAttachingCDict(
+    ZSTD_CCtx* cctx,
+    const ZSTD_CDict* cdict,
+    ZSTD_CCtx_params params,
+    U64 pledgedSrcSize,
+    ZSTD_buffered_policy_e zbuff)
+{
+    {
+        const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+        unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Resize working context table params for input only, since the dict
+         * has its own tables. */
+        params.cParams = ZSTD_adjustCParams_internal(*cdict_cParams, pledgedSrcSize, 0);
+        params.cParams.windowLog = windowLog;
+        ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                ZSTDcrp_continue, zbuff);
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+    }
+
+    {
+        const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
+                                  - cdict->matchState.window.base);
+        const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
+        if (cdictLen == 0) {
+            /* don't even attach dictionaries with no contents */
+            DEBUGLOG(4, "skipping attaching empty dictionary");
+        } else {
+            DEBUGLOG(4, "attaching dictionary into context");
+            cctx->blockState.matchState.dictMatchState = &cdict->matchState;
+
+            /* prep working match state so dict matches never have negative indices
+             * when they are translated to the working context's index space. */
+            if (cctx->blockState.matchState.window.dictLimit < cdictEnd) {
+                cctx->blockState.matchState.window.nextSrc =
+                    cctx->blockState.matchState.window.base + cdictEnd;
+                ZSTD_window_clear(&cctx->blockState.matchState.window);
+            }
+            cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
+        }
+    }
+
+    cctx->dictID = cdict->dictID;
+
+    /* copy block state */
+    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
                             const ZSTD_CDict* cdict,
-                            unsigned windowLog,
-                            ZSTD_frameParameters fParams,
+                            ZSTD_CCtx_params params,
                             U64 pledgedSrcSize,
                             ZSTD_buffered_policy_e zbuff)
 {
-    {   ZSTD_CCtx_params params = cctx->requestedParams;
+    const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+
+    DEBUGLOG(4, "copying dictionary into context");
+
+    {   unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
         /* Copy only compression parameters related to tables. */
-        params.cParams = cdict->cParams;
-        if (windowLog) params.cParams.windowLog = windowLog;
-        params.fParams = fParams;
+        params.cParams = *cdict_cParams;
+        params.cParams.windowLog = windowLog;
         ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
                                 ZSTDcrp_noMemset, zbuff);
-        assert(cctx->appliedParams.cParams.strategy == cdict->cParams.strategy);
-        assert(cctx->appliedParams.cParams.hashLog == cdict->cParams.hashLog);
-        assert(cctx->appliedParams.cParams.chainLog == cdict->cParams.chainLog);
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict_cParams->hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict_cParams->chainLog);
     }
 
     /* copy tables */
-    {   size_t const chainSize = (cdict->cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict->cParams.chainLog);
-        size_t const hSize =  (size_t)1 << cdict->cParams.hashLog;
+    {   size_t const chainSize = (cdict_cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict_cParams->chainLog);
+        size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
         size_t const tableSpace = (chainSize + hSize) * sizeof(U32);
         assert((U32*)cctx->blockState.matchState.chainTable == (U32*)cctx->blockState.matchState.hashTable + hSize);  /* chainTable must follow hashTable */
         assert((U32*)cctx->blockState.matchState.hashTable3 == (U32*)cctx->blockState.matchState.chainTable + chainSize);
@@ -1127,6 +1413,7 @@ static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
         assert((U32*)cdict->matchState.hashTable3 == (U32*)cdict->matchState.chainTable + chainSize);
         memcpy(cctx->blockState.matchState.hashTable, cdict->matchState.hashTable, tableSpace);   /* presumes all tables follow each other */
     }
+
     /* Zero the hashTable3, since the cdict never fills it */
     {   size_t const h3Size = (size_t)1 << cctx->blockState.matchState.hashLog3;
         assert(cdict->matchState.hashLog3 == 0);
@@ -1134,14 +1421,14 @@ static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
     }
 
     /* copy dictionary offsets */
-    {
-        ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+    {   ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
         ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
         dstMatchState->window       = srcMatchState->window;
         dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
         dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
         dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
     }
+
     cctx->dictID = cdict->dictID;
 
     /* copy block state */
@@ -1150,6 +1437,27 @@ static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
     return 0;
 }
 
+/* We have a choice between copying the dictionary context into the working
+ * context, or referencing the dictionary context from the working context
+ * in-place. We decide here which strategy to use. */
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            ZSTD_CCtx_params params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+
+    DEBUGLOG(4, "ZSTD_resetCCtx_usingCDict (pledgedSrcSize=%u)", (U32)pledgedSrcSize);
+
+    if (ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize)) {
+        return ZSTD_resetCCtx_byAttachingCDict(
+            cctx, cdict, params, pledgedSrcSize, zbuff);
+    } else {
+        return ZSTD_resetCCtx_byCopyingCDict(
+            cctx, cdict, params, pledgedSrcSize, zbuff);
+    }
+}
+
 /*! ZSTD_copyCCtx_internal() :
  *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
  *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
@@ -1192,7 +1500,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
 
     /* copy dictionary offsets */
     {
-        ZSTD_matchState_t const* srcMatchState = &srcCCtx->blockState.matchState;
+        const ZSTD_matchState_t* srcMatchState = &srcCCtx->blockState.matchState;
         ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
         dstMatchState->window       = srcMatchState->window;
         dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
@@ -1294,15 +1602,15 @@ static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
 
 /* See doc/zstd_compression_format.md for detailed format description */
 
-size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+static size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
 {
+    U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
     if (srcSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    MEM_writeLE24(dst, cBlockHeader24);
     memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
-    MEM_writeLE24(dst, (U32)(srcSize << 2) + (U32)bt_raw);
-    return ZSTD_blockHeaderSize+srcSize;
+    return ZSTD_blockHeaderSize + srcSize;
 }
 
-
 static size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
     BYTE* const ostart = (BYTE* const)dst;
@@ -1356,16 +1664,24 @@ static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, cons
 }
 
 
-static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
+/* ZSTD_minGain() :
+ * minimum compression required
+ * to generate a compress block or a compressed literals section.
+ * note : use same formula for both situations */
+static size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
+{
+    U32 const minlog = (strat==ZSTD_btultra) ? 7 : 6;
+    return (srcSize >> minlog) + 2;
+}
 
-static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
-                                     ZSTD_entropyCTables_t* nextEntropy,
+static size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
+                                     ZSTD_hufCTables_t* nextHuf,
                                      ZSTD_strategy strategy, int disableLiteralCompression,
                                      void* dst, size_t dstCapacity,
                                const void* src, size_t srcSize,
                                      U32* workspace, const int bmi2)
 {
-    size_t const minGain = ZSTD_minGain(srcSize);
+    size_t const minGain = ZSTD_minGain(srcSize, strategy);
     size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
     BYTE*  const ostart = (BYTE*)dst;
     U32 singleStream = srcSize < 256;
@@ -1376,27 +1692,25 @@ static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
                 disableLiteralCompression);
 
     /* Prepare nextEntropy assuming reusing the existing table */
-    nextEntropy->hufCTable_repeatMode = prevEntropy->hufCTable_repeatMode;
-    memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable,
-           sizeof(prevEntropy->hufCTable));
+    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
     if (disableLiteralCompression)
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
     /* small ? don't even attempt compression (speed opt) */
 #   define COMPRESS_LITERALS_SIZE_MIN 63
-    {   size_t const minLitSize = (prevEntropy->hufCTable_repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
         if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
 
     if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
-    {   HUF_repeat repeat = prevEntropy->hufCTable_repeatMode;
+    {   HUF_repeat repeat = prevHuf->repeatMode;
         int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
         if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
         cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2)
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2)
                                 : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2);
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
         if (repeat != HUF_repeat_none) {
             /* reused the existing table */
             hType = set_repeat;
@@ -1404,17 +1718,17 @@ static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
     }
 
     if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
-        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
+        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
     if (cLitSize==1) {
-        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
+        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
     }
 
     if (hType == set_compressed) {
         /* using a newly constructed table */
-        nextEntropy->hufCTable_repeatMode = HUF_repeat_check;
+        nextHuf->repeatMode = HUF_repeat_check;
     }
 
     /* Build header */
@@ -1451,6 +1765,7 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
     BYTE* const mlCodeTable = seqStorePtr->mlCode;
     U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     U32 u;
+    assert(nbSeq <= seqStorePtr->maxNbSeq);
     for (u=0; u<nbSeq; u++) {
         U32 const llv = sequences[u].litLength;
         U32 const mlv = sequences[u].matchLength;
@@ -1464,61 +1779,234 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
 }
 
+
+/**
+ * -log2(x / 256) lookup table for x in [0, 256).
+ * If x == 0: Return 0
+ * Else: Return floor(-log2(x / 256) * 256)
+ */
+static unsigned const kInverseProbabiltyLog256[256] = {
+    0,    2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
+    1130, 1100, 1073, 1047, 1024, 1001, 980,  960,  941,  923,  906,  889,
+    874,  859,  844,  830,  817,  804,  791,  779,  768,  756,  745,  734,
+    724,  714,  704,  694,  685,  676,  667,  658,  650,  642,  633,  626,
+    618,  610,  603,  595,  588,  581,  574,  567,  561,  554,  548,  542,
+    535,  529,  523,  517,  512,  506,  500,  495,  489,  484,  478,  473,
+    468,  463,  458,  453,  448,  443,  438,  434,  429,  424,  420,  415,
+    411,  407,  402,  398,  394,  390,  386,  382,  377,  373,  370,  366,
+    362,  358,  354,  350,  347,  343,  339,  336,  332,  329,  325,  322,
+    318,  315,  311,  308,  305,  302,  298,  295,  292,  289,  286,  282,
+    279,  276,  273,  270,  267,  264,  261,  258,  256,  253,  250,  247,
+    244,  241,  239,  236,  233,  230,  228,  225,  222,  220,  217,  215,
+    212,  209,  207,  204,  202,  199,  197,  194,  192,  190,  187,  185,
+    182,  180,  178,  175,  173,  171,  168,  166,  164,  162,  159,  157,
+    155,  153,  151,  149,  146,  144,  142,  140,  138,  136,  134,  132,
+    130,  128,  126,  123,  121,  119,  117,  115,  114,  112,  110,  108,
+    106,  104,  102,  100,  98,   96,   94,   93,   91,   89,   87,   85,
+    83,   82,   80,   78,   76,   74,   73,   71,   69,   67,   66,   64,
+    62,   61,   59,   57,   55,   54,   52,   50,   49,   47,   46,   44,
+    42,   41,   39,   37,   36,   34,   33,   31,   30,   28,   26,   25,
+    23,   22,   20,   19,   17,   16,   14,   13,   11,   10,   8,    7,
+    5,    4,    2,    1,
+};
+
+
+/**
+ * Returns the cost in bits of encoding the distribution described by count
+ * using the entropy bound.
+ */
+static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
+{
+    unsigned cost = 0;
+    unsigned s;
+    for (s = 0; s <= max; ++s) {
+        unsigned norm = (unsigned)((256 * count[s]) / total);
+        if (count[s] != 0 && norm == 0)
+            norm = 1;
+        assert(count[s] < total);
+        cost += count[s] * kInverseProbabiltyLog256[norm];
+    }
+    return cost >> 8;
+}
+
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using the
+ * table described by norm. The max symbol support by norm is assumed >= max.
+ * norm must be valid for every symbol with non-zero probability in count.
+ */
+static size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                                    unsigned const* count, unsigned const max)
+{
+    unsigned const shift = 8 - accuracyLog;
+    size_t cost = 0;
+    unsigned s;
+    assert(accuracyLog <= 8);
+    for (s = 0; s <= max; ++s) {
+        unsigned const normAcc = norm[s] != -1 ? norm[s] : 1;
+        unsigned const norm256 = normAcc << shift;
+        assert(norm256 > 0);
+        assert(norm256 < 256);
+        cost += count[s] * kInverseProbabiltyLog256[norm256];
+    }
+    return cost >> 8;
+}
+
+
+static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
+  void const* ptr = ctable;
+  U16 const* u16ptr = (U16 const*)ptr;
+  U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
+  return maxSymbolValue;
+}
+
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using ctable.
+ * Returns an error if ctable cannot represent all the symbols in count.
+ */
+static size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max)
+{
+    unsigned const kAccuracyLog = 8;
+    size_t cost = 0;
+    unsigned s;
+    FSE_CState_t cstate;
+    FSE_initCState(&cstate, ctable);
+    if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
+        DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
+                    ZSTD_getFSEMaxSymbolValue(ctable), max);
+        return ERROR(GENERIC);
+    }
+    for (s = 0; s <= max; ++s) {
+        unsigned const tableLog = cstate.stateLog;
+        unsigned const badCost = (tableLog + 1) << kAccuracyLog;
+        unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
+        if (count[s] == 0)
+            continue;
+        if (bitCost >= badCost) {
+            DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
+            return ERROR(GENERIC);
+        }
+        cost += count[s] * bitCost;
+    }
+    return cost >> kAccuracyLog;
+}
+
+/**
+ * Returns the cost in bytes of encoding the normalized count header.
+ * Returns an error if any of the helper functions return an error.
+ */
+static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
+                              size_t const nbSeq, unsigned const FSELog)
+{
+    BYTE wksp[FSE_NCOUNTBOUND];
+    S16 norm[MaxSeq + 1];
+    const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+    CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq, max));
+    return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
+}
+
+
 typedef enum {
     ZSTD_defaultDisallowed = 0,
     ZSTD_defaultAllowed = 1
 } ZSTD_defaultPolicy_e;
 
-MEM_STATIC
-symbolEncodingType_e ZSTD_selectEncodingType(
-        FSE_repeat* repeatMode, size_t const mostFrequent, size_t nbSeq,
-        U32 defaultNormLog, ZSTD_defaultPolicy_e const isDefaultAllowed)
+MEM_STATIC symbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_defaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy)
 {
-#define MIN_SEQ_FOR_DYNAMIC_FSE   64
-#define MAX_SEQ_FOR_STATIC_FSE  1000
     ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
-    if ((mostFrequent == nbSeq) && (!isDefaultAllowed || nbSeq > 2)) {
+    if (mostFrequent == nbSeq) {
+        *repeatMode = FSE_repeat_none;
+        if (isDefaultAllowed && nbSeq <= 2) {
+            /* Prefer set_basic over set_rle when there are 2 or less symbols,
+             * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+             * If basic encoding isn't possible, always choose RLE.
+             */
+            DEBUGLOG(5, "Selected set_basic");
+            return set_basic;
+        }
         DEBUGLOG(5, "Selected set_rle");
-        /* Prefer set_basic over set_rle when there are 2 or less symbols,
-         * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
-         * If basic encoding isn't possible, always choose RLE.
-         */
-        *repeatMode = FSE_repeat_check;
         return set_rle;
     }
-    if ( isDefaultAllowed
-      && (*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-        DEBUGLOG(5, "Selected set_repeat");
-        return set_repeat;
-    }
-    if ( isDefaultAllowed
-      && ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) ) {
-        DEBUGLOG(5, "Selected set_basic");
-        /* The format allows default tables to be repeated, but it isn't useful.
-         * When using simple heuristics to select encoding type, we don't want
-         * to confuse these tables with dictionaries. When running more careful
-         * analysis, we don't need to waste time checking both repeating tables
-         * and default tables.
-         */
-        *repeatMode = FSE_repeat_none;
-        return set_basic;
+    if (strategy < ZSTD_lazy) {
+        if (isDefaultAllowed) {
+            size_t const staticFse_nbSeq_max = 1000;
+            size_t const mult = 10 - strategy;
+            size_t const baseLog = 3;
+            size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog;  /* 28-36 for offset, 56-72 for lengths */
+            assert(defaultNormLog >= 5 && defaultNormLog <= 6);  /* xx_DEFAULTNORMLOG */
+            assert(mult <= 9 && mult >= 7);
+            if ( (*repeatMode == FSE_repeat_valid)
+              && (nbSeq < staticFse_nbSeq_max) ) {
+                DEBUGLOG(5, "Selected set_repeat");
+                return set_repeat;
+            }
+            if ( (nbSeq < dynamicFse_nbSeq_min)
+              || (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
+                DEBUGLOG(5, "Selected set_basic");
+                /* The format allows default tables to be repeated, but it isn't useful.
+                 * When using simple heuristics to select encoding type, we don't want
+                 * to confuse these tables with dictionaries. When running more careful
+                 * analysis, we don't need to waste time checking both repeating tables
+                 * and default tables.
+                 */
+                *repeatMode = FSE_repeat_none;
+                return set_basic;
+            }
+        }
+    } else {
+        size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
+        size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
+        size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
+        size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
+
+        if (isDefaultAllowed) {
+            assert(!ZSTD_isError(basicCost));
+            assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
+        }
+        assert(!ZSTD_isError(NCountCost));
+        assert(compressedCost < ERROR(maxCode));
+        DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
+                    (U32)basicCost, (U32)repeatCost, (U32)compressedCost);
+        if (basicCost <= repeatCost && basicCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_basic");
+            assert(isDefaultAllowed);
+            *repeatMode = FSE_repeat_none;
+            return set_basic;
+        }
+        if (repeatCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_repeat");
+            assert(!ZSTD_isError(repeatCost));
+            return set_repeat;
+        }
+        assert(compressedCost < basicCost && compressedCost < repeatCost);
     }
     DEBUGLOG(5, "Selected set_compressed");
     *repeatMode = FSE_repeat_check;
     return set_compressed;
 }
 
-MEM_STATIC
-size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
-        FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
-        U32* count, U32 max,
-        BYTE const* codeTable, size_t nbSeq,
-        S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
-        FSE_CTable const* prevCTable, size_t prevCTableSize,
-        void* workspace, size_t workspaceSize)
+MEM_STATIC size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+                U32* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* workspace, size_t workspaceSize)
 {
     BYTE* op = (BYTE*)dst;
-    BYTE const* const oend = op + dstCapacity;
+    const BYTE* const oend = op + dstCapacity;
 
     switch (type) {
     case set_rle:
@@ -1674,7 +2162,7 @@ ZSTD_encodeSequences_bmi2(
 
 #endif
 
-size_t ZSTD_encodeSequences(
+static size_t ZSTD_encodeSequences(
             void* dst, size_t dstCapacity,
             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
             FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
@@ -1706,10 +2194,11 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
                               const int bmi2)
 {
     const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
     U32 count[MaxSeq+1];
-    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
-    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
-    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
+    FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
     U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     const BYTE* const ofCodeTable = seqStorePtr->ofCode;
@@ -1720,15 +2209,17 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     BYTE* op = ostart;
     size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
     BYTE* seqHead;
+    BYTE* lastNCount = NULL;
 
     ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
         size_t const litSize = seqStorePtr->lit - literals;
+        int const disableLiteralCompression = (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
         size_t const cSize = ZSTD_compressLiterals(
-                                    prevEntropy, nextEntropy,
-                                    cctxParams->cParams.strategy, cctxParams->disableLiteralCompression,
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy, disableLiteralCompression,
                                     op, dstCapacity,
                                     literals, litSize,
                                     workspace, bmi2);
@@ -1747,13 +2238,9 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     else
         op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
     if (nbSeq==0) {
-      memcpy(nextEntropy->litlengthCTable, prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable));
-      nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
-      memcpy(nextEntropy->offcodeCTable, prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable));
-      nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
-      memcpy(nextEntropy->matchlengthCTable, prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable));
-      nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
-      return op - ostart;
+        /* Copy the old tables over as if we repeated them */
+        memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
+        return op - ostart;
     }
 
     /* seqHead : flags for FSE encoding type */
@@ -1763,43 +2250,53 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     ZSTD_seqToCodes(seqStorePtr);
     /* build CTable for Literal Lengths */
     {   U32 max = MaxLL;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);
+        size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);   /* can't fail */
         DEBUGLOG(5, "Building LL table");
-        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
-        LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog, ZSTD_defaultAllowed);
+        nextEntropy->fse.litlength_repeatMode = prevEntropy->fse.litlength_repeatMode;
+        LLtype = ZSTD_selectEncodingType(&nextEntropy->fse.litlength_repeatMode, count, max, mostFrequent, nbSeq, LLFSELog, prevEntropy->fse.litlengthCTable, LL_defaultNorm, LL_defaultNormLog, ZSTD_defaultAllowed, strategy);
+        assert(set_basic < set_compressed && set_rle < set_compressed);
+        assert(!(LLtype < set_compressed && nextEntropy->fse.litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
         {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
-                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
-                    prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
-                    workspace, HUF_WORKSPACE_SIZE);
+                                                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                                    prevEntropy->fse.litlengthCTable, sizeof(prevEntropy->fse.litlengthCTable),
+                                                    workspace, HUF_WORKSPACE_SIZE);
             if (ZSTD_isError(countSize)) return countSize;
+            if (LLtype == set_compressed)
+                lastNCount = op;
             op += countSize;
     }   }
     /* build CTable for Offsets */
     {   U32 max = MaxOff;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace);
+        size_t const mostFrequent = HIST_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace);  /* can't fail */
         /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
         ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
         DEBUGLOG(5, "Building OF table");
-        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
-        Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog, defaultPolicy);
+        nextEntropy->fse.offcode_repeatMode = prevEntropy->fse.offcode_repeatMode;
+        Offtype = ZSTD_selectEncodingType(&nextEntropy->fse.offcode_repeatMode, count, max, mostFrequent, nbSeq, OffFSELog, prevEntropy->fse.offcodeCTable, OF_defaultNorm, OF_defaultNormLog, defaultPolicy, strategy);
+        assert(!(Offtype < set_compressed && nextEntropy->fse.offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
         {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
-                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
-                    prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
-                    workspace, HUF_WORKSPACE_SIZE);
+                                                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                                    prevEntropy->fse.offcodeCTable, sizeof(prevEntropy->fse.offcodeCTable),
+                                                    workspace, HUF_WORKSPACE_SIZE);
             if (ZSTD_isError(countSize)) return countSize;
+            if (Offtype == set_compressed)
+                lastNCount = op;
             op += countSize;
     }   }
     /* build CTable for MatchLengths */
     {   U32 max = MaxML;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);
+        size_t const mostFrequent = HIST_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);   /* can't fail */
         DEBUGLOG(5, "Building ML table");
-        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
-        MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog, ZSTD_defaultAllowed);
+        nextEntropy->fse.matchlength_repeatMode = prevEntropy->fse.matchlength_repeatMode;
+        MLtype = ZSTD_selectEncodingType(&nextEntropy->fse.matchlength_repeatMode, count, max, mostFrequent, nbSeq, MLFSELog, prevEntropy->fse.matchlengthCTable, ML_defaultNorm, ML_defaultNormLog, ZSTD_defaultAllowed, strategy);
+        assert(!(MLtype < set_compressed && nextEntropy->fse.matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
         {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
-                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
-                    prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
-                    workspace, HUF_WORKSPACE_SIZE);
+                                                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                                    prevEntropy->fse.matchlengthCTable, sizeof(prevEntropy->fse.matchlengthCTable),
+                                                    workspace, HUF_WORKSPACE_SIZE);
             if (ZSTD_isError(countSize)) return countSize;
+            if (MLtype == set_compressed)
+                lastNCount = op;
             op += countSize;
     }   }
 
@@ -1814,21 +2311,37 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
                                         longOffsets, bmi2);
         if (ZSTD_isError(bitstreamSize)) return bitstreamSize;
         op += bitstreamSize;
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() recieves a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+        if (lastNCount && (op - lastNCount) < 4) {
+            /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(op - lastNCount == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
     }
 
     return op - ostart;
 }
 
 MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
-                              ZSTD_entropyCTables_t const* prevEntropy,
+                        const ZSTD_entropyCTables_t* prevEntropy,
                               ZSTD_entropyCTables_t* nextEntropy,
-                              ZSTD_CCtx_params const* cctxParams,
+                        const ZSTD_CCtx_params* cctxParams,
                               void* dst, size_t dstCapacity,
                               size_t srcSize, U32* workspace, int bmi2)
 {
     size_t const cSize = ZSTD_compressSequences_internal(
             seqStorePtr, prevEntropy, nextEntropy, cctxParams, dst, dstCapacity,
             workspace, bmi2);
+    if (cSize == 0) return 0;
     /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
      * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
      */
@@ -1837,40 +2350,55 @@ MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
     if (ZSTD_isError(cSize)) return cSize;
 
     /* Check compressibility */
-    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize);  /* note : fixed formula, maybe should depend on compression level, or strategy */
+    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
         if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
 
-    /* We check that dictionaries have offset codes available for the first
-     * block. After the first block, the offcode table might not have large
-     * enough codes to represent the offsets in the data.
-     */
-    if (nextEntropy->offcode_repeatMode == FSE_repeat_valid)
-        nextEntropy->offcode_repeatMode = FSE_repeat_check;
-
     return cSize;
 }
 
 /* ZSTD_selectBlockCompressor() :
  * Not static, but internal use only (used by long distance matcher)
  * assumption : strat is a valid strategy */
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode)
 {
-    static const ZSTD_blockCompressor blockCompressor[2][(unsigned)ZSTD_btultra+1] = {
+    static const ZSTD_blockCompressor blockCompressor[3][(unsigned)ZSTD_btultra+1] = {
         { ZSTD_compressBlock_fast  /* default for 0 */,
-          ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
-          ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
-          ZSTD_compressBlock_btopt, ZSTD_compressBlock_btultra },
+          ZSTD_compressBlock_fast,
+          ZSTD_compressBlock_doubleFast,
+          ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy,
+          ZSTD_compressBlock_lazy2,
+          ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt,
+          ZSTD_compressBlock_btultra },
         { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
-          ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
-          ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
-          ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btultra_extDict }
+          ZSTD_compressBlock_fast_extDict,
+          ZSTD_compressBlock_doubleFast_extDict,
+          ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,
+          ZSTD_compressBlock_lazy2_extDict,
+          ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict,
+          ZSTD_compressBlock_btultra_extDict },
+        { ZSTD_compressBlock_fast_dictMatchState  /* default for 0 */,
+          ZSTD_compressBlock_fast_dictMatchState,
+          ZSTD_compressBlock_doubleFast_dictMatchState,
+          ZSTD_compressBlock_greedy_dictMatchState,
+          ZSTD_compressBlock_lazy_dictMatchState,
+          ZSTD_compressBlock_lazy2_dictMatchState,
+          ZSTD_compressBlock_btlazy2_dictMatchState,
+          ZSTD_compressBlock_btopt_dictMatchState,
+          ZSTD_compressBlock_btultra_dictMatchState }
     };
+    ZSTD_blockCompressor selectedCompressor;
     ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
 
     assert((U32)strat >= (U32)ZSTD_fast);
     assert((U32)strat <= (U32)ZSTD_btultra);
-    return blockCompressor[extDict!=0][(U32)strat];
+    selectedCompressor = blockCompressor[(int)dictMode][(U32)strat];
+    assert(selectedCompressor != NULL);
+    return selectedCompressor;
 }
 
 static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
@@ -1880,7 +2408,7 @@ static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
     seqStorePtr->lit += lastLLSize;
 }
 
-static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
+void ZSTD_resetSeqStore(seqStore_t* ssPtr)
 {
     ssPtr->lit = ssPtr->litStart;
     ssPtr->sequences = ssPtr->sequencesStart;
@@ -1892,24 +2420,38 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
                                         const void* src, size_t srcSize)
 {
     ZSTD_matchState_t* const ms = &zc->blockState.matchState;
-    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
-                (U32)dstCapacity, ms->window.dictLimit, ms->nextToUpdate);
+    size_t cSize;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%zu, dictLimit=%u, nextToUpdate=%u)",
+                dstCapacity, ms->window.dictLimit, ms->nextToUpdate);
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+
+    /* Assert that we have correctly flushed the ctx params into the ms's copy */
+    ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
+
     if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
         ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.searchLength);
-        return 0;   /* don't even attempt compression below a certain srcSize */
+        cSize = 0;
+        goto out;  /* don't even attempt compression below a certain srcSize */
     }
     ZSTD_resetSeqStore(&(zc->seqStore));
+    ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;   /* required for optimal parser to read stats from dictionary */
+
+    /* a gap between an attached dict and the current window is not safe,
+     * they must remain adjacent, and when that stops being the case, the dict
+     * must be unset */
+    assert(ms->dictMatchState == NULL || ms->loadedDictEnd == ms->window.dictLimit);
 
     /* limited update after a very long match */
     {   const BYTE* const base = ms->window.base;
         const BYTE* const istart = (const BYTE*)src;
         const U32 current = (U32)(istart-base);
+        if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1));   /* ensure no overflow */
         if (current > ms->nextToUpdate + 384)
             ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
     }
 
     /* select and store sequences */
-    {   U32 const extDict = ZSTD_window_hasExtDict(ms->window);
+    {   ZSTD_dictMode_e const dictMode = ZSTD_matchState_dictMode(ms);
         size_t lastLLSize;
         {   int i;
             for (i = 0; i < ZSTD_REP_NUM; ++i)
@@ -1922,8 +2464,7 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
                 ZSTD_ldm_blockCompress(&zc->externSeqStore,
                                        ms, &zc->seqStore,
                                        zc->blockState.nextCBlock->rep,
-                                       &zc->appliedParams.cParams,
-                                       src, srcSize, extDict);
+                                       src, srcSize);
             assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
         } else if (zc->appliedParams.ldmParams.enableLdm) {
             rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
@@ -1939,31 +2480,38 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
                 ZSTD_ldm_blockCompress(&ldmSeqStore,
                                        ms, &zc->seqStore,
                                        zc->blockState.nextCBlock->rep,
-                                       &zc->appliedParams.cParams,
-                                       src, srcSize, extDict);
+                                       src, srcSize);
             assert(ldmSeqStore.pos == ldmSeqStore.size);
         } else {   /* not long range mode */
-            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, extDict);
-            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, &zc->appliedParams.cParams, src, srcSize);
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, dictMode);
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
         }
         {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
             ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
     }   }
 
     /* encode sequences and literals */
-    {   size_t const cSize = ZSTD_compressSequences(&zc->seqStore,
-                                &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
-                                &zc->appliedParams,
-                                dst, dstCapacity,
-                                srcSize, zc->entropyWorkspace, zc->bmi2);
-        if (ZSTD_isError(cSize) || cSize == 0) return cSize;
-        /* confirm repcodes and entropy tables */
-        {   ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
-            zc->blockState.prevCBlock = zc->blockState.nextCBlock;
-            zc->blockState.nextCBlock = tmp;
-        }
-        return cSize;
+    cSize = ZSTD_compressSequences(&zc->seqStore,
+            &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            srcSize, zc->entropyWorkspace, zc->bmi2);
+
+out:
+    if (!ZSTD_isError(cSize) && cSize != 0) {
+        /* confirm repcodes and entropy tables when emitting a compressed block */
+        ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
+        zc->blockState.prevCBlock = zc->blockState.nextCBlock;
+        zc->blockState.nextCBlock = tmp;
     }
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
 }
 
 
@@ -2005,13 +2553,13 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
             ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
             ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
             ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
-
             ZSTD_reduceIndex(cctx, correction);
             if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
             else ms->nextToUpdate -= correction;
             ms->loadedDictEnd = 0;
+            ms->dictMatchState = NULL;
         }
-        ZSTD_window_enforceMaxDist(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd);
+        ZSTD_window_enforceMaxDist(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
         if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
 
         {   size_t cSize = ZSTD_compressBlock_internal(cctx,
@@ -2020,11 +2568,8 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
             if (ZSTD_isError(cSize)) return cSize;
 
             if (cSize == 0) {  /* block is not compressible */
-                U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
-                if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
-                MEM_writeLE32(op, cBlockHeader24);   /* 4th byte will be overwritten */
-                memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
-                cSize = ZSTD_blockHeaderSize + blockSize;
+                cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+                if (ZSTD_isError(cSize)) return cSize;
             } else {
                 U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
                 MEM_writeLE24(op, cBlockHeader24);
@@ -2060,6 +2605,7 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
     BYTE  const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
     size_t pos=0;
 
+    assert(!(params.fParams.contentSizeFlag && pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN));
     if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
     DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
                 !params.fParams.noDictIDFlag, dictID,  dictIDSizeCode);
@@ -2122,7 +2668,7 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
                         const void* src, size_t srcSize,
                                U32 frame, U32 lastFrameChunk)
 {
-    ZSTD_matchState_t* ms = &cctx->blockState.matchState;
+    ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
     size_t fhSize = 0;
 
     DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
@@ -2143,8 +2689,25 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
     if (!ZSTD_window_update(&ms->window, src, srcSize)) {
         ms->nextToUpdate = ms->window.dictLimit;
     }
-    if (cctx->appliedParams.ldmParams.enableLdm)
+    if (cctx->appliedParams.ldmParams.enableLdm) {
         ZSTD_window_update(&cctx->ldmState.window, src, srcSize);
+    }
+
+    if (!frame) {
+        /* overflow check and correction for block mode */
+        if (ZSTD_window_needOverflowCorrection(ms->window, (const char*)src + srcSize)) {
+            U32 const cycleLog = ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy);
+            U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, 1 << cctx->appliedParams.cParams.windowLog, src);
+            ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+            ZSTD_reduceIndex(cctx, correction);
+            if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+            else ms->nextToUpdate -= correction;
+            ms->loadedDictEnd = 0;
+            ms->dictMatchState = NULL;
+        }
+    }
 
     DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (U32)cctx->blockSize);
     {   size_t const cSize = frame ?
@@ -2153,7 +2716,9 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
         if (ZSTD_isError(cSize)) return cSize;
         cctx->consumedSrcSize += srcSize;
         cctx->producedCSize += (cSize + fhSize);
-        if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+        assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+        if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+            ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
             if (cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne) {
                 DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize >= %u",
                     (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
@@ -2184,44 +2749,50 @@ size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const
 {
     size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
     if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
+
     return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
 }
 
 /*! ZSTD_loadDictionaryContent() :
  *  @return : 0, or an error code
  */
-static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* src, size_t srcSize)
+static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
+                                         ZSTD_CCtx_params const* params,
+                                         const void* src, size_t srcSize,
+                                         ZSTD_dictTableLoadMethod_e dtlm)
 {
     const BYTE* const ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
-    ZSTD_compressionParameters const* cParams = &params->cParams;
 
     ZSTD_window_update(&ms->window, src, srcSize);
     ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
 
+    /* Assert that we the ms params match the params we're being given */
+    ZSTD_assertEqualCParams(params->cParams, ms->cParams);
+
     if (srcSize <= HASH_READ_SIZE) return 0;
 
     switch(params->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, cParams, iend);
+        ZSTD_fillHashTable(ms, iend, dtlm);
         break;
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, cParams, iend);
+        ZSTD_fillDoubleHashTable(ms, iend, dtlm);
         break;
 
     case ZSTD_greedy:
     case ZSTD_lazy:
     case ZSTD_lazy2:
         if (srcSize >= HASH_READ_SIZE)
-            ZSTD_insertAndFindFirstIndex(ms, cParams, iend-HASH_READ_SIZE);
+            ZSTD_insertAndFindFirstIndex(ms, iend-HASH_READ_SIZE);
         break;
 
     case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
     case ZSTD_btopt:
     case ZSTD_btultra:
         if (srcSize >= HASH_READ_SIZE)
-            ZSTD_updateTree(ms, cParams, iend-HASH_READ_SIZE, iend);
+            ZSTD_updateTree(ms, iend-HASH_READ_SIZE, iend);
         break;
 
     default:
@@ -2256,7 +2827,12 @@ static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSym
  *  assumptions : magic number supposed already checked
  *                dictSize supposed > 8
  */
-static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* dict, size_t dictSize, void* workspace)
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
+                                      ZSTD_matchState_t* ms,
+                                      ZSTD_CCtx_params const* params,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictTableLoadMethod_e dtlm,
+                                      void* workspace)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
@@ -2265,13 +2841,15 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matc
     size_t dictID;
 
     ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(dictSize > 8);
+    assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
 
     dictPtr += 4;   /* skip magic number */
     dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
     dictPtr += 4;
 
     {   unsigned maxSymbolValue = 255;
-        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.hufCTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
         if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
         if (maxSymbolValue < 255) return ERROR(dictionary_corrupted);
         dictPtr += hufHeaderSize;
@@ -2282,7 +2860,8 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matc
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
         /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
-        CHECK_E( FSE_buildCTable_wksp(bs->entropy.offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
+        /* fill all offset symbols to avoid garbage at end of table */
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.offcodeCTable, offcodeNCount, MaxOff, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
                  dictionary_corrupted);
         dictPtr += offcodeHeaderSize;
     }
@@ -2294,7 +2873,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matc
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
         /* Every match length code must have non-zero probability */
         CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
-        CHECK_E( FSE_buildCTable_wksp(bs->entropy.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
                  dictionary_corrupted);
         dictPtr += matchlengthHeaderSize;
     }
@@ -2306,7 +2885,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matc
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
         /* Every literal length code must have non-zero probability */
         CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
-        CHECK_E( FSE_buildCTable_wksp(bs->entropy.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
                  dictionary_corrupted);
         dictPtr += litlengthHeaderSize;
     }
@@ -2332,22 +2911,25 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matc
                 if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
         }   }
 
-        bs->entropy.hufCTable_repeatMode = HUF_repeat_valid;
-        bs->entropy.offcode_repeatMode = FSE_repeat_valid;
-        bs->entropy.matchlength_repeatMode = FSE_repeat_valid;
-        bs->entropy.litlength_repeatMode = FSE_repeat_valid;
-        CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize));
+        bs->entropy.huf.repeatMode = HUF_repeat_valid;
+        bs->entropy.fse.offcode_repeatMode = FSE_repeat_valid;
+        bs->entropy.fse.matchlength_repeatMode = FSE_repeat_valid;
+        bs->entropy.fse.litlength_repeatMode = FSE_repeat_valid;
+        CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize, dtlm));
         return dictID;
     }
 }
 
 /** ZSTD_compress_insertDictionary() :
 *   @return : dictID, or an error code */
-static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms,
-                                             ZSTD_CCtx_params const* params,
-                                       const void* dict, size_t dictSize,
-                                             ZSTD_dictContentType_e dictContentType,
-                                             void* workspace)
+static size_t
+ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
+                               ZSTD_matchState_t* ms,
+                         const ZSTD_CCtx_params* params,
+                         const void* dict, size_t dictSize,
+                               ZSTD_dictContentType_e dictContentType,
+                               ZSTD_dictTableLoadMethod_e dtlm,
+                               void* workspace)
 {
     DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
     if ((dict==NULL) || (dictSize<=8)) return 0;
@@ -2356,12 +2938,12 @@ static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZS
 
     /* dict restricted modes */
     if (dictContentType == ZSTD_dct_rawContent)
-        return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+        return ZSTD_loadDictionaryContent(ms, params, dict, dictSize, dtlm);
 
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
         if (dictContentType == ZSTD_dct_auto) {
             DEBUGLOG(4, "raw content dictionary detected");
-            return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+            return ZSTD_loadDictionaryContent(ms, params, dict, dictSize, dtlm);
         }
         if (dictContentType == ZSTD_dct_fullDict)
             return ERROR(dictionary_wrong);
@@ -2369,17 +2951,18 @@ static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZS
     }
 
     /* dict as full zstd dictionary */
-    return ZSTD_loadZstdDictionary(bs, ms, params, dict, dictSize, workspace);
+    return ZSTD_loadZstdDictionary(bs, ms, params, dict, dictSize, dtlm, workspace);
 }
 
 /*! ZSTD_compressBegin_internal() :
  * @return : 0, or an error code */
-size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
-                             const void* dict, size_t dictSize,
-                             ZSTD_dictContentType_e dictContentType,
-                             const ZSTD_CDict* cdict,
-                             ZSTD_CCtx_params params, U64 pledgedSrcSize,
-                             ZSTD_buffered_policy_e zbuff)
+static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                                    ZSTD_buffered_policy_e zbuff)
 {
     DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params.cParams.windowLog);
     /* params are supposed to be fully validated at this point */
@@ -2387,9 +2970,7 @@ size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
 
     if (cdict && cdict->dictContentSize>0) {
-        cctx->requestedParams = params;
-        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params.cParams.windowLog,
-                                         params.fParams, pledgedSrcSize, zbuff);
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params, pledgedSrcSize, zbuff);
     }
 
     CHECK_F( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
@@ -2397,7 +2978,7 @@ size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
     {
         size_t const dictID = ZSTD_compress_insertDictionary(
                 cctx->blockState.prevCBlock, &cctx->blockState.matchState,
-                &params, dict, dictSize, dictContentType, cctx->entropyWorkspace);
+                &params, dict, dictSize, dictContentType, dtlm, cctx->entropyWorkspace);
         if (ZSTD_isError(dictID)) return dictID;
         assert(dictID <= (size_t)(U32)-1);
         cctx->dictID = (U32)dictID;
@@ -2408,6 +2989,7 @@ size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
 size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
                                     const void* dict, size_t dictSize,
                                     ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
                                     const ZSTD_CDict* cdict,
                                     ZSTD_CCtx_params params,
                                     unsigned long long pledgedSrcSize)
@@ -2416,7 +2998,7 @@ size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
     /* compression parameters verification and optimization */
     CHECK_F( ZSTD_checkCParams(params.cParams) );
     return ZSTD_compressBegin_internal(cctx,
-                                       dict, dictSize, dictContentType,
+                                       dict, dictSize, dictContentType, dtlm,
                                        cdict,
                                        params, pledgedSrcSize,
                                        ZSTDb_not_buffered);
@@ -2431,7 +3013,7 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
     ZSTD_CCtx_params const cctxParams =
             ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
     return ZSTD_compressBegin_advanced_internal(cctx,
-                                            dict, dictSize, ZSTD_dct_auto,
+                                            dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast,
                                             NULL /*cdict*/,
                                             cctxParams, pledgedSrcSize);
 }
@@ -2442,7 +3024,7 @@ size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t di
     ZSTD_CCtx_params const cctxParams =
             ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
     DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (U32)dictSize);
-    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
                                        cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
 }
 
@@ -2505,7 +3087,9 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
     if (ZSTD_isError(cSize)) return cSize;
     endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
     if (ZSTD_isError(endResult)) return endResult;
-    if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+    assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+    if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+        ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
         DEBUGLOG(4, "end of frame : controlling src size");
         if (cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1) {
             DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize = %u",
@@ -2517,22 +3101,22 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
 
 
 static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
-                               void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize,
-                         const void* dict,size_t dictSize,
-                               ZSTD_parameters params)
+                                      void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const void* dict,size_t dictSize,
+                                      ZSTD_parameters params)
 {
     ZSTD_CCtx_params const cctxParams =
             ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
     DEBUGLOG(4, "ZSTD_compress_internal");
     return ZSTD_compress_advanced_internal(cctx,
-                                          dst, dstCapacity,
-                                          src, srcSize,
-                                          dict, dictSize,
-                                          cctxParams);
+                                           dst, dstCapacity,
+                                           src, srcSize,
+                                           dict, dictSize,
+                                           cctxParams);
 }
 
-size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
+size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
                                void* dst, size_t dstCapacity,
                          const void* src, size_t srcSize,
                          const void* dict,size_t dictSize,
@@ -2540,7 +3124,11 @@ size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
 {
     DEBUGLOG(4, "ZSTD_compress_advanced");
     CHECK_F(ZSTD_checkCParams(params.cParams));
-    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+    return ZSTD_compress_internal(cctx,
+                                  dst, dstCapacity,
+                                  src, srcSize,
+                                  dict, dictSize,
+                                  params);
 }
 
 /* Internal */
@@ -2551,37 +3139,44 @@ size_t ZSTD_compress_advanced_internal(
         const void* dict,size_t dictSize,
         ZSTD_CCtx_params params)
 {
-    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)",
-                (U32)srcSize);
-    CHECK_F( ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
-                                         params, srcSize, ZSTDb_not_buffered) );
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)", (U32)srcSize);
+    CHECK_F( ZSTD_compressBegin_internal(cctx,
+                         dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                         params, srcSize, ZSTDb_not_buffered) );
     return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
 }
 
-size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
-                               const void* dict, size_t dictSize, int compressionLevel)
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict, size_t dictSize,
+                               int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, srcSize ? srcSize : 1, dict ? dictSize : 0);
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, srcSize + (!srcSize), dict ? dictSize : 0);
     ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
     assert(params.fParams.contentSizeFlag == 1);
-    ZSTD_CCtxParam_setParameter(&cctxParams, ZSTD_p_compressLiterals, compressionLevel>=0);
     return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, cctxParams);
 }
 
-size_t ZSTD_compressCCtx (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize,
+                         int compressionLevel)
 {
     DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (U32)srcSize);
+    assert(cctx != NULL);
     return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
 }
 
-size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+size_t ZSTD_compress(void* dst, size_t dstCapacity,
+               const void* src, size_t srcSize,
+                     int compressionLevel)
 {
     size_t result;
     ZSTD_CCtx ctxBody;
-    memset(&ctxBody, 0, sizeof(ctxBody));
-    ctxBody.customMem = ZSTD_defaultCMem;
+    ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
     result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
-    ZSTD_free(ctxBody.workSpace, ZSTD_defaultCMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
+    ZSTD_freeCCtxContent(&ctxBody);   /* can't free ctxBody itself, as it's on stack; free only heap content */
     return result;
 }
 
@@ -2619,9 +3214,9 @@ static size_t ZSTD_initCDict_internal(
                     ZSTD_dictContentType_e dictContentType,
                     ZSTD_compressionParameters cParams)
 {
-    DEBUGLOG(3, "ZSTD_initCDict_internal, dictContentType %u", (U32)dictContentType);
+    DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (U32)dictContentType);
     assert(!ZSTD_checkCParams(cParams));
-    cdict->cParams = cParams;
+    cdict->matchState.cParams = cParams;
     if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
         cdict->dictBuffer = NULL;
         cdict->dictContent = dictBuffer;
@@ -2654,7 +3249,7 @@ static size_t ZSTD_initCDict_internal(
         {   size_t const dictID = ZSTD_compress_insertDictionary(
                     &cdict->cBlockState, &cdict->matchState, &params,
                     cdict->dictContent, cdict->dictContentSize,
-                    dictContentType, cdict->workspace);
+                    dictContentType, ZSTD_dtlm_full, cdict->workspace);
             if (ZSTD_isError(dictID)) return dictID;
             assert(dictID <= (size_t)(U32)-1);
             cdict->dictID = (U32)dictID;
@@ -2775,7 +3370,7 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
 ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
 {
     assert(cdict != NULL);
-    return cdict->cParams;
+    return cdict->matchState.cParams;
 }
 
 /* ZSTD_compressBegin_usingCDict_advanced() :
@@ -2799,7 +3394,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(
         }
         params.fParams = fParams;
         return ZSTD_compressBegin_internal(cctx,
-                                           NULL, 0, ZSTD_dct_auto,
+                                           NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast,
                                            cdict,
                                            params, pledgedSrcSize,
                                            ZSTDb_not_buffered);
@@ -2813,7 +3408,7 @@ size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
 {
     ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
     DEBUGLOG(4, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
-    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
+    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
 size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
@@ -2880,16 +3475,17 @@ size_t ZSTD_CStreamOutSize(void)
 static size_t ZSTD_resetCStream_internal(ZSTD_CStream* cctx,
                     const void* const dict, size_t const dictSize, ZSTD_dictContentType_e const dictContentType,
                     const ZSTD_CDict* const cdict,
-                    ZSTD_CCtx_params const params, unsigned long long const pledgedSrcSize)
+                    ZSTD_CCtx_params params, unsigned long long const pledgedSrcSize)
 {
-    DEBUGLOG(4, "ZSTD_resetCStream_internal (disableLiteralCompression=%i)",
-                params.disableLiteralCompression);
+    DEBUGLOG(4, "ZSTD_resetCStream_internal");
+    /* Finalize the compression parameters */
+    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, dictSize);
     /* params are supposed to be fully validated at this point */
     assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
 
     CHECK_F( ZSTD_compressBegin_internal(cctx,
-                                         dict, dictSize, dictContentType,
+                                         dict, dictSize, dictContentType, ZSTD_dtlm_fast,
                                          cdict,
                                          params, pledgedSrcSize,
                                          ZSTDb_buffered) );
@@ -2912,7 +3508,6 @@ size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
     DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (U32)pledgedSrcSize);
     if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
     params.fParams.contentSizeFlag = 1;
-    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, 0);
     return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dct_auto, zcs->cdict, params, pledgedSrcSize);
 }
 
@@ -2925,6 +3520,7 @@ size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
                     ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
 {
     DEBUGLOG(4, "ZSTD_initCStream_internal");
+    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, dictSize);
     assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
 
@@ -2991,25 +3587,21 @@ size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
                 (U32)pledgedSrcSize, params.fParams.contentSizeFlag);
     CHECK_F( ZSTD_checkCParams(params.cParams) );
     if ((pledgedSrcSize==0) && (params.fParams.contentSizeFlag==0)) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* for compatibility with older programs relying on this behavior. Users should now specify ZSTD_CONTENTSIZE_UNKNOWN. This line will be removed in the future. */
-    {   ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-        return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL /*cdict*/, cctxParams, pledgedSrcSize);
-    }
+    zcs->requestedParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL /*cdict*/, zcs->requestedParams, pledgedSrcSize);
 }
 
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
+    ZSTD_CCtxParams_init(&zcs->requestedParams, compressionLevel);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, zcs->requestedParams, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
 size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
 {
     U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;  /* temporary : 0 interpreted as "unknown" during transition period. Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN. `0` will be interpreted as "empty" in the future */
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
-    ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-    return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, cctxParams, pledgedSrcSize);
+    ZSTD_CCtxParams_init(&zcs->requestedParams, compressionLevel);
+    return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, zcs->requestedParams, pledgedSrcSize);
 }
 
 size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
@@ -3073,7 +3665,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 ip = iend;
                 op += cSize;
                 zcs->frameEnded = 1;
-                ZSTD_startNewCompression(zcs);
+                ZSTD_CCtx_reset(zcs);
                 someMoreWork = 0; break;
             }
             /* complete loading into inBuffer */
@@ -3126,7 +3718,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     if (zcs->frameEnded) {
                         DEBUGLOG(5, "Frame completed directly in outBuffer");
                         someMoreWork = 0;
-                        ZSTD_startNewCompression(zcs);
+                        ZSTD_CCtx_reset(zcs);
                     }
                     break;
                 }
@@ -3154,7 +3746,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 if (zcs->frameEnded) {
                     DEBUGLOG(5, "Frame completed on flush");
                     someMoreWork = 0;
-                    ZSTD_startNewCompression(zcs);
+                    ZSTD_CCtx_reset(zcs);
                     break;
                 }
                 zcs->streamStage = zcss_load;
@@ -3207,19 +3799,16 @@ size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
         params.cParams = ZSTD_getCParamsFromCCtxParams(
                 &cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
 
+
 #ifdef ZSTD_MULTITHREAD
         if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
             params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
         }
         if (params.nbWorkers > 0) {
             /* mt context creation */
-            if (cctx->mtctx == NULL || (params.nbWorkers != ZSTDMT_getNbWorkers(cctx->mtctx))) {
+            if (cctx->mtctx == NULL) {
                 DEBUGLOG(4, "ZSTD_compress_generic: creating new mtctx for nbWorkers=%u",
                             params.nbWorkers);
-                if (cctx->mtctx != NULL)
-                    DEBUGLOG(4, "ZSTD_compress_generic: previous nbWorkers was %u",
-                                ZSTDMT_getNbWorkers(cctx->mtctx));
-                ZSTDMT_freeCCtx(cctx->mtctx);
                 cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbWorkers, cctx->customMem);
                 if (cctx->mtctx == NULL) return ERROR(memory_allocation);
             }
@@ -3251,8 +3840,9 @@ size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
         {   size_t const flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
             if ( ZSTD_isError(flushMin)
               || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
-                ZSTD_startNewCompression(cctx);
+                ZSTD_CCtx_reset(cctx);
             }
+            DEBUGLOG(5, "completed ZSTD_compress_generic delegating to ZSTDMT_compressStream_generic");
             return flushMin;
     }   }
 #endif
@@ -3308,82 +3898,83 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 
 #define ZSTD_MAX_CLEVEL     22
 int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
 
 static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
 {   /* "default" - guarantees a monotonically increasing memory budget */
     /* W,  C,  H,  S,  L, TL, strat */
     { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
-    { 19, 13, 14,  1,  7,  1, ZSTD_fast    },  /* level  1 */
-    { 19, 15, 16,  1,  6,  1, ZSTD_fast    },  /* level  2 */
-    { 20, 16, 17,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
-    { 20, 17, 18,  1,  5,  8, ZSTD_dfast   },  /* level  4 */
-    { 20, 17, 18,  2,  5, 16, ZSTD_greedy  },  /* level  5 */
-    { 21, 17, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
-    { 21, 18, 19,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
-    { 21, 18, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
-    { 21, 19, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
-    { 21, 19, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
-    { 22, 20, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 19, 13, 14,  1,  7,  0, ZSTD_fast    },  /* level  1 */
+    { 19, 15, 16,  1,  6,  0, ZSTD_fast    },  /* level  2 */
+    { 20, 16, 17,  1,  5,  1, ZSTD_dfast   },  /* level  3 */
+    { 20, 18, 18,  1,  5,  1, ZSTD_dfast   },  /* level  4 */
+    { 20, 18, 18,  2,  5,  2, ZSTD_greedy  },  /* level  5 */
+    { 21, 18, 19,  2,  5,  4, ZSTD_lazy    },  /* level  6 */
+    { 21, 18, 19,  3,  5,  8, ZSTD_lazy2   },  /* level  7 */
+    { 21, 19, 19,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 21, 19, 20,  4,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 21, 20, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 21, 21, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
     { 22, 20, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
     { 22, 21, 22,  4,  5, 32, ZSTD_btlazy2 },  /* level 13 */
     { 22, 21, 22,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
     { 22, 22, 22,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
     { 22, 21, 22,  4,  5, 48, ZSTD_btopt   },  /* level 16 */
-    { 23, 22, 22,  4,  4, 48, ZSTD_btopt   },  /* level 17 */
-    { 23, 22, 22,  5,  3, 64, ZSTD_btopt   },  /* level 18 */
-    { 23, 23, 22,  7,  3,128, ZSTD_btopt   },  /* level 19 */
-    { 25, 25, 23,  7,  3,128, ZSTD_btultra },  /* level 20 */
-    { 26, 26, 24,  7,  3,256, ZSTD_btultra },  /* level 21 */
-    { 27, 27, 25,  9,  3,512, ZSTD_btultra },  /* level 22 */
+    { 23, 22, 22,  4,  4, 64, ZSTD_btopt   },  /* level 17 */
+    { 23, 23, 22,  6,  3,256, ZSTD_btopt   },  /* level 18 */
+    { 23, 24, 22,  7,  3,256, ZSTD_btultra },  /* level 19 */
+    { 25, 25, 23,  7,  3,256, ZSTD_btultra },  /* level 20 */
+    { 26, 26, 24,  7,  3,512, ZSTD_btultra },  /* level 21 */
+    { 27, 27, 25,  9,  3,999, ZSTD_btultra },  /* level 22 */
 },
 {   /* for srcSize <= 256 KB */
     /* W,  C,  H,  S,  L,  T, strat */
     { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
-    { 18, 13, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
-    { 18, 14, 13,  1,  5,  8, ZSTD_dfast   },  /* level  2 */
-    { 18, 16, 15,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
-    { 18, 15, 17,  1,  5,  8, ZSTD_greedy  },  /* level  4.*/
-    { 18, 16, 17,  4,  5,  8, ZSTD_greedy  },  /* level  5.*/
-    { 18, 16, 17,  3,  5,  8, ZSTD_lazy    },  /* level  6.*/
-    { 18, 17, 17,  4,  4,  8, ZSTD_lazy    },  /* level  7 */
-    { 18, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
-    { 18, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
-    { 18, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
-    { 18, 18, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 11.*/
-    { 18, 18, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 12.*/
-    { 18, 19, 17,  7,  4,  8, ZSTD_btlazy2 },  /* level 13 */
-    { 18, 18, 18,  4,  4, 16, ZSTD_btopt   },  /* level 14.*/
-    { 18, 18, 18,  4,  3, 16, ZSTD_btopt   },  /* level 15.*/
-    { 18, 19, 18,  6,  3, 32, ZSTD_btopt   },  /* level 16.*/
-    { 18, 19, 18,  8,  3, 64, ZSTD_btopt   },  /* level 17.*/
-    { 18, 19, 18,  9,  3,128, ZSTD_btopt   },  /* level 18.*/
-    { 18, 19, 18, 10,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 18, 19, 18, 11,  3,512, ZSTD_btultra },  /* level 20.*/
-    { 18, 19, 18, 12,  3,512, ZSTD_btultra },  /* level 21.*/
-    { 18, 19, 18, 13,  3,512, ZSTD_btultra },  /* level 22.*/
+    { 18, 13, 14,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 14,  1,  5,  1, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 16,  1,  4,  1, ZSTD_dfast   },  /* level  3 */
+    { 18, 16, 17,  2,  5,  2, ZSTD_greedy  },  /* level  4.*/
+    { 18, 18, 18,  3,  5,  2, ZSTD_greedy  },  /* level  5.*/
+    { 18, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6.*/
+    { 18, 18, 19,  4,  4,  4, ZSTD_lazy    },  /* level  7 */
+    { 18, 18, 19,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 18, 19,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 18, 19,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 19,  5,  4, 16, ZSTD_btlazy2 },  /* level 11.*/
+    { 18, 19, 19,  6,  4, 16, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 19, 19,  8,  4, 16, ZSTD_btlazy2 },  /* level 13 */
+    { 18, 18, 19,  4,  4, 24, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 19,  4,  3, 24, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 19,  6,  3, 64, ZSTD_btopt   },  /* level 16.*/
+    { 18, 19, 19,  8,  3,128, ZSTD_btopt   },  /* level 17.*/
+    { 18, 19, 19, 10,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 18, 19, 19, 10,  3,256, ZSTD_btultra },  /* level 19.*/
+    { 18, 19, 19, 11,  3,512, ZSTD_btultra },  /* level 20.*/
+    { 18, 19, 19, 12,  3,512, ZSTD_btultra },  /* level 21.*/
+    { 18, 19, 19, 13,  3,999, ZSTD_btultra },  /* level 22.*/
 },
 {   /* for srcSize <= 128 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* level  0 - not used */
-    { 17, 12, 13,  1,  6,  1, ZSTD_fast    },  /* level  1 */
-    { 17, 13, 16,  1,  5,  1, ZSTD_fast    },  /* level  2 */
-    { 17, 16, 16,  2,  5,  8, ZSTD_dfast   },  /* level  3 */
-    { 17, 13, 15,  3,  4,  8, ZSTD_greedy  },  /* level  4 */
-    { 17, 15, 17,  4,  4,  8, ZSTD_greedy  },  /* level  5 */
-    { 17, 16, 17,  3,  4,  8, ZSTD_lazy    },  /* level  6 */
-    { 17, 15, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 17, 12, 13,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 15,  1,  5,  0, ZSTD_fast    },  /* level  2 */
+    { 17, 15, 16,  2,  5,  1, ZSTD_dfast   },  /* level  3 */
+    { 17, 17, 17,  2,  4,  1, ZSTD_dfast   },  /* level  4 */
+    { 17, 16, 17,  3,  4,  2, ZSTD_greedy  },  /* level  5 */
+    { 17, 17, 17,  3,  4,  4, ZSTD_lazy    },  /* level  6 */
+    { 17, 17, 17,  3,  4,  8, ZSTD_lazy2   },  /* level  7 */
     { 17, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
     { 17, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
     { 17, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
     { 17, 17, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 11 */
-    { 17, 17, 17,  8,  4,  8, ZSTD_lazy2   },  /* level 12 */
-    { 17, 18, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13.*/
-    { 17, 17, 17,  7,  3,  8, ZSTD_btopt   },  /* level 14.*/
-    { 17, 17, 17,  7,  3, 16, ZSTD_btopt   },  /* level 15.*/
-    { 17, 18, 17,  7,  3, 32, ZSTD_btopt   },  /* level 16.*/
-    { 17, 18, 17,  7,  3, 64, ZSTD_btopt   },  /* level 17.*/
-    { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 18.*/
-    { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 17, 18, 17,  6,  4, 16, ZSTD_btlazy2 },  /* level 12 */
+    { 17, 18, 17,  8,  4, 16, ZSTD_btlazy2 },  /* level 13.*/
+    { 17, 18, 17,  4,  4, 32, ZSTD_btopt   },  /* level 14.*/
+    { 17, 18, 17,  6,  3, 64, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  7,  3,128, ZSTD_btopt   },  /* level 16.*/
+    { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 17.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btultra },  /* level 19.*/
     { 17, 18, 17,  9,  3,256, ZSTD_btultra },  /* level 20.*/
     { 17, 18, 17, 10,  3,256, ZSTD_btultra },  /* level 21.*/
     { 17, 18, 17, 11,  3,512, ZSTD_btultra },  /* level 22.*/
@@ -3391,28 +3982,28 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
 {   /* for srcSize <= 16 KB */
     /* W,  C,  H,  S,  L,  T, strat */
     { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
-    { 14, 14, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
-    { 14, 14, 14,  1,  4,  1, ZSTD_fast    },  /* level  2 */
-    { 14, 14, 14,  1,  4,  6, ZSTD_dfast   },  /* level  3.*/
-    { 14, 14, 14,  4,  4,  6, ZSTD_greedy  },  /* level  4.*/
-    { 14, 14, 14,  3,  4,  6, ZSTD_lazy    },  /* level  5.*/
-    { 14, 14, 14,  4,  4,  6, ZSTD_lazy2   },  /* level  6 */
-    { 14, 14, 14,  5,  4,  6, ZSTD_lazy2   },  /* level  7 */
-    { 14, 14, 14,  6,  4,  6, ZSTD_lazy2   },  /* level  8.*/
-    { 14, 15, 14,  6,  4,  6, ZSTD_btlazy2 },  /* level  9.*/
-    { 14, 15, 14,  3,  3,  6, ZSTD_btopt   },  /* level 10.*/
-    { 14, 15, 14,  6,  3,  8, ZSTD_btopt   },  /* level 11.*/
+    { 14, 14, 15,  1,  5,  0, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 15,  1,  4,  0, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 14,  2,  4,  1, ZSTD_dfast   },  /* level  3.*/
+    { 14, 14, 14,  4,  4,  2, ZSTD_greedy  },  /* level  4.*/
+    { 14, 14, 14,  3,  4,  4, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  8, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  6,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  8,  4,  8, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  5,  4,  8, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  9,  4,  8, ZSTD_btlazy2 },  /* level 10.*/
+    { 14, 15, 14,  3,  4, 12, ZSTD_btopt   },  /* level 11.*/
     { 14, 15, 14,  6,  3, 16, ZSTD_btopt   },  /* level 12.*/
     { 14, 15, 14,  6,  3, 24, ZSTD_btopt   },  /* level 13.*/
     { 14, 15, 15,  6,  3, 48, ZSTD_btopt   },  /* level 14.*/
     { 14, 15, 15,  6,  3, 64, ZSTD_btopt   },  /* level 15.*/
     { 14, 15, 15,  6,  3, 96, ZSTD_btopt   },  /* level 16.*/
     { 14, 15, 15,  6,  3,128, ZSTD_btopt   },  /* level 17.*/
-    { 14, 15, 15,  6,  3,256, ZSTD_btopt   },  /* level 18.*/
-    { 14, 15, 15,  7,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 14, 15, 15,  6,  3,256, ZSTD_btultra },  /* level 19.*/
     { 14, 15, 15,  8,  3,256, ZSTD_btultra },  /* level 20.*/
     { 14, 15, 15,  9,  3,256, ZSTD_btultra },  /* level 21.*/
-    { 14, 15, 15, 10,  3,256, ZSTD_btultra },  /* level 22.*/
+    { 14, 15, 15, 10,  3,512, ZSTD_btultra },  /* level 22.*/
 },
 };
 
diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h
index 81f12ca6df5e..43f7c1486a9c 100644
--- a/lib/compress/zstd_compress_internal.h
+++ b/lib/compress/zstd_compress_internal.h
@@ -27,6 +27,7 @@
 extern "C" {
 #endif
 
+
 /*-*************************************
 *  Constants
 ***************************************/
@@ -37,7 +38,8 @@ extern "C" {
                                        It's not a big deal though : candidate will just be sorted again.
                                        Additionnally, candidate position 1 will be lost.
                                        But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
-                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be misdhandled after table re-use with a different strategy */
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be misdhandled after table re-use with a different strategy
+                                       Constant required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
 
 
 /*-*************************************
@@ -46,6 +48,12 @@ extern "C" {
 typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
 typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
 
+typedef enum {
+    ZSTD_dictDefaultAttach = 0,
+    ZSTD_dictForceAttach = 1,
+    ZSTD_dictForceCopy = -1,
+} ZSTD_dictAttachPref_e;
+
 typedef struct ZSTD_prefixDict_s {
     const void* dict;
     size_t dictSize;
@@ -53,14 +61,22 @@ typedef struct ZSTD_prefixDict_s {
 } ZSTD_prefixDict;
 
 typedef struct {
-    U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
+    U32 CTable[HUF_CTABLE_SIZE_U32(255)];
+    HUF_repeat repeatMode;
+} ZSTD_hufCTables_t;
+
+typedef struct {
     FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
     FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
     FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
-    HUF_repeat hufCTable_repeatMode;
     FSE_repeat offcode_repeatMode;
     FSE_repeat matchlength_repeatMode;
     FSE_repeat litlength_repeatMode;
+} ZSTD_fseCTables_t;
+
+typedef struct {
+    ZSTD_hufCTables_t huf;
+    ZSTD_fseCTables_t fse;
 } ZSTD_entropyCTables_t;
 
 typedef struct {
@@ -76,26 +92,27 @@ typedef struct {
     U32 rep[ZSTD_REP_NUM];
 } ZSTD_optimal_t;
 
+typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
+
 typedef struct {
     /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
-    U32* litFreq;               /* table of literals statistics, of size 256 */
-    U32* litLengthFreq;         /* table of litLength statistics, of size (MaxLL+1) */
-    U32* matchLengthFreq;       /* table of matchLength statistics, of size (MaxML+1) */
-    U32* offCodeFreq;           /* table of offCode statistics, of size (MaxOff+1) */
-    ZSTD_match_t* matchTable;   /* list of found matches, of size ZSTD_OPT_NUM+1 */
-    ZSTD_optimal_t* priceTable; /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+    U32* litFreq;                /* table of literals statistics, of size 256 */
+    U32* litLengthFreq;          /* table of litLength statistics, of size (MaxLL+1) */
+    U32* matchLengthFreq;        /* table of matchLength statistics, of size (MaxML+1) */
+    U32* offCodeFreq;            /* table of offCode statistics, of size (MaxOff+1) */
+    ZSTD_match_t* matchTable;    /* list of found matches, of size ZSTD_OPT_NUM+1 */
+    ZSTD_optimal_t* priceTable;  /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
 
     U32  litSum;                 /* nb of literals */
     U32  litLengthSum;           /* nb of litLength codes */
     U32  matchLengthSum;         /* nb of matchLength codes */
     U32  offCodeSum;             /* nb of offset codes */
-    /* begin updated by ZSTD_setLog2Prices */
-    U32  log2litSum;             /* pow2 to compare log2(litfreq) to */
-    U32  log2litLengthSum;       /* pow2 to compare log2(llfreq) to */
-    U32  log2matchLengthSum;     /* pow2 to compare log2(mlfreq) to */
-    U32  log2offCodeSum;         /* pow2 to compare log2(offreq) to */
-    /* end : updated by ZSTD_setLog2Prices */
-    U32  staticPrices;           /* prices follow a pre-defined cost structure, statistics are irrelevant */
+    U32  litSumBasePrice;        /* to compare to log2(litfreq) */
+    U32  litLengthSumBasePrice;  /* to compare to log2(llfreq)  */
+    U32  matchLengthSumBasePrice;/* to compare to log2(mlfreq)  */
+    U32  offCodeSumBasePrice;    /* to compare to log2(offreq)  */
+    ZSTD_OptPrice_e priceType;   /* prices can be determined dynamically, or follow a pre-defined cost structure */
+    const ZSTD_entropyCTables_t* symbolCosts;  /* pre-calculated dictionary statistics */
 } optState_t;
 
 typedef struct {
@@ -111,17 +128,20 @@ typedef struct {
     U32 lowLimit;           /* below that point, no more data */
 } ZSTD_window_t;
 
-typedef struct {
-    ZSTD_window_t window;      /* State for window round buffer management */
-    U32 loadedDictEnd;         /* index of end of dictionary */
-    U32 nextToUpdate;          /* index from which to continue table update */
-    U32 nextToUpdate3;         /* index from which to continue table update */
-    U32 hashLog3;              /* dispatch table : larger == faster, more memory */
+typedef struct ZSTD_matchState_t ZSTD_matchState_t;
+struct ZSTD_matchState_t {
+    ZSTD_window_t window;   /* State for window round buffer management */
+    U32 loadedDictEnd;      /* index of end of dictionary */
+    U32 nextToUpdate;       /* index from which to continue table update */
+    U32 nextToUpdate3;      /* index from which to continue table update */
+    U32 hashLog3;           /* dispatch table : larger == faster, more memory */
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
     optState_t opt;         /* optimal parser state */
-} ZSTD_matchState_t;
+    const ZSTD_matchState_t *dictMatchState;
+    ZSTD_compressionParameters cParams;
+};
 
 typedef struct {
     ZSTD_compressedBlockState_t* prevCBlock;
@@ -161,7 +181,7 @@ typedef struct {
   rawSeq* seq;     /* The start of the sequences */
   size_t pos;      /* The position where reading stopped. <= size. */
   size_t size;     /* The number of sequences. <= capacity. */
-  size_t capacity; /* The capacity of the `seq` pointer */
+  size_t capacity; /* The capacity starting from `seq` pointer */
 } rawSeqStore_t;
 
 struct ZSTD_CCtx_params_s {
@@ -170,10 +190,11 @@ struct ZSTD_CCtx_params_s {
     ZSTD_frameParameters fParams;
 
     int compressionLevel;
-    int disableLiteralCompression;
     int forceWindow;           /* force back-references to respect limit of
                                 * 1<<wLog, even for dictionary */
 
+    ZSTD_dictAttachPref_e attachDictPref;
+
     /* Multithreading: used to pass parameters to mtctx */
     unsigned nbWorkers;
     unsigned jobSize;
@@ -193,6 +214,8 @@ struct ZSTD_CCtx_s {
     ZSTD_CCtx_params requestedParams;
     ZSTD_CCtx_params appliedParams;
     U32   dictID;
+
+    int workSpaceOversizedDuration;
     void* workSpace;
     size_t workSpaceSize;
     size_t blockSize;
@@ -235,11 +258,15 @@ struct ZSTD_CCtx_s {
 #endif
 };
 
+typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
+
+typedef enum { ZSTD_noDict = 0, ZSTD_extDict = 1, ZSTD_dictMatchState = 2 } ZSTD_dictMode_e;
+
 
 typedef size_t (*ZSTD_blockCompressor) (
         ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict);
+        void const* src, size_t srcSize);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode);
 
 
 MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
@@ -280,16 +307,18 @@ MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
 */
 MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t mlBase)
 {
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 6)
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL >= 6)
     static const BYTE* g_start = NULL;
     if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
     {   U32 const pos = (U32)((const BYTE*)literals - g_start);
-        DEBUGLOG(6, "Cpos%7u :%3u literals, match%3u bytes at dist.code%7u",
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offCode%7u",
                pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offsetCode);
     }
 #endif
+    assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
     /* copy Literals */
-    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + 128 KB);
+    assert(seqStorePtr->maxNbLit <= 128 KB);
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
     ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
     seqStorePtr->lit += litLength;
 
@@ -420,6 +449,11 @@ ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
     const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
     size_t const matchLength = ZSTD_count(ip, match, vEnd);
     if (match + matchLength != mEnd) return matchLength;
+    DEBUGLOG(7, "ZSTD_count_2segments: found a 2-parts match (current length==%zu)", matchLength);
+    DEBUGLOG(7, "distance from match beginning to end dictionary = %zi", mEnd - match);
+    DEBUGLOG(7, "distance from current pos to end buffer = %zi", iEnd - ip);
+    DEBUGLOG(7, "next byte : ip==%02X, istart==%02X", ip[matchLength], *iStart);
+    DEBUGLOG(7, "final match length = %zu", matchLength + ZSTD_count(ip+matchLength, iStart, iEnd));
     return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
 }
 
@@ -497,6 +531,20 @@ MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
 }
 
 /**
+ * ZSTD_matchState_dictMode():
+ * Inspects the provided matchState and figures out what dictMode should be
+ * passed to the compressor.
+ */
+MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
+{
+    return ZSTD_window_hasExtDict(ms->window) ?
+        ZSTD_extDict :
+        ms->dictMatchState != NULL ?
+            ZSTD_dictMatchState :
+            ZSTD_noDict;
+}
+
+/**
  * ZSTD_window_needOverflowCorrection():
  * Returns non-zero if the indices are getting too large and need overflow
  * protection.
@@ -563,31 +611,41 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
  * ZSTD_window_enforceMaxDist():
  * Updates lowLimit so that:
  *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
+ *
  * This allows a simple check that index >= lowLimit to see if index is valid.
  * This must be called before a block compression call, with srcEnd as the block
  * source end.
+ *
  * If loadedDictEndPtr is not NULL, we set it to zero once we update lowLimit.
  * This is because dictionaries are allowed to be referenced as long as the last
  * byte of the dictionary is in the window, but once they are out of range,
  * they cannot be referenced. If loadedDictEndPtr is NULL, we use
  * loadedDictEnd == 0.
+ *
+ * In normal dict mode, the dict is between lowLimit and dictLimit. In
+ * dictMatchState mode, lowLimit and dictLimit are the same, and the dictionary
+ * is below them. forceWindow and dictMatchState are therefore incompatible.
  */
 MEM_STATIC void ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
                                            void const* srcEnd, U32 maxDist,
-                                           U32* loadedDictEndPtr)
+                                           U32* loadedDictEndPtr,
+                                           const ZSTD_matchState_t** dictMatchStatePtr)
 {
     U32 const current = (U32)((BYTE const*)srcEnd - window->base);
     U32 loadedDictEnd = loadedDictEndPtr != NULL ? *loadedDictEndPtr : 0;
+    DEBUGLOG(5, "ZSTD_window_enforceMaxDist: current=%u, maxDist=%u", current, maxDist);
     if (current > maxDist + loadedDictEnd) {
         U32 const newLowLimit = current - maxDist;
         if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
         if (window->dictLimit < window->lowLimit) {
-            DEBUGLOG(5, "Update dictLimit from %u to %u", window->dictLimit,
-                     window->lowLimit);
+            DEBUGLOG(5, "Update dictLimit to match lowLimit, from %u to %u",
+                        window->dictLimit, window->lowLimit);
             window->dictLimit = window->lowLimit;
         }
         if (loadedDictEndPtr)
             *loadedDictEndPtr = 0;
+        if (dictMatchStatePtr)
+            *dictMatchStatePtr = NULL;
     }
 }
 
@@ -603,12 +661,12 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
 {
     BYTE const* const ip = (BYTE const*)src;
     U32 contiguous = 1;
+    DEBUGLOG(5, "ZSTD_window_update");
     /* Check if blocks follow each other */
     if (src != window->nextSrc) {
         /* not contiguous */
         size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
-        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u",
-                 window->dictLimit);
+        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u", window->dictLimit);
         window->lowLimit = window->dictLimit;
         assert(distanceFromBase == (size_t)(U32)distanceFromBase);  /* should never overflow */
         window->dictLimit = (U32)distanceFromBase;
@@ -625,10 +683,38 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
         ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
         U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
         window->lowLimit = lowLimitMax;
+        DEBUGLOG(5, "Overlapping extDict and input : new lowLimit = %u", window->lowLimit);
     }
     return contiguous;
 }
 
+
+/* debug functions */
+
+MEM_STATIC double ZSTD_fWeight(U32 rawStat)
+{
+    U32 const fp_accuracy = 8;
+    U32 const fp_multiplier = (1 << fp_accuracy);
+    U32 const stat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(stat);
+    U32 const BWeight = hb * fp_multiplier;
+    U32 const FWeight = (stat << fp_accuracy) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + fp_accuracy < 31);
+    return (double)weight / fp_multiplier;
+}
+
+MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
+{
+    unsigned u, sum;
+    for (u=0, sum=0; u<=max; u++) sum += table[u];
+    DEBUGLOG(2, "total nb elts: %u", sum);
+    for (u=0; u<=max; u++) {
+        DEBUGLOG(2, "%2u: %5u  (%.2f)",
+                u, table[u], ZSTD_fWeight(sum) - ZSTD_fWeight(table[u]) );
+    }
+}
+
 #if defined (__cplusplus)
 }
 #endif
@@ -640,7 +726,7 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
  * ============================================================== */
 
 /* ZSTD_getCParamsFromCCtxParams() :
- * cParams are built depending on compressionLevel, src size hints, 
+ * cParams are built depending on compressionLevel, src size hints,
  * LDM and manually set compression parameters.
  */
 ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
@@ -656,6 +742,8 @@ size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
                      const ZSTD_CDict* cdict,
                      ZSTD_CCtx_params  params, unsigned long long pledgedSrcSize);
 
+void ZSTD_resetSeqStore(seqStore_t* ssPtr);
+
 /*! ZSTD_compressStream_generic() :
  *  Private use only. To be called from zstdmt_compress.c in single-thread mode. */
 size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
@@ -672,6 +760,7 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
 size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
                                     const void* dict, size_t dictSize,
                                     ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
                                     const ZSTD_CDict* cdict,
                                     ZSTD_CCtx_params params,
                                     unsigned long long pledgedSrcSize);
diff --git a/lib/compress/zstd_double_fast.c b/lib/compress/zstd_double_fast.c
index 86e6b39621b2..7b9e18e7ed9d 100644
--- a/lib/compress/zstd_double_fast.c
+++ b/lib/compress/zstd_double_fast.c
@@ -13,9 +13,9 @@
 
 
 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
-                              ZSTD_compressionParameters const* cParams,
-                              void const* end)
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashLarge = ms->hashTable;
     U32  const hBitsL = cParams->hashLog;
     U32  const mls = cParams->searchLength;
@@ -40,6 +40,9 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
                 hashSmall[smHash] = current + i;
             if (i == 0 || hashLarge[lgHash] == 0)
                 hashLarge[lgHash] = current + i;
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
         }
     }
 }
@@ -48,9 +51,10 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_doubleFast_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
-        U32 const mls /* template */)
+        void const* src, size_t srcSize,
+        U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
 {
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
     U32* const hashLong = ms->hashTable;
     const U32 hBitsL = cParams->hashLog;
     U32* const hashSmall = ms->chainTable;
@@ -59,70 +63,188 @@ size_t ZSTD_compressBlock_doubleFast_generic(
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32 lowestIndex = ms->window.dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
+    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
     U32 offsetSaved = 0;
 
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams =
+                                     dictMode == ZSTD_dictMatchState ?
+                                     &dms->cParams : NULL;
+    const U32* const dictHashLong  = dictMode == ZSTD_dictMatchState ?
+                                     dms->hashTable : NULL;
+    const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
+                                     dms->chainTable : NULL;
+    const U32 dictStartIndex       = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.base : NULL;
+    const BYTE* const dictStart    = dictMode == ZSTD_dictMatchState ?
+                                     dictBase + dictStartIndex : NULL;
+    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictHBitsL           = dictMode == ZSTD_dictMatchState ?
+                                     dictCParams->hashLog : hBitsL;
+    const U32 dictHBitsS           = dictMode == ZSTD_dictMatchState ?
+                                     dictCParams->chainLog : hBitsS;
+    const U32 dictAndPrefixLength  = (U32)(ip - prefixLowest + dictEnd - dictStart);
+
+    assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
+
     /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const maxRep = (U32)(ip - prefixLowest);
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
+    if (dictMode == ZSTD_dictMatchState) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
 
     /* Main Search Loop */
     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
         size_t mLength;
+        U32 offset;
         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
         U32 const current = (U32)(ip-base);
         U32 const matchIndexL = hashLong[h2];
-        U32 const matchIndexS = hashSmall[h];
+        U32 matchIndexS = hashSmall[h];
         const BYTE* matchLong = base + matchIndexL;
         const BYTE* match = base + matchIndexS;
+        const U32 repIndex = current + 1 - offset_1;
+        const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+                            && repIndex < prefixLowestIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
         hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
 
-        assert(offset_1 <= current);   /* supposed guaranteed by construction */
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
-            /* favor repcode */
+        /* check dictMatchState repcode */
+        if (dictMode == ZSTD_dictMatchState
+            && ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            goto _match_stored;
+        }
+
+        /* check noDict repcode */
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             ip++;
             ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            U32 offset;
-            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
+            goto _match_stored;
+        }
+
+        if (matchIndexL > prefixLowestIndex) {
+            /* check prefix long match */
+            if (MEM_read64(matchLong) == MEM_read64(ip)) {
                 mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
                 offset = (U32)(ip-matchLong);
-                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
-            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
-                size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-                U32 const matchIndexL3 = hashLong[hl3];
-                const BYTE* matchL3 = base + matchIndexL3;
-                hashLong[hl3] = current + 1;
-                if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) {
+                while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        } else if (dictMode == ZSTD_dictMatchState) {
+            /* check dictMatchState long match */
+            U32 const dictMatchIndexL = dictHashLong[dictHL];
+            const BYTE* dictMatchL = dictBase + dictMatchIndexL;
+            assert(dictMatchL < dictEnd);
+
+            if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
+                mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
+                offset = (U32)(current - dictMatchIndexL - dictIndexDelta);
+                while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        }
+
+        if (matchIndexS > prefixLowestIndex) {
+            /* check prefix short match */
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+            }
+        } else if (dictMode == ZSTD_dictMatchState) {
+            /* check dictMatchState short match */
+            U32 const dictMatchIndexS = dictHashSmall[dictHS];
+            match = dictBase + dictMatchIndexS;
+            matchIndexS = dictMatchIndexS + dictIndexDelta;
+
+            if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+            }
+        }
+
+        ip += ((ip-anchor) >> kSearchStrength) + 1;
+        continue;
+
+_search_next_long:
+
+        {
+            size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+            size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            U32 const matchIndexL3 = hashLong[hl3];
+            const BYTE* matchL3 = base + matchIndexL3;
+            hashLong[hl3] = current + 1;
+
+            /* check prefix long +1 match */
+            if (matchIndexL3 > prefixLowestIndex) {
+                if (MEM_read64(matchL3) == MEM_read64(ip+1)) {
                     mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
                     ip++;
                     offset = (U32)(ip-matchL3);
-                    while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
-                } else {
-                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-                    offset = (U32)(ip-match);
-                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                    while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                    goto _match_found;
+                }
+            } else if (dictMode == ZSTD_dictMatchState) {
+                /* check dict long +1 match */
+                U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
+                const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
+                assert(dictMatchL3 < dictEnd);
+                if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
+                    mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
+                    ip++;
+                    offset = (U32)(current + 1 - dictMatchIndexL3 - dictIndexDelta);
+                    while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
+                    goto _match_found;
                 }
-            } else {
-                ip += ((ip-anchor) >> kSearchStrength) + 1;
-                continue;
             }
+        }
 
-            offset_2 = offset_1;
-            offset_1 = offset;
-
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        /* if no long +1 match, explore the short match we found */
+        if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
+            mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
+            offset = (U32)(current - matchIndexS);
+            while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        } else {
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            offset = (U32)(ip - match);
+            while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
         }
 
+        /* fall-through */
+
+_match_found:
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+_match_stored:
         /* match found */
         ip += mLength;
         anchor = ip;
@@ -135,19 +257,44 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                 hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
 
             /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
-                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
-                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
+            if (dictMode == ZSTD_dictMatchState) {
+                while (ip <= ilimit) {
+                    U32 const current2 = (U32)(ip-base);
+                    U32 const repIndex2 = current2 - offset_2;
+                    const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
+                        && repIndex2 < prefixLowestIndex ?
+                            dictBase - dictIndexDelta + repIndex2 :
+                            base + repIndex2;
+                    if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                       && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                        const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
+                        size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
+                        U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                        ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                        hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                        ip += repLength2;
+                        anchor = ip;
+                        continue;
+                    }
+                    break;
+                }
+            }
+
+            if (dictMode == ZSTD_noDict) {
+                while ( (ip <= ilimit)
+                     && ( (offset_2>0)
+                        & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                    /* store sequence */
+                    size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                    U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                    ip += rLength;
+                    anchor = ip;
+                    continue;   /* faster when present ... (?) */
+    }   }   }   }
 
     /* save reps for next block */
     rep[0] = offset_1 ? offset_1 : offsetSaved;
@@ -160,102 +307,126 @@ size_t ZSTD_compressBlock_doubleFast_generic(
 
 size_t ZSTD_compressBlock_doubleFast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
+{
+    const U32 mls = ms->cParams.searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
+    }
+}
+
+
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    const U32 mls = cParams->searchLength;
+    const U32 mls = ms->cParams.searchLength;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
     }
 }
 
 
 static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+        void const* src, size_t srcSize,
         U32 const mls /* template */)
 {
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
     U32* const hashLong = ms->hashTable;
     U32  const hBitsL = cParams->hashLog;
     U32* const hashSmall = ms->chainTable;
     U32  const hBitsS = cParams->chainLog;
-    const BYTE* const base = ms->window.base;
-    const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = ms->window.lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ms->window.dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
+    const U32   prefixStartIndex = ms->window.dictLimit;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const U32   dictStartIndex = ms->window.lowLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
     U32 offset_1=rep[0], offset_2=rep[1];
 
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
+
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
         const U32 matchIndex = hashSmall[hSmall];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
         const BYTE* match = matchBase + matchIndex;
 
         const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
         const U32 matchLongIndex = hashLong[hLong];
-        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
+        const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
         const BYTE* matchLong = matchLongBase + matchLongIndex;
 
         const U32 current = (U32)(ip-base);
         const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
         size_t mLength;
         hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
 
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+        if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
+            & (repIndex > dictStartIndex))
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
             ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
         } else {
-            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
-                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
+            if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
                 U32 offset;
-                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
                 offset = current - matchLongIndex;
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
                 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
-            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+            } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
                 U32 const matchIndex3 = hashLong[h3];
-                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
+                const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
                 const BYTE* match3 = match3Base + matchIndex3;
                 U32 offset;
                 hashLong[h3] = current + 1;
-                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
-                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
+                if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
                     ip++;
                     offset = current+1 - matchIndex3;
                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
                 } else {
-                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                    const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
                     offset = current - matchIndex;
                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
                 }
@@ -282,12 +453,13 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
                 U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
-                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3)   /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
+                    & (repIndex2 > dictStartIndex))
+                  && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
                     ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
@@ -309,19 +481,19 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
 
 size_t ZSTD_compressBlock_doubleFast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    U32 const mls = cParams->searchLength;
+    U32 const mls = ms->cParams.searchLength;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
     }
 }
diff --git a/lib/compress/zstd_double_fast.h b/lib/compress/zstd_double_fast.h
index 6d80b2774c0e..4fa31acfc0d6 100644
--- a/lib/compress/zstd_double_fast.h
+++ b/lib/compress/zstd_double_fast.h
@@ -19,14 +19,16 @@ extern "C" {
 #include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
 
 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
-                              ZSTD_compressionParameters const* cParams,
-                              void const* end);
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm);
 size_t ZSTD_compressBlock_doubleFast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_doubleFast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 
 
 #if defined (__cplusplus)
diff --git a/lib/compress/zstd_fast.c b/lib/compress/zstd_fast.c
index df4d28b3402b..247746517cd5 100644
--- a/lib/compress/zstd_fast.c
+++ b/lib/compress/zstd_fast.c
@@ -13,9 +13,9 @@
 
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
-                        ZSTD_compressionParameters const* cParams,
-                        void const* end)
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
     U32  const hBits = cParams->hashLog;
     U32  const mls = cParams->searchLength;
@@ -34,6 +34,9 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
             size_t const hash = ZSTD_hashPtr(ip + i, hBits, mls);
             if (i == 0 || hashTable[hash] == 0)
                 hashTable[hash] = current + i;
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
         }
     }
 }
@@ -42,26 +45,65 @@ FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_fast_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
-        U32 const hlog, U32 const stepSize, U32 const mls)
+        U32 const mls, ZSTD_dictMode_e const dictMode)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
     const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = ms->window.dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
+    const U32   prefixStartIndex = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
     U32 offsetSaved = 0;
 
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams =
+                                     dictMode == ZSTD_dictMatchState ?
+                                     &dms->cParams : NULL;
+    const U32* const dictHashTable = dictMode == ZSTD_dictMatchState ?
+                                     dms->hashTable : NULL;
+    const U32 dictStartIndex       = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.base : NULL;
+    const BYTE* const dictStart    = dictMode == ZSTD_dictMatchState ?
+                                     dictBase + dictStartIndex : NULL;
+    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+                                     prefixStartIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
+    const U32 dictHLog             = dictMode == ZSTD_dictMatchState ?
+                                     dictCParams->hashLog : hlog;
+
+    assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
+
+    /* otherwise, we would get index underflow when translating a dict index
+     * into a local index */
+    assert(dictMode != ZSTD_dictMatchState
+        || prefixStartIndex >= (U32)(dictEnd - dictBase));
+
     /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const maxRep = (U32)(ip - prefixStart);
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
+    if (dictMode == ZSTD_dictMatchState) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
 
     /* Main Search Loop */
     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
@@ -70,26 +112,67 @@ size_t ZSTD_compressBlock_fast_generic(
         U32 const current = (U32)(ip-base);
         U32 const matchIndex = hashTable[h];
         const BYTE* match = base + matchIndex;
+        const U32 repIndex = current + 1 - offset_1;
+        const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+                            && repIndex < prefixStartIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
         hashTable[h] = current;   /* update hash table */
 
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+        if ( (dictMode == ZSTD_dictMatchState)
+          && ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else if ( dictMode == ZSTD_noDict
+                 && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             ip++;
             ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            if ( (matchIndex <= lowestIndex)
-              || (MEM_read32(match) != MEM_read32(ip)) ) {
+        } else if ( (matchIndex <= prefixStartIndex) ) {
+            if (dictMode == ZSTD_dictMatchState) {
+                size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
+                U32 const dictMatchIndex = dictHashTable[dictHash];
+                const BYTE* dictMatch = dictBase + dictMatchIndex;
+                if (dictMatchIndex <= dictStartIndex ||
+                    MEM_read32(dictMatch) != MEM_read32(ip)) {
+                    assert(stepSize >= 1);
+                    ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+                    continue;
+                } else {
+                    /* found a dict match */
+                    U32 const offset = (U32)(current-dictMatchIndex-dictIndexDelta);
+                    mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
+                    while (((ip>anchor) & (dictMatch>dictStart))
+                         && (ip[-1] == dictMatch[-1])) {
+                        ip--; dictMatch--; mLength++;
+                    } /* catch up */
+                    offset_2 = offset_1;
+                    offset_1 = offset;
+                    ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                }
+            } else {
                 assert(stepSize >= 1);
                 ip += ((ip-anchor) >> kSearchStrength) + stepSize;
                 continue;
             }
+        } else if (MEM_read32(match) != MEM_read32(ip)) {
+            /* it's not a match, and we're not going to check the dictionary */
+            assert(stepSize >= 1);
+            ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+            continue;
+        } else {
+            /* found a regular match */
+            U32 const offset = (U32)(ip-match);
             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-            {   U32 const offset = (U32)(ip-match);
-                while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
-                offset_2 = offset_1;
-                offset_1 = offset;
-                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-        }   }
+            while (((ip>anchor) & (match>prefixStart))
+                 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+            offset_2 = offset_1;
+            offset_1 = offset;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }
 
         /* match found */
         ip += mLength;
@@ -97,21 +180,46 @@ size_t ZSTD_compressBlock_fast_generic(
 
         if (ip <= ilimit) {
             /* Fill Table */
+            assert(base+current+2 > istart);  /* check base overflow */
             hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;  /* here because current+2 could be > iend-8 */
             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+
             /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
-                hashTable[ZSTD_hashPtr(ip, hlog, mls)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
+            if (dictMode == ZSTD_dictMatchState) {
+                while (ip <= ilimit) {
+                    U32 const current2 = (U32)(ip-base);
+                    U32 const repIndex2 = current2 - offset_2;
+                    const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
+                            dictBase - dictIndexDelta + repIndex2 :
+                            base + repIndex2;
+                    if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                       && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                        const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                        size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                        U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                        ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                        hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
+                        ip += repLength2;
+                        anchor = ip;
+                        continue;
+                    }
+                    break;
+                }
+            }
+
+            if (dictMode == ZSTD_noDict) {
+                while ( (ip <= ilimit)
+                     && ( (offset_2>0)
+                        & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                    /* store sequence */
+                    size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                    U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = (U32)(ip-base);
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                    ip += rLength;
+                    anchor = ip;
+                    continue;   /* faster when present ... (?) */
+    }   }   }   }
 
     /* save reps for next block */
     rep[0] = offset_1 ? offset_1 : offsetSaved;
@@ -124,42 +232,66 @@ size_t ZSTD_compressBlock_fast_generic(
 
 size_t ZSTD_compressBlock_fast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    U32 const hlog = cParams->hashLog;
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32 const mls = cParams->searchLength;
+    assert(ms->dictMatchState == NULL);
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
+    case 5 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
+    case 6 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
+    case 7 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
+    }
+}
+
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
     U32 const mls = cParams->searchLength;
-    U32 const stepSize = cParams->targetLength;
+    assert(ms->dictMatchState != NULL);
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
     case 5 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
     case 6 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
     case 7 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
     }
 }
 
 
 static size_t ZSTD_compressBlock_fast_extDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        void const* src, size_t srcSize,
-        U32 const hlog, U32 const stepSize, U32 const mls)
+        void const* src, size_t srcSize, U32 const mls)
 {
-    U32* hashTable = ms->hashTable;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
     const BYTE* const base = ms->window.base;
     const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = ms->window.lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ms->window.dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32   dictStartIndex = ms->window.lowLimit;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const U32   prefixStartIndex = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
     U32 offset_1=rep[0], offset_2=rep[1];
@@ -167,33 +299,34 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
         const size_t h = ZSTD_hashPtr(ip, hlog, mls);
-        const U32 matchIndex = hashTable[h];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
-        const BYTE* match = matchBase + matchIndex;
-        const U32 current = (U32)(ip-base);
-        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
+        const U32    matchIndex = hashTable[h];
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
+        const BYTE*  match = matchBase + matchIndex;
+        const U32    current = (U32)(ip-base);
+        const U32    repIndex = current + 1 - offset_1;
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
         size_t mLength;
         hashTable[h] = current;   /* update hash table */
+        assert(offset_1 <= current +1);   /* check repIndex */
 
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+        if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex))
            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
             ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
         } else {
-            if ( (matchIndex < lowestIndex) ||
+            if ( (matchIndex < dictStartIndex) ||
                  (MEM_read32(match) != MEM_read32(ip)) ) {
                 assert(stepSize >= 1);
                 ip += ((ip-anchor) >> kSearchStrength) + stepSize;
                 continue;
             }
-            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+            {   const BYTE* matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
                 U32 offset;
-                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
                 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
                 offset = current - matchIndex;
                 offset_2 = offset_1;
@@ -213,11 +346,11 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
                 U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (repIndex2 > dictStartIndex))  /* intentional overflow */
                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
                     ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
                     hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
@@ -239,21 +372,20 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
 
 size_t ZSTD_compressBlock_fast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    U32 const hlog = cParams->hashLog;
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
     U32 const mls = cParams->searchLength;
-    U32 const stepSize = cParams->targetLength;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
     case 5 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
     case 6 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
     case 7 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
     }
 }
diff --git a/lib/compress/zstd_fast.h b/lib/compress/zstd_fast.h
index f0438ad5b411..b74a88c57c81 100644
--- a/lib/compress/zstd_fast.h
+++ b/lib/compress/zstd_fast.h
@@ -19,14 +19,16 @@ extern "C" {
 #include "zstd_compress_internal.h"
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
-                        ZSTD_compressionParameters const* cParams,
-                        void const* end);
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm);
 size_t ZSTD_compressBlock_fast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_fast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 
 #if defined (__cplusplus)
 }
diff --git a/lib/compress/zstd_lazy.c b/lib/compress/zstd_lazy.c
index 9f158123f077..af615e07763d 100644
--- a/lib/compress/zstd_lazy.c
+++ b/lib/compress/zstd_lazy.c
@@ -16,11 +16,12 @@
 *  Binary Tree search
 ***************************************/
 
-void ZSTD_updateDUBT(
-                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+static void
+ZSTD_updateDUBT(ZSTD_matchState_t* ms,
                 const BYTE* ip, const BYTE* iend,
                 U32 mls)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
     U32  const hashLog = cParams->hashLog;
 
@@ -59,11 +60,12 @@ void ZSTD_updateDUBT(
  *  sort one already inserted but unsorted position
  *  assumption : current >= btlow == (current - btmask)
  *  doesn't fail */
-static void ZSTD_insertDUBT1(
-                 ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+static void
+ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
                  U32 current, const BYTE* inputEnd,
-                 U32 nbCompares, U32 btLow, int extDict)
+                 U32 nbCompares, U32 btLow, const ZSTD_dictMode_e dictMode)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32*   const bt = ms->chainTable;
     U32    const btLog  = cParams->chainLog - 1;
     U32    const btMask = (1 << btLog) - 1;
@@ -92,10 +94,12 @@ static void ZSTD_insertDUBT1(
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         assert(matchIndex < current);
 
-        if ( (!extDict)
+        if ( (dictMode != ZSTD_extDict)
           || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
           || (current < dictLimit) /* both in extDict */) {
-            const BYTE* const mBase = !extDict || ((matchIndex+matchLength) >= dictLimit) ? base : dictBase;
+            const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
+                                     || (matchIndex+matchLength >= dictLimit)) ?
+                                        base : dictBase;
             assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
                  || (current < dictLimit) );
             match = mBase + matchIndex;
@@ -138,13 +142,92 @@ static void ZSTD_insertDUBT1(
 }
 
 
-static size_t ZSTD_DUBT_findBestMatch (
-                            ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                            const BYTE* const ip, const BYTE* const iend,
-                            size_t* offsetPtr,
-                            U32 const mls,
-                            U32 const extDict)
+static size_t
+ZSTD_DUBT_findBetterDictMatch (
+        ZSTD_matchState_t* ms,
+        const BYTE* const ip, const BYTE* const iend,
+        size_t* offsetPtr,
+        size_t bestLength,
+        U32 nbCompares,
+        U32 const mls,
+        const ZSTD_dictMode_e dictMode)
 {
+    const ZSTD_matchState_t * const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dmsCParams = &dms->cParams;
+    const U32 * const dictHashTable = dms->hashTable;
+    U32         const hashLog = dmsCParams->hashLog;
+    size_t      const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32               dictMatchIndex = dictHashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    U32         const current = (U32)(ip-base);
+    const BYTE* const dictBase = dms->window.base;
+    const BYTE* const dictEnd = dms->window.nextSrc;
+    U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
+    U32         const dictLowLimit = dms->window.lowLimit;
+    U32         const dictIndexDelta = ms->window.lowLimit - dictHighLimit;
+
+    U32*        const dictBt = dms->chainTable;
+    U32         const btLog  = dmsCParams->chainLog - 1;
+    U32         const btMask = (1 << btLog) - 1;
+    U32         const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask;
+
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+
+    (void)dictMode;
+    assert(dictMode == ZSTD_dictMatchState);
+
+    while (nbCompares-- && (dictMatchIndex > dictLowLimit)) {
+        U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match = dictBase + dictMatchIndex;
+        matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+        if (dictMatchIndex+matchLength >= dictHighLimit)
+            match = base + dictMatchIndex + dictIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+        if (matchLength > bestLength) {
+            U32 matchIndex = dictMatchIndex + dictIndexDelta;
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
+                DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
+                    current, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, ZSTD_REP_MOVE + current - matchIndex, dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+            }
+            if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
+                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthLarger = matchLength;
+            dictMatchIndex = nextPtr[0];
+        }
+    }
+
+    if (bestLength >= MINMATCH) {
+        U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+        DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                    current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+    }
+    return bestLength;
+
+}
+
+
+static size_t
+ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iend,
+                        size_t* offsetPtr,
+                        U32 const mls,
+                        const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32*   const hashTable = ms->hashTable;
     U32    const hashLog = cParams->hashLog;
     size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
@@ -195,8 +278,8 @@ static size_t ZSTD_DUBT_findBestMatch (
     while (matchIndex) {  /* will end on matchIndex == 0 */
         U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
         U32 const nextCandidateIdx = *nextCandidateIdxPtr;
-        ZSTD_insertDUBT1(ms, cParams, matchIndex, iend,
-                         nbCandidates, unsortLimit, extDict);
+        ZSTD_insertDUBT1(ms, matchIndex, iend,
+                         nbCandidates, unsortLimit, dictMode);
         matchIndex = nextCandidateIdx;
         nbCandidates++;
     }
@@ -221,7 +304,7 @@ static size_t ZSTD_DUBT_findBestMatch (
             size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
             const BYTE* match;
 
-            if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) {
                 match = base + matchIndex;
                 matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
             } else {
@@ -237,6 +320,11 @@ static size_t ZSTD_DUBT_findBestMatch (
                 if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
                     bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
                 if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    if (dictMode == ZSTD_dictMatchState) {
+                        nbCompares = 0; /* in addition to avoiding checking any
+                                         * further in this loop, make sure we
+                                         * skip checking in the dictionary. */
+                    }
                     break;   /* drop, to guarantee consistency (miss a little bit of compression) */
                 }
             }
@@ -259,6 +347,13 @@ static size_t ZSTD_DUBT_findBestMatch (
 
         *smallerPtr = *largerPtr = 0;
 
+        if (dictMode == ZSTD_dictMatchState && nbCompares) {
+            bestLength = ZSTD_DUBT_findBetterDictMatch(
+                    ms, ip, iend,
+                    offsetPtr, bestLength, nbCompares,
+                    mls, dictMode);
+        }
+
         assert(matchEndIdx > current+8); /* ensure nextToUpdate is increased */
         ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
         if (bestLength >= MINMATCH) {
@@ -272,61 +367,64 @@ static size_t ZSTD_DUBT_findBestMatch (
 
 
 /** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 mls /* template */)
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iLimit,
+                      size_t* offsetPtr,
+                const U32 mls /* template */,
+                const ZSTD_dictMode_e dictMode)
 {
     DEBUGLOG(7, "ZSTD_BtFindBestMatch");
     if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
-    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 0);
+    ZSTD_updateDUBT(ms, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offsetPtr, mls, dictMode);
 }
 
 
-static size_t ZSTD_BtFindBestMatch_selectMLS (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
+static size_t
+ZSTD_BtFindBestMatch_selectMLS (  ZSTD_matchState_t* ms,
+                            const BYTE* ip, const BYTE* const iLimit,
+                                  size_t* offsetPtr)
 {
-    switch(cParams->searchLength)
+    switch(ms->cParams.searchLength)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 4);
-    case 5 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
     case 7 :
-    case 6 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 6);
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
     }
 }
 
 
-/** Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch_extDict (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 mls)
+static size_t ZSTD_BtFindBestMatch_dictMatchState_selectMLS (
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
 {
-    DEBUGLOG(7, "ZSTD_BtFindBestMatch_extDict");
-    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
-    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 1);
+    switch(ms->cParams.searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
+    }
 }
 
 
-static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+static size_t ZSTD_BtFindBestMatch_extDict_selectMLS (
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
                         size_t* offsetPtr)
 {
-    switch(cParams->searchLength)
+    switch(ms->cParams.searchLength)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 4);
-    case 5 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
     case 7 :
-    case 6 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 6);
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
     }
 }
 
@@ -340,7 +438,8 @@ static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
 /* Update chains up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
 static U32 ZSTD_insertAndFindFirstIndex_internal(
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
+                        const ZSTD_compressionParameters* const cParams,
                         const BYTE* ip, U32 const mls)
 {
     U32* const hashTable  = ms->hashTable;
@@ -362,22 +461,21 @@ static U32 ZSTD_insertAndFindFirstIndex_internal(
     return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
 }
 
-U32 ZSTD_insertAndFindFirstIndex(
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                        const BYTE* ip)
-{
-    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, cParams->searchLength);
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.searchLength);
 }
 
 
 /* inlining is important to hardwire a hot branch (template emulation) */
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_HcFindBestMatch_generic (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
                         size_t* offsetPtr,
-                        const U32 mls, const U32 extDict)
+                        const U32 mls, const ZSTD_dictMode_e dictMode)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const chainTable = ms->chainTable;
     const U32 chainSize = (1 << cParams->chainLog);
     const U32 chainMask = chainSize-1;
@@ -397,7 +495,7 @@ size_t ZSTD_HcFindBestMatch_generic (
 
     for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
         size_t currentMl=0;
-        if ((!extDict) || matchIndex >= dictLimit) {
+        if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
             const BYTE* const match = base + matchIndex;
             if (match[ml] == ip[ml])   /* potentially better */
                 currentMl = ZSTD_count(ip, match, iLimit);
@@ -419,38 +517,87 @@ size_t ZSTD_HcFindBestMatch_generic (
         matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
     }
 
+    if (dictMode == ZSTD_dictMatchState) {
+        const ZSTD_matchState_t* const dms = ms->dictMatchState;
+        const U32* const dmsChainTable = dms->chainTable;
+        const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
+        const U32 dmsChainMask         = dmsChainSize - 1;
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+        const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0;
+
+        matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
+
+        for ( ; (matchIndex>dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
+            size_t currentMl=0;
+            const BYTE* const match = dmsBase + matchIndex;
+            assert(match+4 <= dmsEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = current - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+
+            if (matchIndex <= dmsMinChain) break;
+            matchIndex = dmsChainTable[matchIndex & dmsChainMask];
+        }
+    }
+
     return ml;
 }
 
 
 FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(ms->cParams.searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
+    }
+}
+
+
+static size_t ZSTD_HcFindBestMatch_dictMatchState_selectMLS (
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
                         size_t* offsetPtr)
 {
-    switch(cParams->searchLength)
+    switch(ms->cParams.searchLength)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 0);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 0);
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
     case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 0);
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
     }
 }
 
 
 FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
-                        size_t* const offsetPtr)
+                        size_t* offsetPtr)
 {
-    switch(cParams->searchLength)
+    switch(ms->cParams.searchLength)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 1);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 1);
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
     case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 1);
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
     }
 }
 
@@ -462,30 +609,55 @@ FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_lazy_generic(
                         ZSTD_matchState_t* ms, seqStore_t* seqStore,
                         U32 rep[ZSTD_REP_NUM],
-                        ZSTD_compressionParameters const* cParams,
                         const void* src, size_t srcSize,
-                        const U32 searchMethod, const U32 depth)
+                        const U32 searchMethod, const U32 depth,
+                        ZSTD_dictMode_e const dictMode)
 {
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ms->window.base + ms->window.dictLimit;
+    const BYTE* const base = ms->window.base;
+    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
 
     typedef size_t (*searchMax_f)(
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
-    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
+    searchMax_f const searchMax = dictMode == ZSTD_dictMatchState ?
+        (searchMethod ? ZSTD_BtFindBestMatch_dictMatchState_selectMLS : ZSTD_HcFindBestMatch_dictMatchState_selectMLS) :
+        (searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS);
     U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
 
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const U32 dictLowestIndex      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.base : NULL;
+    const BYTE* const dictLowest   = dictMode == ZSTD_dictMatchState ?
+                                     dictBase + dictLowestIndex : NULL;
+    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictAndPrefixLength = (U32)(ip - prefixLowest + dictEnd - dictLowest);
+
     /* init */
-    ip += (ip==base);
+    ip += (dictAndPrefixLength == 0);
     ms->nextToUpdate3 = ms->nextToUpdate;
-    {   U32 const maxRep = (U32)(ip-base);
+    if (dictMode == ZSTD_noDict) {
+        U32 const maxRep = (U32)(ip - prefixLowest);
         if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
     }
+    if (dictMode == ZSTD_dictMatchState) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
 
     /* Match Loop */
     while (ip < ilimit) {
@@ -494,15 +666,28 @@ size_t ZSTD_compressBlock_lazy_generic(
         const BYTE* start=ip+1;
 
         /* check repCode */
-        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
-            /* repcode : we take it */
+        if (dictMode == ZSTD_dictMatchState) {
+            const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
+            const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+                                && repIndex < prefixLowestIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+                const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                if (depth==0) goto _storeSequence;
+            }
+        }
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             if (depth==0) goto _storeSequence;
         }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+        {   size_t offsetFound = 999999999;
+            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
             if (ml2 > matchLength)
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
@@ -516,15 +701,31 @@ size_t ZSTD_compressBlock_lazy_generic(
         if (depth>=1)
         while (ip<ilimit) {
             ip ++;
-            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+            if ( (dictMode == ZSTD_noDict)
+              && (offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                 size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                 int const gain2 = (int)(mlRep * 3);
                 int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
                 if ((mlRep >= 4) && (gain2 > gain1))
                     matchLength = mlRep, offset = 0, start = ip;
             }
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+            if (dictMode == ZSTD_dictMatchState) {
+                const U32 repIndex = (U32)(ip - base) - offset_1;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+                if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                    int const gain2 = (int)(mlRep * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offset = 0, start = ip;
+                }
+            }
+            {   size_t offset2=999999999;
+                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -535,15 +736,31 @@ size_t ZSTD_compressBlock_lazy_generic(
             /* let's find an even better one */
             if ((depth==2) && (ip<ilimit)) {
                 ip ++;
-                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
-                    size_t const ml2 = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
-                    int const gain2 = (int)(ml2 * 4);
+                if ( (dictMode == ZSTD_noDict)
+                  && (offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(mlRep * 4);
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
-                    if ((ml2 >= 4) && (gain2 > gain1))
-                        matchLength = ml2, offset = 0, start = ip;
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offset = 0, start = ip;
+                }
+                if (dictMode == ZSTD_dictMatchState) {
+                    const U32 repIndex = (U32)(ip - base) - offset_1;
+                    const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+                    if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                        && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                        const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                        size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                        int const gain2 = (int)(mlRep * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        if ((mlRep >= 4) && (gain2 > gain1))
+                            matchLength = mlRep, offset = 0, start = ip;
+                    }
                 }
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                {   size_t offset2=999999999;
+                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -560,9 +777,17 @@ size_t ZSTD_compressBlock_lazy_generic(
          */
         /* catch up */
         if (offset) {
-            while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > base))
-                 && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
-                { start--; matchLength++; }
+            if (dictMode == ZSTD_noDict) {
+                while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > prefixLowest))
+                     && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
+                    { start--; matchLength++; }
+            }
+            if (dictMode == ZSTD_dictMatchState) {
+                U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+                const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
+                const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
+                while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            }
             offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
         }
         /* store sequence */
@@ -573,16 +798,39 @@ _storeSequence:
         }
 
         /* check immediate repcode */
-        while ( ((ip <= ilimit) & (offset_2>0))
-             && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
-            /* store sequence */
-            matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
-            ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
-            ip += matchLength;
-            anchor = ip;
-            continue;   /* faster when present ... (?) */
-    }   }
+        if (dictMode == ZSTD_dictMatchState) {
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex = current2 - offset_2;
+                const BYTE* repMatch = dictMode == ZSTD_dictMatchState
+                    && repIndex < prefixLowestIndex ?
+                        dictBase - dictIndexDelta + repIndex :
+                        base + repIndex;
+                if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
+                    offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+                    ip += matchLength;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+
+        if (dictMode == ZSTD_noDict) {
+            while ( ((ip <= ilimit) & (offset_2>0))
+                 && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+                /* store sequence */
+                matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
 
     /* Save reps for next block */
     rep[0] = offset_1 ? offset_1 : savedOffset;
@@ -595,30 +843,58 @@ _storeSequence:
 
 size_t ZSTD_compressBlock_btlazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 1, 2, ZSTD_noDict);
 }
 
 size_t ZSTD_compressBlock_lazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 2, ZSTD_noDict);
 }
 
 size_t ZSTD_compressBlock_lazy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 1, ZSTD_noDict);
 }
 
 size_t ZSTD_compressBlock_greedy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 0, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 1, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 1, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, 0, 0, ZSTD_dictMatchState);
 }
 
 
@@ -626,7 +902,6 @@ FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_lazy_extDict_generic(
                         ZSTD_matchState_t* ms, seqStore_t* seqStore,
                         U32 rep[ZSTD_REP_NUM],
-                        ZSTD_compressionParameters const* cParams,
                         const void* src, size_t srcSize,
                         const U32 searchMethod, const U32 depth)
 {
@@ -644,9 +919,9 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
     const BYTE* const dictStart  = dictBase + lowestIndex;
 
     typedef size_t (*searchMax_f)(
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
-    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
+    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_extDict_selectMLS : ZSTD_HcFindBestMatch_extDict_selectMLS;
 
     U32 offset_1 = rep[0], offset_2 = rep[1];
 
@@ -674,8 +949,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         }   }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+        {   size_t offsetFound = 999999999;
+            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
             if (ml2 > matchLength)
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
@@ -707,8 +982,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             }   }
 
             /* search match, depth 1 */
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+            {   size_t offset2=999999999;
+                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -737,8 +1012,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                 }   }
 
                 /* search match, depth 2 */
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                {   size_t offset2=999999999;
+                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -794,31 +1069,31 @@ _storeSequence:
 
 size_t ZSTD_compressBlock_greedy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, 0, 0);
 }
 
 size_t ZSTD_compressBlock_lazy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, 0, 1);
 }
 
 size_t ZSTD_compressBlock_lazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, 0, 2);
 }
 
 size_t ZSTD_compressBlock_btlazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        void const* src, size_t srcSize)
 
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, 1, 2);
 }
diff --git a/lib/compress/zstd_lazy.h b/lib/compress/zstd_lazy.h
index bda064f1997b..ef85a6df9c8f 100644
--- a/lib/compress/zstd_lazy.h
+++ b/lib/compress/zstd_lazy.h
@@ -17,37 +17,48 @@ extern "C" {
 
 #include "zstd_compress_internal.h"
 
-U32 ZSTD_insertAndFindFirstIndex(
-        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-        const BYTE* ip);
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
 
 void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). pre-emptively increase value of ZSTD_DUBT_UNSORTED_MARK */
 
 size_t ZSTD_compressBlock_btlazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_lazy2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_lazy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_greedy(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
 size_t ZSTD_compressBlock_greedy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_lazy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_lazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btlazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 
 #if defined (__cplusplus)
 }
diff --git a/lib/compress/zstd_ldm.c b/lib/compress/zstd_ldm.c
index bffd8a3dfaa5..6238ddecf24f 100644
--- a/lib/compress/zstd_ldm.c
+++ b/lib/compress/zstd_ldm.c
@@ -9,6 +9,7 @@
 
 #include "zstd_ldm.h"
 
+#include "debug.h"
 #include "zstd_fast.h"          /* ZSTD_fillHashTable() */
 #include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
 
@@ -20,7 +21,7 @@
 void ZSTD_ldm_adjustParameters(ldmParams_t* params,
                                ZSTD_compressionParameters const* cParams)
 {
-    U32 const windowLog = cParams->windowLog;
+    params->windowLog = cParams->windowLog;
     ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
     DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
     if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
@@ -33,12 +34,13 @@ void ZSTD_ldm_adjustParameters(ldmParams_t* params,
       params->minMatchLength = minMatch;
     }
     if (params->hashLog == 0) {
-        params->hashLog = MAX(ZSTD_HASHLOG_MIN, windowLog - LDM_HASH_RLOG);
+        params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
         assert(params->hashLog <= ZSTD_HASHLOG_MAX);
     }
     if (params->hashEveryLog == 0) {
-        params->hashEveryLog =
-                windowLog < params->hashLog ? 0 : windowLog - params->hashLog;
+        params->hashEveryLog = params->windowLog < params->hashLog
+                                   ? 0
+                                   : params->windowLog - params->hashLog;
     }
     params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
 }
@@ -216,21 +218,18 @@ static size_t ZSTD_ldm_countBackwardsMatch(
  *  The tables for the other strategies are filled within their
  *  block compressors. */
 static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
-                                      ZSTD_compressionParameters const* cParams,
                                       void const* end)
 {
     const BYTE* const iend = (const BYTE*)end;
 
-    switch(cParams->strategy)
+    switch(ms->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, cParams, iend);
-        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
         break;
 
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, cParams, iend);
-        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
         break;
 
     case ZSTD_greedy:
@@ -508,7 +507,7 @@ size_t ZSTD_ldm_generateSequences(
          *       * Try invalidation after the sequence generation and test the
          *         the offset against maxDist directly.
          */
-        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL);
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL, NULL);
         /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
         newLeftoverSize = ZSTD_ldm_generateSequences_internal(
             ldmState, sequences, params, chunkStart, chunkSize);
@@ -591,19 +590,19 @@ static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
 
 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-    ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
-    int const extDict)
+    void const* src, size_t srcSize)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     unsigned const minMatch = cParams->searchLength;
     ZSTD_blockCompressor const blockCompressor =
-        ZSTD_selectBlockCompressor(cParams->strategy, extDict);
-    BYTE const* const base = ms->window.base;
+        ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
     /* Input bounds */
     BYTE const* const istart = (BYTE const*)src;
     BYTE const* const iend = istart + srcSize;
     /* Input positions */
     BYTE const* ip = istart;
 
+    DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
     assert(rawSeqStore->pos <= rawSeqStore->size);
     assert(rawSeqStore->size <= rawSeqStore->capacity);
     /* Loop through each sequence and apply the block compressor to the lits */
@@ -621,14 +620,13 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
 
         /* Fill tables for block compressor */
         ZSTD_ldm_limitTableUpdate(ms, ip);
-        ZSTD_ldm_fillFastTables(ms, cParams, ip);
+        ZSTD_ldm_fillFastTables(ms, ip);
         /* Run the block compressor */
+        DEBUGLOG(5, "calling block compressor on segment of size %u", sequence.litLength);
         {
             size_t const newLitLength =
-                blockCompressor(ms, seqStore, rep, cParams, ip,
-                                sequence.litLength);
+                blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
             ip += sequence.litLength;
-            ms->nextToUpdate = (U32)(ip - base);
             /* Update the repcodes */
             for (i = ZSTD_REP_NUM - 1; i > 0; i--)
                 rep[i] = rep[i-1];
@@ -642,12 +640,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     }
     /* Fill the tables for the block compressor */
     ZSTD_ldm_limitTableUpdate(ms, ip);
-    ZSTD_ldm_fillFastTables(ms, cParams, ip);
+    ZSTD_ldm_fillFastTables(ms, ip);
     /* Compress the last literals */
-    {
-        size_t const lastLiterals = blockCompressor(ms, seqStore, rep, cParams,
-                                                    ip, iend - ip);
-        ms->nextToUpdate = (U32)(iend - base);
-        return lastLiterals;
-    }
+    return blockCompressor(ms, seqStore, rep, ip, iend - ip);
 }
diff --git a/lib/compress/zstd_ldm.h b/lib/compress/zstd_ldm.h
index 0c3789ff137c..21fba4d591a4 100644
--- a/lib/compress/zstd_ldm.h
+++ b/lib/compress/zstd_ldm.h
@@ -61,9 +61,7 @@ size_t ZSTD_ldm_generateSequences(
  */
 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
             ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-            ZSTD_compressionParameters const* cParams,
-            void const* src, size_t srcSize,
-            int const extDict);
+            void const* src, size_t srcSize);
 
 /**
  * ZSTD_ldm_skipSequences():
diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
index f63f0c585224..8af69a91d46e 100644
--- a/lib/compress/zstd_opt.c
+++ b/lib/compress/zstd_opt.c
@@ -9,10 +9,11 @@
  */
 
 #include "zstd_compress_internal.h"
+#include "hist.h"
 #include "zstd_opt.h"
 
 
-#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats. Also used for matchSum (?) */
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
 #define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */
 #define ZSTD_MAX_PRICE     (1<<30)
 
@@ -20,128 +21,210 @@
 /*-*************************************
 *  Price functions for optimal parser
 ***************************************/
-static void ZSTD_setLog2Prices(optState_t* optPtr)
+
+#if 0    /* approximation at bit level */
+#  define BITCOST_ACCURACY 0
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat)  ((void)opt, ZSTD_bitWeight(stat))
+#elif 0  /* fractional bit accuracy */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))
+#else    /* opt==approx, ultra==accurate */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) (opt ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#endif
+
+MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
+{
+    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
+}
+
+MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
 {
-    optPtr->log2litSum = ZSTD_highbit32(optPtr->litSum+1);
-    optPtr->log2litLengthSum = ZSTD_highbit32(optPtr->litLengthSum+1);
-    optPtr->log2matchLengthSum = ZSTD_highbit32(optPtr->matchLengthSum+1);
-    optPtr->log2offCodeSum = ZSTD_highbit32(optPtr->offCodeSum+1);
+    U32 const stat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(stat);
+    U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + BITCOST_ACCURACY < 31);
+    return weight;
 }
 
+/* debugging function, @return price in bytes */
+MEM_STATIC double ZSTD_fCost(U32 price)
+{
+    return (double)price / (BITCOST_MULTIPLIER*8);
+}
+
+static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
+{
+    optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);
+    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);
+    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);
+    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);
+}
+
+
+static U32 ZSTD_downscaleStat(U32* table, U32 lastEltIndex, int malus)
+{
+    U32 s, sum=0;
+    assert(ZSTD_FREQ_DIV+malus > 0 && ZSTD_FREQ_DIV+malus < 31);
+    for (s=0; s<=lastEltIndex; s++) {
+        table[s] = 1 + (table[s] >> (ZSTD_FREQ_DIV+malus));
+        sum += table[s];
+    }
+    return sum;
+}
 
 static void ZSTD_rescaleFreqs(optState_t* const optPtr,
-                              const BYTE* const src, size_t const srcSize)
+                              const BYTE* const src, size_t const srcSize,
+                              int optLevel)
 {
-    optPtr->staticPrices = 0;
-
-    if (optPtr->litLengthSum == 0) {  /* first init */
-        unsigned u;
-        if (srcSize <= 1024) optPtr->staticPrices = 1;
-
-        assert(optPtr->litFreq!=NULL);
-        for (u=0; u<=MaxLit; u++)
-            optPtr->litFreq[u] = 0;
-        for (u=0; u<srcSize; u++)
-            optPtr->litFreq[src[u]]++;
-        optPtr->litSum = 0;
-        for (u=0; u<=MaxLit; u++) {
-            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> ZSTD_FREQ_DIV);
-            optPtr->litSum += optPtr->litFreq[u];
-        }
+    optPtr->priceType = zop_dynamic;
+
+    if (optPtr->litLengthSum == 0) {  /* first block : init */
+        if (srcSize <= 1024)   /* heuristic */
+            optPtr->priceType = zop_predef;
+
+        assert(optPtr->symbolCosts != NULL);
+        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) { /* huffman table presumed generated by dictionary */
+            optPtr->priceType = zop_dynamic;
+
+            assert(optPtr->litFreq != NULL);
+            optPtr->litSum = 0;
+            {   unsigned lit;
+                for (lit=0; lit<=MaxLit; lit++) {
+                    U32 const scaleLog = 11;   /* scale to 2K */
+                    U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);
+                    assert(bitCost <= scaleLog);
+                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litSum += optPtr->litFreq[lit];
+            }   }
+
+            {   unsigned ll;
+                FSE_CState_t llstate;
+                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
+                optPtr->litLengthSum = 0;
+                for (ll=0; ll<=MaxLL; ll++) {
+                    U32 const scaleLog = 10;   /* scale to 1K */
+                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);
+                    assert(bitCost < scaleLog);
+                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];
+            }   }
+
+            {   unsigned ml;
+                FSE_CState_t mlstate;
+                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
+                optPtr->matchLengthSum = 0;
+                for (ml=0; ml<=MaxML; ml++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);
+                    assert(bitCost < scaleLog);
+                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];
+            }   }
+
+            {   unsigned of;
+                FSE_CState_t ofstate;
+                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
+                optPtr->offCodeSum = 0;
+                for (of=0; of<=MaxOff; of++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);
+                    assert(bitCost < scaleLog);
+                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->offCodeSum += optPtr->offCodeFreq[of];
+            }   }
+
+        } else {  /* not a dictionary */
+
+            assert(optPtr->litFreq != NULL);
+            {   unsigned lit = MaxLit;
+                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
+            }
+            optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+
+            {   unsigned ll;
+                for (ll=0; ll<=MaxLL; ll++)
+                    optPtr->litLengthFreq[ll] = 1;
+            }
+            optPtr->litLengthSum = MaxLL+1;
+
+            {   unsigned ml;
+                for (ml=0; ml<=MaxML; ml++)
+                    optPtr->matchLengthFreq[ml] = 1;
+            }
+            optPtr->matchLengthSum = MaxML+1;
+
+            {   unsigned of;
+                for (of=0; of<=MaxOff; of++)
+                    optPtr->offCodeFreq[of] = 1;
+            }
+            optPtr->offCodeSum = MaxOff+1;
 
-        for (u=0; u<=MaxLL; u++)
-            optPtr->litLengthFreq[u] = 1;
-        optPtr->litLengthSum = MaxLL+1;
-        for (u=0; u<=MaxML; u++)
-            optPtr->matchLengthFreq[u] = 1;
-        optPtr->matchLengthSum = MaxML+1;
-        for (u=0; u<=MaxOff; u++)
-            optPtr->offCodeFreq[u] = 1;
-        optPtr->offCodeSum = (MaxOff+1);
-
-    } else {
-        unsigned u;
-
-        optPtr->litSum = 0;
-        for (u=0; u<=MaxLit; u++) {
-            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> (ZSTD_FREQ_DIV+1));
-            optPtr->litSum += optPtr->litFreq[u];
-        }
-        optPtr->litLengthSum = 0;
-        for (u=0; u<=MaxLL; u++) {
-            optPtr->litLengthFreq[u] = 1 + (optPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
-            optPtr->litLengthSum += optPtr->litLengthFreq[u];
-        }
-        optPtr->matchLengthSum = 0;
-        for (u=0; u<=MaxML; u++) {
-            optPtr->matchLengthFreq[u] = 1 + (optPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
-            optPtr->matchLengthSum += optPtr->matchLengthFreq[u];
-        }
-        optPtr->offCodeSum = 0;
-        for (u=0; u<=MaxOff; u++) {
-            optPtr->offCodeFreq[u] = 1 + (optPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
-            optPtr->offCodeSum += optPtr->offCodeFreq[u];
         }
+
+    } else {   /* new block : re-use previous statistics, scaled down */
+
+        optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+        optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);
+        optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);
+        optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);
     }
 
-    ZSTD_setLog2Prices(optPtr);
+    ZSTD_setBasePrices(optPtr, optLevel);
 }
 
-
 /* ZSTD_rawLiteralsCost() :
- * cost of literals (only) in given segment (which length can be null)
- * does not include cost of literalLength symbol */
+ * price of literals (only) in specified segment (which length can be 0).
+ * does not include price of literalLength symbol */
 static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
-                                const optState_t* const optPtr)
+                                const optState_t* const optPtr,
+                                int optLevel)
 {
-    if (optPtr->staticPrices) return (litLength*6);  /* 6 bit per literal - no statistic used */
     if (litLength == 0) return 0;
-
-    /* literals */
-    {   U32 u;
-        U32 cost = litLength * optPtr->log2litSum;
-        for (u=0; u < litLength; u++)
-            cost -= ZSTD_highbit32(optPtr->litFreq[literals[u]]+1);
-        return cost;
+    if (optPtr->priceType == zop_predef)
+        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
+
+    /* dynamic statistics */
+    {   U32 price = litLength * optPtr->litSumBasePrice;
+        U32 u;
+        for (u=0; u < litLength; u++) {
+            assert(WEIGHT(optPtr->litFreq[literals[u]], optLevel) <= optPtr->litSumBasePrice);   /* literal cost should never be negative */
+            price -= WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+        }
+        return price;
     }
 }
 
 /* ZSTD_litLengthPrice() :
  * cost of literalLength symbol */
-static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr)
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
 {
-    if (optPtr->staticPrices) return ZSTD_highbit32((U32)litLength+1);
+    if (optPtr->priceType == zop_predef) return WEIGHT(litLength, optLevel);
 
-    /* literal Length */
+    /* dynamic statistics */
     {   U32 const llCode = ZSTD_LLcode(litLength);
-        U32 const price = LL_bits[llCode] + optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
-        return price;
+        return (LL_bits[llCode] * BITCOST_MULTIPLIER) + (optPtr->litLengthSumBasePrice - WEIGHT(optPtr->litLengthFreq[llCode], optLevel));
     }
 }
 
-/* ZSTD_litLengthPrice() :
- * cost of the literal part of a sequence,
- * including literals themselves, and literalLength symbol */
-static U32 ZSTD_fullLiteralsCost(const BYTE* const literals, U32 const litLength,
-                                 const optState_t* const optPtr)
-{
-    return ZSTD_rawLiteralsCost(literals, litLength, optPtr)
-         + ZSTD_litLengthPrice(litLength, optPtr);
-}
-
 /* ZSTD_litLengthContribution() :
  * @return ( cost(litlength) - cost(0) )
  * this value can then be added to rawLiteralsCost()
  * to provide a cost which is directly comparable to a match ending at same position */
-static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr)
+static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr, int optLevel)
 {
-    if (optPtr->staticPrices) return ZSTD_highbit32(litLength+1);
+    if (optPtr->priceType >= zop_predef) return WEIGHT(litLength, optLevel);
 
-    /* literal Length */
+    /* dynamic statistics */
     {   U32 const llCode = ZSTD_LLcode(litLength);
-        int const contribution = LL_bits[llCode]
-                        + ZSTD_highbit32(optPtr->litLengthFreq[0]+1)
-                        - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+        int const contribution = (LL_bits[llCode] * BITCOST_MULTIPLIER)
+                               + WEIGHT(optPtr->litLengthFreq[0], optLevel)   /* note: log2litLengthSum cancel out */
+                               - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
 #if 1
         return contribution;
 #else
@@ -155,10 +238,11 @@ static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* con
  * which can be compared to the ending cost of a match
  * should a new match start at this position */
 static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLength,
-                                     const optState_t* const optPtr)
+                                     const optState_t* const optPtr,
+                                     int optLevel)
 {
-    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr)
-                           + ZSTD_litLengthContribution(litLength, optPtr);
+    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr, optLevel)
+                           + ZSTD_litLengthContribution(litLength, optPtr, optLevel);
     return contribution;
 }
 
@@ -166,31 +250,38 @@ static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLe
  * Provides the cost of the match part (offset + matchLength) of a sequence
  * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
  * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */
-FORCE_INLINE_TEMPLATE U32 ZSTD_getMatchPrice(
-                                    U32 const offset, U32 const matchLength,
-                                    const optState_t* const optPtr,
-                                    int const optLevel)
+FORCE_INLINE_TEMPLATE U32
+ZSTD_getMatchPrice(U32 const offset,
+                   U32 const matchLength,
+                   const optState_t* const optPtr,
+                   int const optLevel)
 {
     U32 price;
     U32 const offCode = ZSTD_highbit32(offset+1);
     U32 const mlBase = matchLength - MINMATCH;
     assert(matchLength >= MINMATCH);
 
-    if (optPtr->staticPrices)  /* fixed scheme, do not use statistics */
-        return ZSTD_highbit32((U32)mlBase+1) + 16 + offCode;
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, do not use statistics */
+        return WEIGHT(mlBase, optLevel) + ((16 + offCode) * BITCOST_MULTIPLIER);
 
-    price = offCode + optPtr->log2offCodeSum - ZSTD_highbit32(optPtr->offCodeFreq[offCode]+1);
-    if ((optLevel<2) /*static*/ && offCode >= 20) price += (offCode-19)*2; /* handicap for long distance offsets, favor decompression speed */
+    /* dynamic statistics */
+    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
+    if ((optLevel<2) /*static*/ && offCode >= 20)
+        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */
 
     /* match Length */
     {   U32 const mlCode = ZSTD_MLcode(mlBase);
-        price += ML_bits[mlCode] + optPtr->log2matchLengthSum - ZSTD_highbit32(optPtr->matchLengthFreq[mlCode]+1);
+        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));
     }
 
+    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */
+
     DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
     return price;
 }
 
+/* ZSTD_updateStats() :
+ * assumption : literals + litLengtn <= iend */
 static void ZSTD_updateStats(optState_t* const optPtr,
                              U32 litLength, const BYTE* literals,
                              U32 offsetCode, U32 matchLength)
@@ -269,10 +360,11 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE*
  *  ip : assumed <= iend-8 .
  * @return : nb of positions added */
 static U32 ZSTD_insertBt1(
-                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iend,
-                U32 const mls, U32 const extDict)
+                U32 const mls, const int extDict)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32*   const hashTable = ms->hashTable;
     U32    const hashLog = cParams->hashLog;
     size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
@@ -293,6 +385,7 @@ static U32 ZSTD_insertBt1(
     U32* largerPtr  = smallerPtr + 1;
     U32 dummy32;   /* to be nullified at the end */
     U32 const windowLow = ms->window.lowLimit;
+    U32 const matchLow = windowLow ? windowLow : 1;
     U32 matchEndIdx = current+8+1;
     size_t bestLength = 8;
     U32 nbCompares = 1U << cParams->searchLog;
@@ -308,7 +401,7 @@ static U32 ZSTD_insertBt1(
     assert(ip <= iend-8);   /* required for h calculation */
     hashTable[h] = current;   /* Update Hash Table */
 
-    while (nbCompares-- && (matchIndex > windowLow)) {
+    while (nbCompares-- && (matchIndex >= matchLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         assert(matchIndex < current);
@@ -334,8 +427,8 @@ static U32 ZSTD_insertBt1(
         }
 #endif
 
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if extDict is incorrectly set to 0 */
+        if (!extDict || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */
             match = base + matchIndex;
             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
         } else {
@@ -379,35 +472,33 @@ static U32 ZSTD_insertBt1(
 
 FORCE_INLINE_TEMPLATE
 void ZSTD_updateTree_internal(
-                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iend,
-                const U32 mls, const U32 extDict)
+                const U32 mls, const ZSTD_dictMode_e dictMode)
 {
     const BYTE* const base = ms->window.base;
     U32 const target = (U32)(ip - base);
     U32 idx = ms->nextToUpdate;
-    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (extDict:%u)",
-                idx, target, extDict);
+    DEBUGLOG(5, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
+                idx, target, dictMode);
 
     while(idx < target)
-        idx += ZSTD_insertBt1(ms, cParams, base+idx, iend, mls, extDict);
+        idx += ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
     ms->nextToUpdate = target;
 }
 
-void ZSTD_updateTree(
-                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                const BYTE* ip, const BYTE* iend)
-{
-    ZSTD_updateTree_internal(ms, cParams, ip, iend, cParams->searchLength, 0 /*extDict*/);
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
+    ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.searchLength, ZSTD_noDict);
 }
 
 FORCE_INLINE_TEMPLATE
 U32 ZSTD_insertBtAndGetAllMatches (
-                    ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                    const BYTE* const ip, const BYTE* const iLimit, int const extDict,
+                    ZSTD_matchState_t* ms,
+                    const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
                     U32 rep[ZSTD_REP_NUM], U32 const ll0,
                     ZSTD_match_t* matches, const U32 lengthToBeat, U32 const mls /* template */)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     const BYTE* const base = ms->window.base;
     U32 const current = (U32)(ip-base);
@@ -426,6 +517,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     const BYTE* const prefixStart = base + dictLimit;
     U32 const btLow = btMask >= current ? 0 : current - btMask;
     U32 const windowLow = ms->window.lowLimit;
+    U32 const matchLow = windowLow ? windowLow : 1;
     U32* smallerPtr = bt + 2*(current&btMask);
     U32* largerPtr  = bt + 2*(current&btMask) + 1;
     U32 matchEndIdx = current+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
@@ -433,8 +525,21 @@ U32 ZSTD_insertBtAndGetAllMatches (
     U32 mnum = 0;
     U32 nbCompares = 1U << cParams->searchLog;
 
+    const ZSTD_matchState_t* dms    = dictMode == ZSTD_dictMatchState ? ms->dictMatchState : NULL;
+    const ZSTD_compressionParameters* const dmsCParams =
+                                      dictMode == ZSTD_dictMatchState ? &dms->cParams : NULL;
+    const BYTE* const dmsBase       = dictMode == ZSTD_dictMatchState ? dms->window.base : NULL;
+    const BYTE* const dmsEnd        = dictMode == ZSTD_dictMatchState ? dms->window.nextSrc : NULL;
+    U32         const dmsHighLimit  = dictMode == ZSTD_dictMatchState ? (U32)(dmsEnd - dmsBase) : 0;
+    U32         const dmsLowLimit   = dictMode == ZSTD_dictMatchState ? dms->window.lowLimit : 0;
+    U32         const dmsIndexDelta = dictMode == ZSTD_dictMatchState ? windowLow - dmsHighLimit : 0;
+    U32         const dmsHashLog    = dictMode == ZSTD_dictMatchState ? dmsCParams->hashLog : hashLog;
+    U32         const dmsBtLog      = dictMode == ZSTD_dictMatchState ? dmsCParams->chainLog - 1 : btLog;
+    U32         const dmsBtMask     = dictMode == ZSTD_dictMatchState ? (1U << dmsBtLog) - 1 : 0;
+    U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
+
     size_t bestLength = lengthToBeat-1;
-    DEBUGLOG(7, "ZSTD_insertBtAndGetAllMatches");
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", current);
 
     /* check repCode */
     {   U32 const lastR = ZSTD_REP_NUM + ll0;
@@ -449,18 +554,26 @@ U32 ZSTD_insertBtAndGetAllMatches (
                     repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
                 }
             } else {  /* repIndex < dictLimit || repIndex >= current */
-                const BYTE* const repMatch = dictBase + repIndex;
+                const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
+                                             dmsBase + repIndex - dmsIndexDelta :
+                                             dictBase + repIndex;
                 assert(current >= windowLow);
-                if ( extDict /* this case only valid in extDict mode */
+                if ( dictMode == ZSTD_extDict
                   && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
                      & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
                   && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
                     repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+                }
+                if (dictMode == ZSTD_dictMatchState
+                  && ( ((repOffset-1) /*intentional overflow*/ < current - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `current > repIndex >= dmsLowLimit` */
+                     & ((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
             }   }
             /* save longer solution */
             if (repLen > bestLength) {
-                DEBUGLOG(8, "found rep-match %u of length %u",
-                            repCode - ll0, (U32)repLen);
+                DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
+                            repCode, ll0, repOffset, repLen);
                 bestLength = repLen;
                 matches[mnum].off = repCode - ll0;
                 matches[mnum].len = (U32)repLen;
@@ -473,10 +586,10 @@ U32 ZSTD_insertBtAndGetAllMatches (
     /* HC3 match finder */
     if ((mls == 3) /*static*/ && (bestLength < mls)) {
         U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, ip);
-        if ((matchIndex3 > windowLow)
+        if ((matchIndex3 >= matchLow)
           & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
             size_t mlen;
-            if ((!extDict) /*static*/ || (matchIndex3 >= dictLimit)) {
+            if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
                 const BYTE* const match = base + matchIndex3;
                 mlen = ZSTD_count(ip, match, iLimit);
             } else {
@@ -498,17 +611,21 @@ U32 ZSTD_insertBtAndGetAllMatches (
                      (ip+mlen == iLimit) ) {  /* best possible length */
                     ms->nextToUpdate = current+1;  /* skip insertion */
                     return 1;
-    }   }   }   }
+                }
+            }
+        }
+        /* no dictMatchState lookup: dicts don't have a populated HC3 table */
+    }
 
     hashTable[h] = current;   /* Update Hash Table */
 
-    while (nbCompares-- && (matchIndex > windowLow)) {
+    while (nbCompares-- && (matchIndex >= matchLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         const BYTE* match;
         assert(current > matchIndex);
 
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+        if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
             assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
             match = base + matchIndex;
             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
@@ -520,8 +637,8 @@ U32 ZSTD_insertBtAndGetAllMatches (
         }
 
         if (matchLength > bestLength) {
-            DEBUGLOG(8, "found match of length %u at distance %u",
-                    (U32)matchLength, current - matchIndex);
+            DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
+                    (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
             assert(matchEndIdx > matchIndex);
             if (matchLength > matchEndIdx - matchIndex)
                 matchEndIdx = matchIndex + (U32)matchLength;
@@ -529,9 +646,10 @@ U32 ZSTD_insertBtAndGetAllMatches (
             matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
             matches[mnum].len = (U32)matchLength;
             mnum++;
-            if (matchLength > ZSTD_OPT_NUM) break;
-            if (ip+matchLength == iLimit) {  /* equal : no way to know if inf or sup */
-                break;   /* drop, to preserve bt consistency (miss a little bit of compression) */
+            if ( (matchLength > ZSTD_OPT_NUM)
+               | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
+                break; /* drop, to preserve bt consistency (miss a little bit of compression) */
             }
         }
 
@@ -552,6 +670,47 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
     *smallerPtr = *largerPtr = 0;
 
+    if (dictMode == ZSTD_dictMatchState && nbCompares) {
+        size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
+        U32 dictMatchIndex = dms->hashTable[dmsH];
+        const U32* const dmsBt = dms->chainTable;
+        commonLengthSmaller = commonLengthLarger = 0;
+        while (nbCompares-- && (dictMatchIndex > dmsLowLimit)) {
+            const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match = dmsBase + dictMatchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dmsEnd, prefixStart);
+            if (dictMatchIndex+matchLength >= dmsHighLimit)
+                match = base + dictMatchIndex + dmsIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+            if (matchLength > bestLength) {
+                matchIndex = dictMatchIndex + dmsIndexDelta;
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
+                        (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                bestLength = matchLength;
+                matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+                matches[mnum].len = (U32)matchLength;
+                mnum++;
+                if ( (matchLength > ZSTD_OPT_NUM)
+                   | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
+            if (match[matchLength] < ip[matchLength]) {
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                commonLengthLarger = matchLength;
+                dictMatchIndex = nextPtr[0];
+            }
+        }
+    }
+
     assert(matchEndIdx > current+8);
     ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
     return mnum;
@@ -559,23 +718,24 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
 
 FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
-                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-                        const BYTE* ip, const BYTE* const iHighLimit, int const extDict,
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iHighLimit, const ZSTD_dictMode_e dictMode,
                         U32 rep[ZSTD_REP_NUM], U32 const ll0,
                         ZSTD_match_t* matches, U32 const lengthToBeat)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const matchLengthSearch = cParams->searchLength;
-    DEBUGLOG(7, "ZSTD_BtGetAllMatches");
+    DEBUGLOG(8, "ZSTD_BtGetAllMatches");
     if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_internal(ms, cParams, ip, iHighLimit, matchLengthSearch, extDict);
+    ZSTD_updateTree_internal(ms, ip, iHighLimit, matchLengthSearch, dictMode);
     switch(matchLengthSearch)
     {
-    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 3);
+    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 3);
     default :
-    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 4);
-    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 5);
+    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 5);
     case 7 :
-    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 6);
+    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 6);
     }
 }
 
@@ -587,7 +747,7 @@ typedef struct repcodes_s {
     U32 rep[3];
 } repcodes_t;
 
-repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
+static repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
 {
     repcodes_t newReps;
     if (offset >= ZSTD_REP_NUM) {  /* full offset */
@@ -609,65 +769,17 @@ repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
 }
 
 
-typedef struct {
-    const BYTE* anchor;
-    U32 litlen;
-    U32 rawLitCost;
-} cachedLiteralPrice_t;
-
-static U32 ZSTD_rawLiteralsCost_cached(
-                            cachedLiteralPrice_t* const cachedLitPrice,
-                            const BYTE* const anchor, U32 const litlen,
-                            const optState_t* const optStatePtr)
-{
-    U32 startCost;
-    U32 remainingLength;
-    const BYTE* startPosition;
-
-    if (anchor == cachedLitPrice->anchor) {
-        startCost = cachedLitPrice->rawLitCost;
-        startPosition = anchor + cachedLitPrice->litlen;
-        assert(litlen >= cachedLitPrice->litlen);
-        remainingLength = litlen - cachedLitPrice->litlen;
-    } else {
-        startCost = 0;
-        startPosition = anchor;
-        remainingLength = litlen;
-    }
-
-    {   U32 const rawLitCost = startCost + ZSTD_rawLiteralsCost(startPosition, remainingLength, optStatePtr);
-        cachedLitPrice->anchor = anchor;
-        cachedLitPrice->litlen = litlen;
-        cachedLitPrice->rawLitCost = rawLitCost;
-        return rawLitCost;
-    }
-}
-
-static U32 ZSTD_fullLiteralsCost_cached(
-                            cachedLiteralPrice_t* const cachedLitPrice,
-                            const BYTE* const anchor, U32 const litlen,
-                            const optState_t* const optStatePtr)
-{
-    return ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
-         + ZSTD_litLengthPrice(litlen, optStatePtr);
-}
-
-static int ZSTD_literalsContribution_cached(
-                            cachedLiteralPrice_t* const cachedLitPrice,
-                            const BYTE* const anchor, U32 const litlen,
-                            const optState_t* const optStatePtr)
+static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
 {
-    int const contribution = ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
-                           + ZSTD_litLengthContribution(litlen, optStatePtr);
-    return contribution;
+    return sol.litlen + sol.mlen;
 }
 
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore,
-                                      U32 rep[ZSTD_REP_NUM],
-                                      ZSTD_compressionParameters const* cParams,
-                                      const void* src, size_t srcSize,
-                                      const int optLevel, const int extDict)
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
+                               seqStore_t* seqStore,
+                               U32 rep[ZSTD_REP_NUM],
+                               const void* src, size_t srcSize,
+                               const int optLevel, const ZSTD_dictMode_e dictMode)
 {
     optState_t* const optStatePtr = &ms->opt;
     const BYTE* const istart = (const BYTE*)src;
@@ -677,72 +789,76 @@ size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore
     const BYTE* const ilimit = iend - 8;
     const BYTE* const base = ms->window.base;
     const BYTE* const prefixStart = base + ms->window.dictLimit;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
 
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     U32 const minMatch = (cParams->searchLength == 3) ? 3 : 4;
 
     ZSTD_optimal_t* const opt = optStatePtr->priceTable;
     ZSTD_match_t* const matches = optStatePtr->matchTable;
-    cachedLiteralPrice_t cachedLitPrice;
+    ZSTD_optimal_t lastSequence;
 
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_opt_generic");
+    assert(optLevel <= 2);
     ms->nextToUpdate3 = ms->nextToUpdate;
-    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
     ip += (ip==prefixStart);
-    memset(&cachedLitPrice, 0, sizeof(cachedLitPrice));
 
     /* Match Loop */
     while (ip < ilimit) {
         U32 cur, last_pos = 0;
-        U32 best_mlen, best_off;
 
         /* find first match */
         {   U32 const litlen = (U32)(ip - anchor);
             U32 const ll0 = !litlen;
-            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, ip, iend, extDict, rep, ll0, matches, minMatch);
+            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, ip, iend, dictMode, rep, ll0, matches, minMatch);
             if (!nbMatches) { ip++; continue; }
 
             /* initialize opt[0] */
             { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-            opt[0].mlen = 1;
+            opt[0].mlen = 0;  /* means is_a_literal */
             opt[0].litlen = litlen;
+            opt[0].price = ZSTD_literalsContribution(anchor, litlen, optStatePtr, optLevel);
 
             /* large match -> immediate encoding */
             {   U32 const maxML = matches[nbMatches-1].len;
-                DEBUGLOG(7, "found %u matches of maxLength=%u and offset=%u at cPos=%u => start new serie",
-                            nbMatches, maxML, matches[nbMatches-1].off, (U32)(ip-prefixStart));
+                U32 const maxOffset = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffCode=%u at cPos=%u => start new serie",
+                            nbMatches, maxML, maxOffset, (U32)(ip-prefixStart));
 
                 if (maxML > sufficient_len) {
-                    best_mlen = maxML;
-                    best_off = matches[nbMatches-1].off;
-                    DEBUGLOG(7, "large match (%u>%u), immediate encoding",
-                                best_mlen, sufficient_len);
+                    lastSequence.litlen = litlen;
+                    lastSequence.mlen = maxML;
+                    lastSequence.off = maxOffset;
+                    DEBUGLOG(6, "large match (%u>%u), immediate encoding",
+                                maxML, sufficient_len);
                     cur = 0;
-                    last_pos = 1;
+                    last_pos = ZSTD_totalLen(lastSequence);
                     goto _shortestPath;
             }   }
 
             /* set prices for first matches starting position == 0 */
-            {   U32 const literalsPrice = ZSTD_fullLiteralsCost_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
+            {   U32 const literalsPrice = opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
                 U32 pos;
                 U32 matchNb;
-                for (pos = 0; pos < minMatch; pos++) {
-                    opt[pos].mlen = 1;
-                    opt[pos].price = ZSTD_MAX_PRICE;
+                for (pos = 1; pos < minMatch; pos++) {
+                    opt[pos].price = ZSTD_MAX_PRICE;   /* mlen, litlen and price will be fixed during forward scanning */
                 }
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
                     U32 const offset = matches[matchNb].off;
                     U32 const end = matches[matchNb].len;
                     repcodes_t const repHistory = ZSTD_updateRep(rep, offset, ll0);
                     for ( ; pos <= end ; pos++ ) {
-                        U32 const matchPrice = literalsPrice + ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
-                        DEBUGLOG(7, "rPos:%u => set initial price : %u",
-                                    pos, matchPrice);
+                        U32 const matchPrice = ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
+                        U32 const sequencePrice = literalsPrice + matchPrice;
+                        DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
+                                    pos, ZSTD_fCost(sequencePrice));
                         opt[pos].mlen = pos;
                         opt[pos].off = offset;
                         opt[pos].litlen = litlen;
-                        opt[pos].price = matchPrice;
+                        opt[pos].price = sequencePrice;
+                        ZSTD_STATIC_ASSERT(sizeof(opt[pos].rep) == sizeof(repHistory));
                         memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
                 }   }
                 last_pos = pos-1;
@@ -753,55 +869,67 @@ size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore
         for (cur = 1; cur <= last_pos; cur++) {
             const BYTE* const inr = ip + cur;
             assert(cur < ZSTD_OPT_NUM);
+            DEBUGLOG(7, "cPos:%zi==rPos:%u", inr-istart, cur)
 
             /* Fix current position with one literal if cheaper */
-            {   U32 const litlen = (opt[cur-1].mlen == 1) ? opt[cur-1].litlen + 1 : 1;
-                int price;  /* note : contribution can be negative */
-                if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_literalsContribution(inr-litlen, litlen, optStatePtr);
-                } else {
-                    price = ZSTD_literalsContribution_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
-                }
+            {   U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
+                int const price = opt[cur-1].price
+                                + ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+                                + ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
+                                - ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
                 assert(price < 1000000000); /* overflow check */
                 if (price <= opt[cur].price) {
-                    DEBUGLOG(7, "rPos:%u : better price (%u<%u) using literal",
-                                cur, price, opt[cur].price);
-                    opt[cur].mlen = 1;
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
+                                opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
+                    opt[cur].mlen = 0;
                     opt[cur].off = 0;
                     opt[cur].litlen = litlen;
                     opt[cur].price = price;
                     memcpy(opt[cur].rep, opt[cur-1].rep, sizeof(opt[cur].rep));
-            }   }
+                } else {
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price),
+                                opt[cur].rep[0], opt[cur].rep[1], opt[cur].rep[2]);
+                }
+            }
 
             /* last match must start at a minimum distance of 8 from oend */
             if (inr > ilimit) continue;
 
             if (cur == last_pos) break;
 
-             if ( (optLevel==0) /*static*/
-               && (opt[cur+1].price <= opt[cur].price) )
+            if ( (optLevel==0) /*static_test*/
+              && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
+                DEBUGLOG(7, "move to next rPos:%u : price is <=", cur+1);
                 continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+            }
 
-            {   U32 const ll0 = (opt[cur].mlen != 1);
-                U32 const litlen = (opt[cur].mlen == 1) ? opt[cur].litlen : 0;
-                U32 const previousPrice = (cur > litlen) ? opt[cur-litlen].price : 0;
-                U32 const basePrice = previousPrice + ZSTD_fullLiteralsCost(inr-litlen, litlen, optStatePtr);
-                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, inr, iend, extDict, opt[cur].rep, ll0, matches, minMatch);
+            {   U32 const ll0 = (opt[cur].mlen != 0);
+                U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
+                U32 const previousPrice = opt[cur].price;
+                U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, inr, iend, dictMode, opt[cur].rep, ll0, matches, minMatch);
                 U32 matchNb;
-                if (!nbMatches) continue;
+                if (!nbMatches) {
+                    DEBUGLOG(7, "rPos:%u : no match found", cur);
+                    continue;
+                }
 
                 {   U32 const maxML = matches[nbMatches-1].len;
-                    DEBUGLOG(7, "rPos:%u, found %u matches, of maxLength=%u",
-                                cur, nbMatches, maxML);
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u, found %u matches, of maxLength=%u",
+                                inr-istart, cur, nbMatches, maxML);
 
                     if ( (maxML > sufficient_len)
-                       | (cur + maxML >= ZSTD_OPT_NUM) ) {
-                        best_mlen = maxML;
-                        best_off = matches[nbMatches-1].off;
-                        last_pos = cur + 1;
+                      || (cur + maxML >= ZSTD_OPT_NUM) ) {
+                        lastSequence.mlen = maxML;
+                        lastSequence.off = matches[nbMatches-1].off;
+                        lastSequence.litlen = litlen;
+                        cur -= (opt[cur].mlen==0) ? opt[cur].litlen : 0;  /* last sequence is actually only literals, fix cur to last match - note : may underflow, in which case, it's first sequence, and it's okay */
+                        last_pos = cur + ZSTD_totalLen(lastSequence);
+                        if (cur > ZSTD_OPT_NUM) cur = 0;   /* underflow => first match */
                         goto _shortestPath;
-                    }
-                }
+                }   }
 
                 /* set prices using matches found at position == cur */
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
@@ -811,81 +939,97 @@ size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore
                     U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
                     U32 mlen;
 
-                    DEBUGLOG(7, "testing match %u => offCode=%u, mlen=%u, llen=%u",
+                    DEBUGLOG(7, "testing match %u => offCode=%4u, mlen=%2u, llen=%2u",
                                 matchNb, matches[matchNb].off, lastML, litlen);
 
-                    for (mlen = lastML; mlen >= startML; mlen--) {
+                    for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
                         U32 const pos = cur + mlen;
                         int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
 
                         if ((pos > last_pos) || (price < opt[pos].price)) {
-                            DEBUGLOG(7, "rPos:%u => new better price (%u<%u)",
-                                        pos, price, opt[pos].price);
-                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }   /* fill empty positions */
                             opt[pos].mlen = mlen;
                             opt[pos].off = offset;
                             opt[pos].litlen = litlen;
                             opt[pos].price = price;
+                            ZSTD_STATIC_ASSERT(sizeof(opt[pos].rep) == sizeof(repHistory));
                             memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
                         } else {
-                            if (optLevel==0) break;  /* gets ~+10% speed for about -0.01 ratio loss */
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
                         }
             }   }   }
         }  /* for (cur = 1; cur <= last_pos; cur++) */
 
-        best_mlen = opt[last_pos].mlen;
-        best_off = opt[last_pos].off;
-        cur = last_pos - best_mlen;
+        lastSequence = opt[last_pos];
+        cur = last_pos > ZSTD_totalLen(lastSequence) ? last_pos - ZSTD_totalLen(lastSequence) : 0;  /* single sequence, and it starts before `ip` */
+        assert(cur < ZSTD_OPT_NUM);  /* control overflow*/
 
 _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
-        assert(opt[0].mlen == 1);
-
-        /* reverse traversal */
-        DEBUGLOG(7, "start reverse traversal (last_pos:%u, cur:%u)",
-                    last_pos, cur);
-        {   U32 selectedMatchLength = best_mlen;
-            U32 selectedOffset = best_off;
-            U32 pos = cur;
-            while (1) {
-                U32 const mlen = opt[pos].mlen;
-                U32 const off = opt[pos].off;
-                opt[pos].mlen = selectedMatchLength;
-                opt[pos].off = selectedOffset;
-                selectedMatchLength = mlen;
-                selectedOffset = off;
-                if (mlen > pos) break;
-                pos -= mlen;
-        }   }
-
-        /* save sequences */
-        {   U32 pos;
-            for (pos=0; pos < last_pos; ) {
-                U32 const llen = (U32)(ip - anchor);
-                U32 const mlen = opt[pos].mlen;
-                U32 const offset = opt[pos].off;
-                if (mlen == 1) { ip++; pos++; continue; }  /* literal position => move on */
-                pos += mlen; ip += mlen;
-
-                /* repcodes update : like ZSTD_updateRep(), but update in place */
-                if (offset >= ZSTD_REP_NUM) {  /* full offset */
-                    rep[2] = rep[1];
-                    rep[1] = rep[0];
-                    rep[0] = offset - ZSTD_REP_MOVE;
-                } else {   /* repcode */
-                    U32 const repCode = offset + (llen==0);
-                    if (repCode) {  /* note : if repCode==0, no change */
-                        U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
-                        if (repCode >= 2) rep[2] = rep[1];
-                        rep[1] = rep[0];
-                        rep[0] = currentOffset;
+        assert(opt[0].mlen == 0);
+
+        {   U32 const storeEnd = cur + 1;
+            U32 storeStart = storeEnd;
+            U32 seqPos = cur;
+
+            DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
+                        last_pos, cur); (void)last_pos;
+            assert(storeEnd < ZSTD_OPT_NUM);
+            DEBUGLOG(6, "last sequence copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                        storeEnd, lastSequence.litlen, lastSequence.mlen, lastSequence.off);
+            opt[storeEnd] = lastSequence;
+            while (seqPos > 0) {
+                U32 const backDist = ZSTD_totalLen(opt[seqPos]);
+                storeStart--;
+                DEBUGLOG(6, "sequence from rPos=%u copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                            seqPos, storeStart, opt[seqPos].litlen, opt[seqPos].mlen, opt[seqPos].off);
+                opt[storeStart] = opt[seqPos];
+                seqPos = (seqPos > backDist) ? seqPos - backDist : 0;
+            }
+
+            /* save sequences */
+            DEBUGLOG(6, "sending selected sequences into seqStore")
+            {   U32 storePos;
+                for (storePos=storeStart; storePos <= storeEnd; storePos++) {
+                    U32 const llen = opt[storePos].litlen;
+                    U32 const mlen = opt[storePos].mlen;
+                    U32 const offCode = opt[storePos].off;
+                    U32 const advance = llen + mlen;
+                    DEBUGLOG(6, "considering seq starting at %zi, llen=%u, mlen=%u",
+                                anchor - istart, llen, mlen);
+
+                    if (mlen==0) {  /* only literals => must be last "sequence", actually starting a new stream of sequences */
+                        assert(storePos == storeEnd);   /* must be last sequence */
+                        ip = anchor + llen;     /* last "sequence" is a bunch of literals => don't progress anchor */
+                        continue;   /* will finish */
                     }
-                }
 
-                ZSTD_updateStats(optStatePtr, llen, anchor, offset, mlen);
-                ZSTD_storeSeq(seqStore, llen, anchor, offset, mlen-MINMATCH);
-                anchor = ip;
-        }   }
-        ZSTD_setLog2Prices(optStatePtr);
+                    /* repcodes update : like ZSTD_updateRep(), but update in place */
+                    if (offCode >= ZSTD_REP_NUM) {  /* full offset */
+                        rep[2] = rep[1];
+                        rep[1] = rep[0];
+                        rep[0] = offCode - ZSTD_REP_MOVE;
+                    } else {   /* repcode */
+                        U32 const repCode = offCode + (llen==0);
+                        if (repCode) {  /* note : if repCode==0, no change */
+                            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+                            if (repCode >= 2) rep[2] = rep[1];
+                            rep[1] = rep[0];
+                            rep[0] = currentOffset;
+                    }   }
+
+                    assert(anchor + llen <= iend);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, offCode, mlen-MINMATCH);
+                    anchor += advance;
+                    ip = anchor;
+            }   }
+            ZSTD_setBasePrices(optStatePtr, optLevel);
+        }
+
     }   /* while (ip < ilimit) */
 
     /* Return the last literals size */
@@ -895,29 +1039,94 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
 
 size_t ZSTD_compressBlock_btopt(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_compressBlock_btopt");
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 0 /*extDict*/);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_noDict);
+}
+
+
+/* used in 2-pass strategy */
+static U32 ZSTD_upscaleStat(U32* table, U32 lastEltIndex, int bonus)
+{
+    U32 s, sum=0;
+    assert(ZSTD_FREQ_DIV+bonus > 0);
+    for (s=0; s<=lastEltIndex; s++) {
+        table[s] <<= ZSTD_FREQ_DIV+bonus;
+        table[s]--;
+        sum += table[s];
+    }
+    return sum;
+}
+
+/* used in 2-pass strategy */
+MEM_STATIC void ZSTD_upscaleStats(optState_t* optPtr)
+{
+    optPtr->litSum = ZSTD_upscaleStat(optPtr->litFreq, MaxLit, 0);
+    optPtr->litLengthSum = ZSTD_upscaleStat(optPtr->litLengthFreq, MaxLL, 1);
+    optPtr->matchLengthSum = ZSTD_upscaleStat(optPtr->matchLengthFreq, MaxML, 1);
+    optPtr->offCodeSum = ZSTD_upscaleStat(optPtr->offCodeFreq, MaxOff, 1);
 }
 
 size_t ZSTD_compressBlock_btultra(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
+#if 0
+    /* 2-pass strategy (disabled)
+     * this strategy makes a first pass over first block to collect statistics
+     * and seed next round's statistics with it.
+     * The compression ratio gain is generally small (~0.5% on first block),
+     * the cost is 2x cpu time on first block. */
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    if ( (ms->opt.litLengthSum==0)   /* first block */
+      && (seqStore->sequences == seqStore->sequencesStart)   /* no ldm */
+      && (ms->window.dictLimit == ms->window.lowLimit) ) {   /* no dictionary */
+        U32 tmpRep[ZSTD_REP_NUM];
+        DEBUGLOG(5, "ZSTD_compressBlock_btultra: first block: collecting statistics");
+        assert(ms->nextToUpdate >= ms->window.dictLimit
+            && ms->nextToUpdate <= ms->window.dictLimit + 1);
+        memcpy(tmpRep, rep, sizeof(tmpRep));
+        ZSTD_compressBlock_opt_generic(ms, seqStore, tmpRep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);   /* generate stats into ms->opt*/
+        ZSTD_resetSeqStore(seqStore);
+        /* invalidate first scan from history */
+        ms->window.base -= srcSize;
+        ms->window.dictLimit += (U32)srcSize;
+        ms->window.lowLimit = ms->window.dictLimit;
+        ms->nextToUpdate = ms->window.dictLimit;
+        ms->nextToUpdate3 = ms->window.dictLimit;
+        /* re-inforce weight of collected statistics */
+        ZSTD_upscaleStats(&ms->opt);
+    }
+#endif
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 0 /*extDict*/);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_dictMatchState);
 }
 
 size_t ZSTD_compressBlock_btopt_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 1 /*extDict*/);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_extDict);
 }
 
 size_t ZSTD_compressBlock_btultra_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 1 /*extDict*/);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_extDict);
 }
diff --git a/lib/compress/zstd_opt.h b/lib/compress/zstd_opt.h
index b8dc389f317a..eeadb604c6a8 100644
--- a/lib/compress/zstd_opt.h
+++ b/lib/compress/zstd_opt.h
@@ -17,23 +17,29 @@ extern "C" {
 
 #include "zstd_compress_internal.h"
 
-void ZSTD_updateTree(
-        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
-        const BYTE* ip, const BYTE* iend);  /* used in ZSTD_loadDictionaryContent() */
+/* used in ZSTD_loadDictionaryContent() */
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend);
 
 size_t ZSTD_compressBlock_btopt(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btultra(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
 size_t ZSTD_compressBlock_btopt_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 size_t ZSTD_compressBlock_btultra_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+        void const* src, size_t srcSize);
 
 #if defined (__cplusplus)
 }
diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c
index c7a205d8c757..f4aba1d2c494 100644
--- a/lib/compress/zstdmt_compress.c
+++ b/lib/compress/zstdmt_compress.c
@@ -37,18 +37,19 @@
 #define ZSTD_RESIZE_SEQPOOL 0
 
 /* ======   Debug   ====== */
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=2) \
+    && !defined(_MSC_VER) \
+    && !defined(__MINGW32__)
 
 #  include <stdio.h>
 #  include <unistd.h>
 #  include <sys/times.h>
-#  define DEBUGLOGRAW(l, ...) if (l<=ZSTD_DEBUG) { fprintf(stderr, __VA_ARGS__); }
 
 #  define DEBUG_PRINTHEX(l,p,n) {            \
     unsigned debug_u;                        \
     for (debug_u=0; debug_u<(n); debug_u++)  \
-        DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
-    DEBUGLOGRAW(l, " \n");                   \
+        RAWLOG(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
+    RAWLOG(l, " \n");                        \
 }
 
 static unsigned long long GetCurrentClockTimeMicroseconds(void)
@@ -62,7 +63,7 @@ static unsigned long long GetCurrentClockTimeMicroseconds(void)
 
 #define MUTEX_WAIT_TIME_DLEVEL 6
 #define ZSTD_PTHREAD_MUTEX_LOCK(mutex) {          \
-    if (ZSTD_DEBUG >= MUTEX_WAIT_TIME_DLEVEL) {   \
+    if (DEBUGLEVEL >= MUTEX_WAIT_TIME_DLEVEL) {   \
         unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
         ZSTD_pthread_mutex_lock(mutex);           \
         {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
@@ -160,6 +161,25 @@ static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const
     ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
 }
 
+
+static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, U32 nbWorkers)
+{
+    unsigned const maxNbBuffers = 2*nbWorkers + 3;
+    if (srcBufPool==NULL) return NULL;
+    if (srcBufPool->totalBuffers >= maxNbBuffers) /* good enough */
+        return srcBufPool;
+    /* need a larger buffer pool */
+    {   ZSTD_customMem const cMem = srcBufPool->cMem;
+        size_t const bSize = srcBufPool->bufferSize;   /* forward parameters */
+        ZSTDMT_bufferPool* newBufPool;
+        ZSTDMT_freeBufferPool(srcBufPool);
+        newBufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+        if (newBufPool==NULL) return newBufPool;
+        ZSTDMT_setBufferSize(newBufPool, bSize);
+        return newBufPool;
+    }
+}
+
 /** ZSTDMT_getBuffer() :
  *  assumption : bufPool must be valid
  * @return : a buffer, with start pointer and size
@@ -229,8 +249,8 @@ static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
 /* store buffer for later re-use, up to pool capacity */
 static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
 {
-    if (buf.start == NULL) return;   /* compatible with release on NULL */
     DEBUGLOG(5, "ZSTDMT_releaseBuffer");
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
     ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
     if (bufPool->nbBuffers < bufPool->totalBuffers) {
         bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
@@ -300,7 +320,8 @@ static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
 
 static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
 {
-    ZSTDMT_seqPool* seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    if (seqPool == NULL) return NULL;
     ZSTDMT_setNbSeq(seqPool, 0);
     return seqPool;
 }
@@ -310,6 +331,10 @@ static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
     ZSTDMT_freeBufferPool(seqPool);
 }
 
+static ZSTDMT_seqPool* ZSTDMT_expandSeqPool(ZSTDMT_seqPool* pool, U32 nbWorkers)
+{
+    return ZSTDMT_expandBufferPool(pool, nbWorkers);
+}
 
 
 /* =====   CCtx Pool   ===== */
@@ -355,6 +380,18 @@ static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbWorkers,
     return cctxPool;
 }
 
+static ZSTDMT_CCtxPool* ZSTDMT_expandCCtxPool(ZSTDMT_CCtxPool* srcPool,
+                                              unsigned nbWorkers)
+{
+    if (srcPool==NULL) return NULL;
+    if (nbWorkers <= srcPool->totalCCtx) return srcPool;   /* good enough */
+    /* need a larger cctx pool */
+    {   ZSTD_customMem const cMem = srcPool->cMem;
+        ZSTDMT_freeCCtxPool(srcPool);
+        return ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    }
+}
+
 /* only works during initialization phase, not during compression */
 static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
 {
@@ -425,12 +462,11 @@ typedef struct {
     ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
 } serialState_t;
 
-static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool* seqPool, ZSTD_CCtx_params params)
+static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool* seqPool, ZSTD_CCtx_params params, size_t jobSize)
 {
     /* Adjust parameters */
     if (params.ldmParams.enableLdm) {
         DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
-        params.ldmParams.windowLog = params.cParams.windowLog;
         ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
         assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
         assert(params.ldmParams.hashEveryLog < 32);
@@ -453,7 +489,7 @@ static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool*
             serialState->params.ldmParams.hashLog -
             serialState->params.ldmParams.bucketSizeLog;
         /* Size the seq pool tables */
-        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, params.jobSize));
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, jobSize));
         /* Reset the window */
         ZSTD_window_clear(&serialState->ldmState.window);
         serialState->ldmWindow = serialState->ldmState.window;
@@ -473,6 +509,7 @@ static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool*
         memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
     }
     serialState->params = params;
+    serialState->params.jobSize = (U32)jobSize;
     return 0;
 }
 
@@ -505,6 +542,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
     /* Wait for our turn */
     ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
     while (serialState->nextJobID < jobID) {
+        DEBUGLOG(5, "wait for serialState->cond");
         ZSTD_pthread_cond_wait(&serialState->cond, &serialState->mutex);
     }
     /* A future job may error and skip our job */
@@ -514,6 +552,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
             size_t error;
             assert(seqStore.seq != NULL && seqStore.pos == 0 &&
                    seqStore.size == 0 && seqStore.capacity > 0);
+            assert(src.size <= serialState->params.jobSize);
             ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
             error = ZSTD_ldm_generateSequences(
                 &serialState->ldmState, &seqStore,
@@ -593,14 +632,32 @@ typedef struct {
     unsigned frameChecksumNeeded;        /* used only by mtctx */
 } ZSTDMT_jobDescription;
 
+#define JOB_ERROR(e) {                          \
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);   \
+    job->cSize = e;                             \
+    ZSTD_pthread_mutex_unlock(&job->job_mutex); \
+    goto _endJob;                               \
+}
+
 /* ZSTDMT_compressionJob() is a POOL_function type */
-void ZSTDMT_compressionJob(void* jobDescription)
+static void ZSTDMT_compressionJob(void* jobDescription)
 {
     ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
     ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
     ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
     rawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
     buffer_t dstBuff = job->dstBuff;
+    size_t lastCBlockSize = 0;
+
+    /* ressources */
+    if (cctx==NULL) JOB_ERROR(ERROR(memory_allocation));
+    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
+        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
+    }
+    if (jobParams.ldmParams.enableLdm && rawSeqStore.seq == NULL)
+        JOB_ERROR(ERROR(memory_allocation));
 
     /* Don't compute the checksum for chunks, since we compute it externally,
      * but write it in the header.
@@ -609,47 +666,31 @@ void ZSTDMT_compressionJob(void* jobDescription)
     /* Don't run LDM for the chunks, since we handle it externally */
     jobParams.ldmParams.enableLdm = 0;
 
-    /* ressources */
-    if (cctx==NULL) {
-        job->cSize = ERROR(memory_allocation);
-        goto _endJob;
-    }
-    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
-        dstBuff = ZSTDMT_getBuffer(job->bufPool);
-        if (dstBuff.start==NULL) {
-            job->cSize = ERROR(memory_allocation);
-            goto _endJob;
-        }
-        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
-    }
 
     /* init */
     if (job->cdict) {
-        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, job->cdict, jobParams, job->fullFrameSize);
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast, job->cdict, jobParams, job->fullFrameSize);
         assert(job->firstJob);  /* only allowed for first job */
-        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
+        if (ZSTD_isError(initError)) JOB_ERROR(initError);
     } else {  /* srcStart points at reloaded section */
         U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
         {   size_t const forceWindowError = ZSTD_CCtxParam_setParameter(&jobParams, ZSTD_p_forceMaxWindow, !job->firstJob);
-            if (ZSTD_isError(forceWindowError)) {
-                job->cSize = forceWindowError;
-                goto _endJob;
-        }   }
+            if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
+        }
         {   size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
                                         job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
+                                        ZSTD_dtlm_fast,
                                         NULL, /*cdict*/
                                         jobParams, pledgedSrcSize);
-            if (ZSTD_isError(initError)) {
-                job->cSize = initError;
-                goto _endJob;
-    }   }   }
+            if (ZSTD_isError(initError)) JOB_ERROR(initError);
+    }   }
 
     /* Perform serial step as early as possible, but after CCtx initialization */
     ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
 
     if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
         size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
-        if (ZSTD_isError(hSize)) { job->cSize = hSize; /* save error code */ goto _endJob; }
+        if (ZSTD_isError(hSize)) JOB_ERROR(hSize);
         DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
         ZSTD_invalidateRepCodes(cctx);
     }
@@ -667,7 +708,7 @@ void ZSTDMT_compressionJob(void* jobDescription)
         assert(job->cSize == 0);
         for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
             size_t const cSize = ZSTD_compressContinue(cctx, op, oend-op, ip, chunkSize);
-            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
             ip += chunkSize;
             op += cSize; assert(op < oend);
             /* stats */
@@ -680,18 +721,16 @@ void ZSTDMT_compressionJob(void* jobDescription)
             ZSTD_pthread_mutex_unlock(&job->job_mutex);
         }
         /* last block */
-        assert(chunkSize > 0); assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
+        assert(chunkSize > 0);
+        assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
         if ((nbChunks > 0) | job->lastJob /*must output a "last block" flag*/ ) {
             size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
             size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
             size_t const cSize = (job->lastJob) ?
                  ZSTD_compressEnd     (cctx, op, oend-op, ip, lastBlockSize) :
                  ZSTD_compressContinue(cctx, op, oend-op, ip, lastBlockSize);
-            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
-            /* stats */
-            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
-            job->cSize += cSize;
-            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
+            lastCBlockSize = cSize;
     }   }
 
 _endJob:
@@ -704,7 +743,9 @@ _endJob:
     ZSTDMT_releaseCCtx(job->cctxPool, cctx);
     /* report */
     ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
-    job->consumed = job->src.size;
+    if (ZSTD_isError(job->cSize)) assert(lastCBlockSize == 0);
+    job->cSize += lastCBlockSize;
+    job->consumed = job->src.size;  /* when job->consumed == job->src.size , compression job is presumed completed */
     ZSTD_pthread_cond_signal(&job->job_cond);
     ZSTD_pthread_mutex_unlock(&job->job_mutex);
 }
@@ -745,9 +786,9 @@ struct ZSTDMT_CCtx_s {
     ZSTD_CCtx_params params;
     size_t targetSectionSize;
     size_t targetPrefixSize;
-    roundBuff_t roundBuff;
+    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create a new job. */
     inBuff_t inBuff;
-    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create another one. */
+    roundBuff_t roundBuff;
     serialState_t serial;
     unsigned singleBlockingThread;
     unsigned jobIDMask;
@@ -798,6 +839,20 @@ static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_custom
     return jobTable;
 }
 
+static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
+    U32 nbJobs = nbWorkers + 2;
+    if (nbJobs > mtctx->jobIDMask+1) {  /* need more job capacity */
+        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+        mtctx->jobIDMask = 0;
+        mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, mtctx->cMem);
+        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
+        assert((nbJobs != 0) && ((nbJobs & (nbJobs - 1)) == 0));  /* ensure nbJobs is a power of 2 */
+        mtctx->jobIDMask = nbJobs - 1;
+    }
+    return 0;
+}
+
+
 /* ZSTDMT_CCtxParam_setNbWorkers():
  * Internal use only */
 size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
@@ -875,7 +930,7 @@ static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
         unsigned const jobID = mtctx->doneJobID & mtctx->jobIDMask;
         ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
         while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
-            DEBUGLOG(5, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
+            DEBUGLOG(4, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
             ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
         }
         ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
@@ -924,6 +979,8 @@ size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
         if ( (value > 0)  /* value==0 => automatic job size */
            & (value < ZSTDMT_JOBSIZE_MIN) )
             value = ZSTDMT_JOBSIZE_MIN;
+        if (value > ZSTDMT_JOBSIZE_MAX)
+            value = ZSTDMT_JOBSIZE_MAX;
         params->jobSize = value;
         return value;
     case ZSTDMT_p_overlapSectionLog :
@@ -950,6 +1007,21 @@ size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter,
     }
 }
 
+size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned* value)
+{
+    switch (parameter) {
+    case ZSTDMT_p_jobSize:
+        *value = mtctx->params.jobSize;
+        break;
+    case ZSTDMT_p_overlapSectionLog:
+        *value = mtctx->params.overlapSizeLog;
+        break;
+    default:
+        return ERROR(parameter_unsupported);
+    }
+    return 0;
+}
+
 /* Sets parameters relevant to the compression job,
  * initializing others to default values. */
 static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
@@ -960,13 +1032,30 @@ static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
     jobParams.cParams = params.cParams;
     jobParams.fParams = params.fParams;
     jobParams.compressionLevel = params.compressionLevel;
-    jobParams.disableLiteralCompression = params.disableLiteralCompression;
 
     return jobParams;
 }
 
+
+/* ZSTDMT_resize() :
+ * @return : error code if fails, 0 on success */
+static size_t ZSTDMT_resize(ZSTDMT_CCtx* mtctx, unsigned nbWorkers)
+{
+    if (POOL_resize(mtctx->factory, nbWorkers)) return ERROR(memory_allocation);
+    CHECK_F( ZSTDMT_expandJobsTable(mtctx, nbWorkers) );
+    mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, nbWorkers);
+    if (mtctx->bufPool == NULL) return ERROR(memory_allocation);
+    mtctx->cctxPool = ZSTDMT_expandCCtxPool(mtctx->cctxPool, nbWorkers);
+    if (mtctx->cctxPool == NULL) return ERROR(memory_allocation);
+    mtctx->seqPool = ZSTDMT_expandSeqPool(mtctx->seqPool, nbWorkers);
+    if (mtctx->seqPool == NULL) return ERROR(memory_allocation);
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    return 0;
+}
+
+
 /*! ZSTDMT_updateCParams_whileCompressing() :
- *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  Updates a selected set of compression parameters, remaining compatible with currently active frame.
  *  New parameters will be applied to next compression job. */
 void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams)
 {
@@ -981,38 +1070,36 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
     }
 }
 
-/* ZSTDMT_getNbWorkers():
- * @return nb threads currently active in mtctx.
- * mtctx must be valid */
-unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx)
-{
-    assert(mtctx != NULL);
-    return mtctx->params.nbWorkers;
-}
-
 /* ZSTDMT_getFrameProgression():
  * tells how much data has been consumed (input) and produced (output) for current frame.
  * able to count progression inside worker threads.
- * Note : mutex will be acquired during statistics collection. */
+ * Note : mutex will be acquired during statistics collection inside workers. */
 ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
 {
     ZSTD_frameProgression fps;
-    DEBUGLOG(6, "ZSTDMT_getFrameProgression");
-    fps.consumed = mtctx->consumed;
-    fps.produced = mtctx->produced;
+    DEBUGLOG(5, "ZSTDMT_getFrameProgression");
     fps.ingested = mtctx->consumed + mtctx->inBuff.filled;
+    fps.consumed = mtctx->consumed;
+    fps.produced = fps.flushed = mtctx->produced;
+    fps.currentJobID = mtctx->nextJobID;
+    fps.nbActiveWorkers = 0;
     {   unsigned jobNb;
         unsigned lastJobNb = mtctx->nextJobID + mtctx->jobReady; assert(mtctx->jobReady <= 1);
         DEBUGLOG(6, "ZSTDMT_getFrameProgression: jobs: from %u to <%u (jobReady:%u)",
                     mtctx->doneJobID, lastJobNb, mtctx->jobReady)
         for (jobNb = mtctx->doneJobID ; jobNb < lastJobNb ; jobNb++) {
             unsigned const wJobID = jobNb & mtctx->jobIDMask;
-            ZSTD_pthread_mutex_lock(&mtctx->jobs[wJobID].job_mutex);
-            {   size_t const cResult = mtctx->jobs[wJobID].cSize;
+            ZSTDMT_jobDescription* jobPtr = &mtctx->jobs[wJobID];
+            ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+            {   size_t const cResult = jobPtr->cSize;
                 size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
-                fps.consumed += mtctx->jobs[wJobID].consumed;
-                fps.ingested += mtctx->jobs[wJobID].src.size;
+                size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+                assert(flushed <= produced);
+                fps.ingested += jobPtr->src.size;
+                fps.consumed += jobPtr->consumed;
                 fps.produced += produced;
+                fps.flushed  += flushed;
+                fps.nbActiveWorkers += (jobPtr->consumed < jobPtr->src.size);
             }
             ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
         }
@@ -1021,6 +1108,34 @@ ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
 }
 
 
+size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
+{
+    size_t toFlush;
+    unsigned const jobID = mtctx->doneJobID;
+    assert(jobID <= mtctx->nextJobID);
+    if (jobID == mtctx->nextJobID) return 0;   /* no active job => nothing to flush */
+
+    /* look into oldest non-fully-flushed job */
+    {   unsigned const wJobID = jobID & mtctx->jobIDMask;
+        ZSTDMT_jobDescription* const jobPtr = &mtctx->jobs[wJobID];
+        ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+        {   size_t const cResult = jobPtr->cSize;
+            size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+            size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+            assert(flushed <= produced);
+            toFlush = produced - flushed;
+            if (toFlush==0 && (jobPtr->consumed >= jobPtr->src.size)) {
+                /* doneJobID is not-fully-flushed, but toFlush==0 : doneJobID should be compressing some more data */
+                assert(jobPtr->consumed < jobPtr->src.size);
+            }
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+    }
+
+    return toFlush;
+}
+
+
 /* ------------------------------------------ */
 /* =====   Multi-threaded compression   ===== */
 /* ------------------------------------------ */
@@ -1087,18 +1202,10 @@ static size_t ZSTDMT_compress_advanced_internal(
 
     assert(avgJobSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
     ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
-    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, avgJobSize))
         return ERROR(memory_allocation);
 
-    if (nbJobs > mtctx->jobIDMask+1) {  /* enlarge job table */
-        U32 jobsTableSize = nbJobs;
-        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
-        mtctx->jobIDMask = 0;
-        mtctx->jobs = ZSTDMT_createJobsTable(&jobsTableSize, mtctx->cMem);
-        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
-        assert((jobsTableSize != 0) && ((jobsTableSize & (jobsTableSize - 1)) == 0));  /* ensure jobsTableSize is a power of 2 */
-        mtctx->jobIDMask = jobsTableSize - 1;
-    }
+    CHECK_F( ZSTDMT_expandJobsTable(mtctx, nbJobs) );  /* only expands if necessary */
 
     {   unsigned u;
         for (u=0; u<nbJobs; u++) {
@@ -1221,17 +1328,18 @@ size_t ZSTDMT_initCStream_internal(
         const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
         unsigned long long pledgedSrcSize)
 {
-    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u, disableLiteralCompression=%i)",
-                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx, params.disableLiteralCompression);
-    /* params are supposed to be fully validated at this point */
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u)",
+                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx);
+
+    /* params supposed partially fully validated at this point */
     assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
-    assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
 
     /* init */
-    if (params.jobSize == 0) {
-        params.jobSize = 1U << ZSTDMT_computeTargetJobLog(params);
-    }
+    if (params.nbWorkers != mtctx->params.nbWorkers)
+        CHECK_F( ZSTDMT_resize(mtctx, params.nbWorkers) );
+
+    if (params.jobSize > 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
     if (params.jobSize > ZSTDMT_JOBSIZE_MAX) params.jobSize = ZSTDMT_JOBSIZE_MAX;
 
     mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN);  /* do not trigger multi-threading when srcSize is too small */
@@ -1270,7 +1378,9 @@ size_t ZSTDMT_initCStream_internal(
     mtctx->targetPrefixSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
     DEBUGLOG(4, "overlapLog=%u => %u KB", params.overlapSizeLog, (U32)(mtctx->targetPrefixSize>>10));
     mtctx->targetSectionSize = params.jobSize;
-    if (mtctx->targetSectionSize < ZSTDMT_JOBSIZE_MIN) mtctx->targetSectionSize = ZSTDMT_JOBSIZE_MIN;
+    if (mtctx->targetSectionSize == 0) {
+        mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(params);
+    }
     if (mtctx->targetSectionSize < mtctx->targetPrefixSize) mtctx->targetSectionSize = mtctx->targetPrefixSize;  /* job size must be >= overlap size */
     DEBUGLOG(4, "Job Size : %u KB (note : set to %u)", (U32)(mtctx->targetSectionSize>>10), params.jobSize);
     DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
@@ -1312,7 +1422,7 @@ size_t ZSTDMT_initCStream_internal(
     mtctx->allJobsCompleted = 0;
     mtctx->consumed = 0;
     mtctx->produced = 0;
-    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, mtctx->targetSectionSize))
         return ERROR(memory_allocation);
     return 0;
 }
@@ -1420,7 +1530,7 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZS
         mtctx->jobs[jobID].jobID = mtctx->nextJobID;
         mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
         mtctx->jobs[jobID].lastJob = endFrame;
-        mtctx->jobs[jobID].frameChecksumNeeded = endFrame && (mtctx->nextJobID>0) && mtctx->params.fParams.checksumFlag;
+        mtctx->jobs[jobID].frameChecksumNeeded = mtctx->params.fParams.checksumFlag && endFrame && (mtctx->nextJobID>0);
         mtctx->jobs[jobID].dstFlushed = 0;
 
         /* Update the round buffer pos and clear the input buffer to be reset */
@@ -1468,6 +1578,8 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZS
 
 
 /*! ZSTDMT_flushProduced() :
+ *  flush whatever data has been produced but not yet flushed in current job.
+ *  move to next job if current one is fully flushed.
  * `output` : `pos` will be updated with amount of data flushed .
  * `blockToFlush` : if >0, the function will block and wait if there is no data available to flush .
  * @return : amount of data remaining within internal buffer, 0 if no more, 1 if unknown but > 0, or an error code */
@@ -1496,7 +1608,7 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
     /* try to flush something */
     {   size_t cSize = mtctx->jobs[wJobID].cSize;                  /* shared */
         size_t const srcConsumed = mtctx->jobs[wJobID].consumed;   /* shared */
-        size_t const srcSize = mtctx->jobs[wJobID].src.size;        /* read-only, could be done after mutex lock, but no-declaration-after-statement */
+        size_t const srcSize = mtctx->jobs[wJobID].src.size;       /* read-only, could be done after mutex lock, but no-declaration-after-statement */
         ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
         if (ZSTD_isError(cSize)) {
             DEBUGLOG(5, "ZSTDMT_flushProduced: job %u : compression error detected : %s",
@@ -1516,6 +1628,7 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
             mtctx->jobs[wJobID].cSize += 4;  /* can write this shared value, as worker is no longer active */
             mtctx->jobs[wJobID].frameChecksumNeeded = 0;
         }
+
         if (cSize > 0) {   /* compression is ongoing or completed */
             size_t const toFlush = MIN(cSize - mtctx->jobs[wJobID].dstFlushed, output->size - output->pos);
             DEBUGLOG(5, "ZSTDMT_flushProduced: Flushing %u bytes from job %u (completion:%u/%u, generated:%u)",
@@ -1529,11 +1642,12 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
             output->pos += toFlush;
             mtctx->jobs[wJobID].dstFlushed += toFlush;  /* can write : this value is only used by mtctx */
 
-            if ( (srcConsumed == srcSize)    /* job completed */
+            if ( (srcConsumed == srcSize)    /* job is completed */
               && (mtctx->jobs[wJobID].dstFlushed == cSize) ) {   /* output buffer fully flushed => free this job position */
                 DEBUGLOG(5, "Job %u completed (%u bytes), moving to next one",
                         mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
                 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[wJobID].dstBuff);
+                DEBUGLOG(5, "dstBuffer released");
                 mtctx->jobs[wJobID].dstBuff = g_nullBuffer;
                 mtctx->jobs[wJobID].cSize = 0;   /* ensure this job slot is considered "not started" in future check */
                 mtctx->consumed += srcSize;
@@ -1610,6 +1724,7 @@ static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
     range_t extDict;
     range_t prefix;
 
+    DEBUGLOG(5, "ZSTDMT_doesOverlapWindow");
     extDict.start = window.dictBase + window.lowLimit;
     extDict.size = window.dictLimit - window.lowLimit;
 
@@ -1630,12 +1745,13 @@ static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
 {
     if (mtctx->params.ldmParams.enableLdm) {
         ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
         DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
                     (size_t)buffer.start,
                     (size_t)buffer.start + buffer.capacity);
         ZSTD_PTHREAD_MUTEX_LOCK(mutex);
         while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
-            DEBUGLOG(6, "Waiting for LDM to finish...");
+            DEBUGLOG(5, "Waiting for LDM to finish...");
             ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
         }
         DEBUGLOG(6, "Done waiting for LDM to finish");
@@ -1655,6 +1771,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
     size_t const target = mtctx->targetSectionSize;
     buffer_t buffer;
 
+    DEBUGLOG(5, "ZSTDMT_tryGetInputRange");
     assert(mtctx->inBuff.buffer.start == NULL);
     assert(mtctx->roundBuff.capacity >= target);
 
@@ -1668,7 +1785,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
         buffer.start = start;
         buffer.capacity = prefixSize;
         if (ZSTDMT_isOverlapped(buffer, inUse)) {
-            DEBUGLOG(6, "Waiting for buffer...");
+            DEBUGLOG(5, "Waiting for buffer...");
             return 0;
         }
         ZSTDMT_waitForLdmComplete(mtctx, buffer);
@@ -1680,7 +1797,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
     buffer.capacity = target;
 
     if (ZSTDMT_isOverlapped(buffer, inUse)) {
-        DEBUGLOG(6, "Waiting for buffer...");
+        DEBUGLOG(5, "Waiting for buffer...");
         return 0;
     }
     assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
@@ -1753,8 +1870,10 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
                 /* It is only possible for this operation to fail if there are
                  * still compression jobs ongoing.
                  */
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange failed");
                 assert(mtctx->doneJobID != mtctx->nextJobID);
-            }
+            } else
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange completed successfully : mtctx->inBuff.buffer.start = %p", mtctx->inBuff.buffer.start);
         }
         if (mtctx->inBuff.buffer.start != NULL) {
             size_t const toLoad = MIN(input->size - input->pos, mtctx->targetSectionSize - mtctx->inBuff.filled);
@@ -1782,6 +1901,7 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
     /* check for potential compressed data ready to be flushed */
     {   size_t const remainingToFlush = ZSTDMT_flushProduced(mtctx, output, !forwardInputProgress, endOp); /* block if there was no forward input progress */
         if (input->pos < input->size) return MAX(remainingToFlush, 1);  /* input not consumed : do not end flush yet */
+        DEBUGLOG(5, "end of ZSTDMT_compressStream_generic: remainingToFlush = %u", (U32)remainingToFlush);
         return remainingToFlush;
     }
 }
diff --git a/lib/compress/zstdmt_compress.h b/lib/compress/zstdmt_compress.h
index f79e3b441800..12ad9f899b57 100644
--- a/lib/compress/zstdmt_compress.h
+++ b/lib/compress/zstdmt_compress.h
@@ -95,6 +95,11 @@ typedef enum {
  * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
 ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value);
 
+/* ZSTDMT_getMTCtxParameter() :
+ * Query the ZSTDMT_CCtx for a parameter value.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned* value);
+
 
 /*! ZSTDMT_compressStream_generic() :
  *  Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
@@ -114,11 +119,21 @@ ZSTDLIB_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
  * ===  Not exposed in libzstd. Never invoke directly   ===
  * ======================================================== */
 
+ /*! ZSTDMT_toFlushNow()
+  *  Tell how many bytes are ready to be flushed immediately.
+  *  Probe the oldest active job (not yet entirely flushed) and check its output buffer.
+  *  If return 0, it means there is no active job,
+  *  or, it means oldest job is still active, but everything produced has been flushed so far,
+  *  therefore flushing is limited by speed of oldest job. */
+size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx);
+
+/*! ZSTDMT_CCtxParam_setMTCtxParameter()
+ *  like ZSTDMT_setMTCtxParameter(), but into a ZSTD_CCtx_Params */
 size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value);
 
-/* ZSTDMT_CCtxParam_setNbWorkers()
- * Set nbWorkers, and clamp it.
- * Also reset jobSize and overlapLog */
+/*! ZSTDMT_CCtxParam_setNbWorkers()
+ *  Set nbWorkers, and clamp it.
+ *  Also reset jobSize and overlapLog */
 size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers);
 
 /*! ZSTDMT_updateCParams_whileCompressing() :
@@ -126,14 +141,9 @@ size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorker
  *  New parameters will be applied to next compression job. */
 void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams);
 
-/* ZSTDMT_getNbWorkers():
- * @return nb threads currently active in mtctx.
- * mtctx must be valid */
-unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx);
-
-/* ZSTDMT_getFrameProgression():
- * tells how much data has been consumed (input) and produced (output) for current frame.
- * able to count progression inside worker threads.
+/*! ZSTDMT_getFrameProgression():
+ *  tells how much data has been consumed (input) and produced (output) for current frame.
+ *  able to count progression inside worker threads.
  */
 ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
 
diff --git a/lib/decompress/huf_decompress.c b/lib/decompress/huf_decompress.c
index 73f5c46c0615..83ecaff01e84 100644
--- a/lib/decompress/huf_decompress.c
+++ b/lib/decompress/huf_decompress.c
@@ -1,6 +1,7 @@
 /* ******************************************************************
-   Huffman decoder, part of New Generation Entropy library
-   Copyright (C) 2013-2016, Yann Collet.
+   huff0 huffman decoder,
+   part of Finite State Entropy library
+   Copyright (C) 2013-present, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -29,16 +30,15 @@
 
     You can contact the author at :
     - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
 ****************************************************************** */
 
 /* **************************************************************
 *  Dependencies
 ****************************************************************/
 #include <string.h>     /* memcpy, memset */
-#include "bitstream.h"  /* BIT_* */
 #include "compiler.h"
-#include "fse.h"        /* header compression */
+#include "bitstream.h"  /* BIT_* */
+#include "fse.h"        /* to compress headers */
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
 #include "error_private.h"
@@ -48,7 +48,6 @@
 *  Error Management
 ****************************************************************/
 #define HUF_isError ERR_isError
-#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
 #define CHECK_F(f) { size_t const err_ = (f); if (HUF_isError(err_)) return err_; }
 
 
@@ -75,15 +74,15 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
 /*-***************************/
 /*  single-symbol decoding   */
 /*-***************************/
-typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decoding */
 
-size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
 {
     U32 tableLog = 0;
     U32 nbSymbols = 0;
     size_t iSize;
     void* const dtPtr = DTable + 1;
-    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+    HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
 
     U32* rankVal;
     BYTE* huffWeight;
@@ -96,7 +95,7 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
 
     if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
 
-    HUF_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
+    DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
     iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
@@ -124,7 +123,7 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
             U32 const w = huffWeight[n];
             U32 const length = (1 << w) >> 1;
             U32 u;
-            HUF_DEltX2 D;
+            HUF_DEltX1 D;
             D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
             for (u = rankVal[w]; u < rankVal[w] + length; u++)
                 dt[u] = D;
@@ -134,17 +133,15 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
     return iSize;
 }
 
-size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_readDTableX2_wksp(DTable, src, srcSize,
+    return HUF_readDTableX1_wksp(DTable, src, srcSize,
                                  workSpace, sizeof(workSpace));
 }
 
-typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
-
 FORCE_INLINE_TEMPLATE BYTE
-HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
     BYTE const c = dt[val].byte;
@@ -152,44 +149,44 @@ HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog
     return c;
 }
 
-#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
-    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+#define HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog)
 
-#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr)  \
+#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)  \
     if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
 
-#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr) \
     if (MEM_64bits()) \
-        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
 
 HINT_INLINE size_t
-HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX1* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
     /* up to 4 symbols at a time */
     while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
     }
 
     /* [0-3] symbols remaining */
     if (MEM_32bits())
         while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
-            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
     /* no more data to retrieve from bitstream, no need to reload */
     while (p < pEnd)
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
     return pEnd-pStart;
 }
 
 FORCE_INLINE_TEMPLATE size_t
-HUF_decompress1X2_usingDTable_internal_body(
+HUF_decompress1X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -197,14 +194,14 @@ HUF_decompress1X2_usingDTable_internal_body(
     BYTE* op = (BYTE*)dst;
     BYTE* const oend = op + dstSize;
     const void* dtPtr = DTable + 1;
-    const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+    const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
     BIT_DStream_t bitD;
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
     U32 const dtLog = dtd.tableLog;
 
     CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
 
-    HUF_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+    HUF_decodeStreamX1(op, &bitD, oend, dt, dtLog);
 
     if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
 
@@ -212,7 +209,7 @@ HUF_decompress1X2_usingDTable_internal_body(
 }
 
 FORCE_INLINE_TEMPLATE size_t
-HUF_decompress4X2_usingDTable_internal_body(
+HUF_decompress4X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -224,7 +221,7 @@ HUF_decompress4X2_usingDTable_internal_body(
         BYTE* const ostart = (BYTE*) dst;
         BYTE* const oend = ostart + dstSize;
         const void* const dtPtr = DTable + 1;
-        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+        const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
 
         /* Init */
         BIT_DStream_t bitD1;
@@ -260,22 +257,22 @@ HUF_decompress4X2_usingDTable_internal_body(
         /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
         endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
         while ( (endSignal==BIT_DStream_unfinished) && (op4<(oend-3)) ) {
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
             BIT_reloadDStream(&bitD1);
             BIT_reloadDStream(&bitD2);
             BIT_reloadDStream(&bitD3);
@@ -291,191 +288,10 @@ HUF_decompress4X2_usingDTable_internal_body(
         /* note : op4 supposed already verified within main loop */
 
         /* finish bitStreams one by one */
-        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
-        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
-        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
-        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
-
-        /* check */
-        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
-          if (!endCheck) return ERROR(corruption_detected); }
-
-        /* decoded size */
-        return dstSize;
-    }
-}
-
-
-FORCE_INLINE_TEMPLATE U32
-HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
-{
-    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 2);
-    BIT_skipBits(DStream, dt[val].nbBits);
-    return dt[val].length;
-}
-
-FORCE_INLINE_TEMPLATE U32
-HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
-{
-    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 1);
-    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
-    else {
-        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
-            BIT_skipBits(DStream, dt[val].nbBits);
-            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
-                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
-                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
-    }   }
-    return 1;
-}
-
-#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
-    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
-    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
-    if (MEM_64bits()) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-HINT_INLINE size_t
-HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
-                const HUF_DEltX4* const dt, const U32 dtLog)
-{
-    BYTE* const pStart = p;
-
-    /* up to 8 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
-    }
-
-    /* closer to end : up to 2 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
-
-    while (p <= pEnd-2)
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
-
-    if (p < pEnd)
-        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
-
-    return p-pStart;
-}
-
-FORCE_INLINE_TEMPLATE size_t
-HUF_decompress1X4_usingDTable_internal_body(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    BIT_DStream_t bitD;
-
-    /* Init */
-    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
-
-    /* decode */
-    {   BYTE* const ostart = (BYTE*) dst;
-        BYTE* const oend = ostart + dstSize;
-        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
-        DTableDesc const dtd = HUF_getDTableDesc(DTable);
-        HUF_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
-    }
-
-    /* check */
-    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
-
-    /* decoded size */
-    return dstSize;
-}
-
-
-FORCE_INLINE_TEMPLATE size_t
-HUF_decompress4X4_usingDTable_internal_body(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
-
-    {   const BYTE* const istart = (const BYTE*) cSrc;
-        BYTE* const ostart = (BYTE*) dst;
-        BYTE* const oend = ostart + dstSize;
-        const void* const dtPtr = DTable+1;
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
-
-        /* Init */
-        BIT_DStream_t bitD1;
-        BIT_DStream_t bitD2;
-        BIT_DStream_t bitD3;
-        BIT_DStream_t bitD4;
-        size_t const length1 = MEM_readLE16(istart);
-        size_t const length2 = MEM_readLE16(istart+2);
-        size_t const length3 = MEM_readLE16(istart+4);
-        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
-        const BYTE* const istart1 = istart + 6;  /* jumpTable */
-        const BYTE* const istart2 = istart1 + length1;
-        const BYTE* const istart3 = istart2 + length2;
-        const BYTE* const istart4 = istart3 + length3;
-        size_t const segmentSize = (dstSize+3) / 4;
-        BYTE* const opStart2 = ostart + segmentSize;
-        BYTE* const opStart3 = opStart2 + segmentSize;
-        BYTE* const opStart4 = opStart3 + segmentSize;
-        BYTE* op1 = ostart;
-        BYTE* op2 = opStart2;
-        BYTE* op3 = opStart3;
-        BYTE* op4 = opStart4;
-        U32 endSignal;
-        DTableDesc const dtd = HUF_getDTableDesc(DTable);
-        U32 const dtLog = dtd.tableLog;
-
-        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
-        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
-        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
-        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
-
-        /* 16-32 symbols per loop (4-8 symbols per stream) */
-        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
-
-            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        }
-
-        /* check corruption */
-        if (op1 > opStart2) return ERROR(corruption_detected);
-        if (op2 > opStart3) return ERROR(corruption_detected);
-        if (op3 > opStart4) return ERROR(corruption_detected);
-        /* note : op4 already verified within main loop */
-
-        /* finish bitStreams one by one */
-        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
-        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
-        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
-        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+        HUF_decodeStreamX1(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX1(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX1(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX1(op4, &bitD4, oend,     dt, dtLog);
 
         /* check */
         { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
@@ -493,7 +309,7 @@ typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
                                                const HUF_DTable *DTable);
 #if DYNAMIC_BMI2
 
-#define X(fn)                                                               \
+#define HUF_DGEN(fn)                                                               \
                                                                             \
     static size_t fn##_default(                                             \
                   void* dst,  size_t dstSize,                               \
@@ -522,7 +338,7 @@ typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
 
 #else
 
-#define X(fn)                                                               \
+#define HUF_DGEN(fn)                                                               \
     static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
                      size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
     {                                                                       \
@@ -532,112 +348,114 @@ typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
 
 #endif
 
-X(HUF_decompress1X2_usingDTable_internal)
-X(HUF_decompress4X2_usingDTable_internal)
-X(HUF_decompress1X4_usingDTable_internal)
-X(HUF_decompress4X4_usingDTable_internal)
+HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
+HUF_DGEN(HUF_decompress4X1_usingDTable_internal)
 
-#undef X
 
 
-size_t HUF_decompress1X2_usingDTable(
+size_t HUF_decompress1X1_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
+    size_t const hSize = HUF_readDTableX1_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
 }
 
 
-size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
                               const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
 
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
-size_t HUF_decompress4X2_usingDTable(
+size_t HUF_decompress4X1_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize, int bmi2)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX2_wksp (dctx, cSrc, cSrcSize,
+    size_t const hSize = HUF_readDTableX1_wksp (dctx, cSrc, cSrcSize,
                                                 workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
 }
 
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
-    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
+    return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
 }
 
 
-size_t HUF_decompress4X2_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
 
 /* *************************/
 /* double-symbols decoding */
 /* *************************/
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2;  /* double-symbols decoding */
 typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
-/* HUF_fillDTableX4Level2() :
+
+/* HUF_fillDTableX2Level2() :
  * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
-static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 consumed,
                            const U32* rankValOrigin, const int minWeight,
                            const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
                            U32 nbBitsBaseline, U16 baseSeq)
 {
-    HUF_DEltX4 DElt;
+    HUF_DEltX2 DElt;
     U32 rankVal[HUF_TABLELOG_MAX + 1];
 
     /* get pre-calculated rankVal */
@@ -672,10 +490,8 @@ static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 co
     }   }
 }
 
-typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
-typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
-static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
                            const sortedSymbol_t* sortedList, const U32 sortedListSize,
                            const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
                            const U32 nbBitsBaseline)
@@ -700,12 +516,12 @@ static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
             int minWeight = nbBits + scaleLog;
             if (minWeight < 1) minWeight = 1;
             sortedRank = rankStart[minWeight];
-            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+            HUF_fillDTableX2Level2(DTable+start, targetLog-nbBits, nbBits,
                            rankValOrigin[nbBits], minWeight,
                            sortedList+sortedRank, sortedListSize-sortedRank,
                            nbBitsBaseline, symbol);
         } else {
-            HUF_DEltX4 DElt;
+            HUF_DEltX2 DElt;
             MEM_writeLE16(&(DElt.sequence), symbol);
             DElt.nbBits = (BYTE)(nbBits);
             DElt.length = 1;
@@ -717,16 +533,16 @@ static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
     }
 }
 
-size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
-                             size_t srcSize, void* workSpace,
-                             size_t wkspSize)
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
+                       const void* src, size_t srcSize,
+                             void* workSpace, size_t wkspSize)
 {
     U32 tableLog, maxW, sizeOfSort, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     U32 const maxTableLog = dtd.maxTableLog;
     size_t iSize;
     void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
-    HUF_DEltX4* const dt = (HUF_DEltX4*)dtPtr;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
     U32 *rankStart;
 
     rankValCol_t* rankVal;
@@ -752,7 +568,7 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
     rankStart = rankStart0 + 1;
     memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
 
-    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
     if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
@@ -806,7 +622,7 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
                     rankValPtr[w] = rankVal0[w] >> consumed;
     }   }   }   }
 
-    HUF_fillDTableX4(dt, maxTableLog,
+    HUF_fillDTableX2(dt, maxTableLog,
                    sortedSymbol, sizeOfSort,
                    rankStart0, rankVal, maxW,
                    tableLog+1);
@@ -817,112 +633,296 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
     return iSize;
 }
 
-size_t HUF_readDTableX4(HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
 {
   U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-  return HUF_readDTableX4_wksp(DTable, src, srcSize,
+  return HUF_readDTableX2_wksp(DTable, src, srcSize,
                                workSpace, sizeof(workSpace));
 }
 
-size_t HUF_decompress1X4_usingDTable(
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
+    }   }
+    return 1;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
+
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BIT_DStream_t bitD;
+
+    /* Init */
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        HUF_decodeStreamX2(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
+HUF_DGEN(HUF_decompress4X2_usingDTable_internal)
+
+size_t HUF_decompress1X2_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX4_wksp(DCtx, cSrc, cSrcSize,
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
                                                workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
 }
 
 
-size_t HUF_decompress1X4_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
                               const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X4_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
 
-size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
-size_t HUF_decompress4X4_usingDTable(
+size_t HUF_decompress4X2_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-static size_t HUF_decompress4X4_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize, int bmi2)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t hSize = HUF_readDTableX4_wksp(dctx, cSrc, cSrcSize,
+    size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
                                          workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
 }
 
-size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
-    return HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
+    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
 }
 
 
-size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
                               const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
 
-size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
 
-/* ********************************/
-/* Generic decompression selector */
-/* ********************************/
+/* ***********************************/
+/* Universal decompression selectors */
+/* ***********************************/
 
 size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize,
                                     const void* cSrc, size_t cSrcSize,
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
@@ -930,8 +930,8 @@ size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
 
@@ -960,12 +960,12 @@ static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, qu
 /** HUF_selectDecoder() :
  *  Tells which decoder is likely to decode faster,
  *  based on a set of pre-computed metrics.
- * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
  *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
 {
     assert(dstSize > 0);
-    assert(dstSize <= 128 KB);
+    assert(dstSize <= 128*1024);
     /* decoder timing evaluation */
     {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
         U32 const D256 = (U32)(dstSize >> 8);
@@ -980,7 +980,7 @@ typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc,
 
 size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    static const decompressionAlgo decompress[2] = { HUF_decompress4X2, HUF_decompress4X4 };
+    static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
 
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -1002,8 +1002,8 @@ size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+        return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
     }
 }
 
@@ -1025,8 +1025,8 @@ size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
     if (cSrcSize == 0) return ERROR(corruption_detected);
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize):
-                        HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize):
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
     }
 }
 
@@ -1041,9 +1041,9 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress1X4_DCtx_wksp(dctx, dst, dstSize, cSrc,
+        return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
                                 cSrcSize, workSpace, wkspSize):
-                        HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                        HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
                                 cSrcSize, workSpace, wkspSize);
     }
 }
@@ -1060,27 +1060,27 @@ size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
 size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
 }
 
-size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
 }
 
 size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
 }
 
 size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
@@ -1090,7 +1090,7 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
     if (cSrcSize == 0) return ERROR(corruption_detected);
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
-                        HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return algoNb ? HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
+                        HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
     }
 }
diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index 3ec6a1cb328b..711b5b6d7aca 100644
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -40,12 +40,24 @@
 #  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_DEFAULTMAX) + 1)
 #endif
 
+/*!
+ *  NO_FORWARD_PROGRESS_MAX :
+ *  maximum allowed nb of calls to ZSTD_decompressStream() and ZSTD_decompress_generic()
+ *  without any forward progress
+ *  (defined as: no byte read from input, and no byte flushed to output)
+ *  before triggering an error.
+ */
+#ifndef ZSTD_NO_FORWARD_PROGRESS_MAX
+#  define ZSTD_NO_FORWARD_PROGRESS_MAX 16
+#endif
+
 
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
 #include <string.h>      /* memcpy, memmove, memset */
-#include "cpu.h"
+#include "compiler.h"    /* prefetch */
+#include "cpu.h"         /* bmi2 */
 #include "mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
@@ -57,6 +69,9 @@
 #  include "zstd_legacy.h"
 #endif
 
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
+
 
 /*-*************************************
 *  Errors
@@ -99,11 +114,10 @@ typedef struct {
 #define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
 
 typedef struct {
-    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];
-    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];
-    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
     HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
-    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
     U32 rep[ZSTD_REP_NUM];
 } ZSTD_entropyDTables_t;
 
@@ -114,9 +128,10 @@ struct ZSTD_DCtx_s
     const ZSTD_seqSymbol* OFTptr;
     const HUF_DTable* HUFptr;
     ZSTD_entropyDTables_t entropy;
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];   /* space needed when building huffman tables */
     const void* previousDstEnd;   /* detect continuity */
-    const void* base;             /* start of current segment */
-    const void* vBase;            /* virtual start of previous segment if it was just before current one */
+    const void* prefixStart;      /* start of current segment */
+    const void* virtualStart;     /* virtual start of previous segment if it was just before current one */
     const void* dictEnd;          /* end of previous segment */
     size_t expected;
     ZSTD_frameHeader fParams;
@@ -127,7 +142,6 @@ struct ZSTD_DCtx_s
     U32 fseEntropy;
     XXH64_state_t xxhState;
     size_t headerSize;
-    U32 dictID;
     ZSTD_format_e format;
     const BYTE* litPtr;
     ZSTD_customMem customMem;
@@ -136,9 +150,13 @@ struct ZSTD_DCtx_s
     size_t staticSize;
     int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
 
-    /* streaming */
+    /* dictionary */
     ZSTD_DDict* ddictLocal;
-    const ZSTD_DDict* ddict;
+    const ZSTD_DDict* ddict;     /* set by ZSTD_initDStream_usingDDict(), or ZSTD_DCtx_refDDict() */
+    U32 dictID;
+    int ddictIsCold;             /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
+
+    /* streaming */
     ZSTD_dStreamStage streamStage;
     char*  inBuff;
     size_t inBuffSize;
@@ -153,6 +171,7 @@ struct ZSTD_DCtx_s
     U32 previousLegacyVersion;
     U32 legacyVersion;
     U32 hostageByte;
+    int noForwardProgress;
 
     /* workspace */
     BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
@@ -173,7 +192,7 @@ size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 static size_t ZSTD_startingInputLength(ZSTD_format_e format)
 {
     size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
-                    ZSTD_frameHeaderSize_prefix - ZSTD_frameIdSize :
+                    ZSTD_frameHeaderSize_prefix - ZSTD_FRAMEIDSIZE :
                     ZSTD_frameHeaderSize_prefix;
     ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
     /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
@@ -188,10 +207,15 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     dctx->ddict       = NULL;
     dctx->ddictLocal  = NULL;
+    dctx->dictEnd     = NULL;
+    dctx->ddictIsCold = 0;
     dctx->inBuff      = NULL;
     dctx->inBuffSize  = 0;
     dctx->outBuffSize = 0;
     dctx->streamStage = zdss_init;
+    dctx->legacyContext = NULL;
+    dctx->previousLegacyVersion = 0;
+    dctx->noForwardProgress = 0;
     dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
 }
 
@@ -215,8 +239,6 @@ ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
     {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
         if (!dctx) return NULL;
         dctx->customMem = customMem;
-        dctx->legacyContext = NULL;
-        dctx->previousLegacyVersion = 0;
         ZSTD_initDCtx_internal(dctx);
         return dctx;
     }
@@ -265,7 +287,7 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
 unsigned ZSTD_isFrame(const void* buffer, size_t size)
 {
-    if (size < ZSTD_frameIdSize) return 0;
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
     {   U32 const magic = MEM_readLE32(buffer);
         if (magic == ZSTD_MAGICNUMBER) return 1;
         if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
@@ -298,25 +320,28 @@ static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZST
 
 /** ZSTD_frameHeaderSize() :
  *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
- * @return : size of the Frame Header */
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
 size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
 {
     return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
 }
 
 
-/** ZSTD_getFrameHeader_internal() :
+/** ZSTD_getFrameHeader_advanced() :
  *  decode Frame Header, or require larger `srcSize`.
  *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
  * @return : 0, `zfhPtr` is correctly filled,
  *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
  *           or an error code, which can be tested using ZSTD_isError() */
-static size_t ZSTD_getFrameHeader_internal(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
+size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
 {
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
+    memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
     if (srcSize < minInputSize) return minInputSize;
+    if (src==NULL) return ERROR(GENERIC);   /* invalid parameter */
 
     if ( (format != ZSTD_f_zstd1_magicless)
       && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
@@ -325,7 +350,7 @@ static size_t ZSTD_getFrameHeader_internal(ZSTD_frameHeader* zfhPtr, const void*
             if (srcSize < ZSTD_skippableHeaderSize)
                 return ZSTD_skippableHeaderSize; /* magic number + frame length */
             memset(zfhPtr, 0, sizeof(*zfhPtr));
-            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_frameIdSize);
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
             zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
         }
@@ -394,7 +419,7 @@ static size_t ZSTD_getFrameHeader_internal(ZSTD_frameHeader* zfhPtr, const void*
  *           or an error code, which can be tested using ZSTD_isError() */
 size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
 {
-    return ZSTD_getFrameHeader_internal(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+    return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
 }
 
 
@@ -437,7 +462,7 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
             size_t skippableSize;
             if (srcSize < ZSTD_skippableHeaderSize)
                 return ERROR(srcSize_wrong);
-            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_frameIdSize)
+            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_FRAMEIDSIZE)
                           + ZSTD_skippableHeaderSize;
             if (srcSize < skippableSize) {
                 return ZSTD_CONTENTSIZE_ERROR;
@@ -491,7 +516,7 @@ unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
 *   @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
 static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
 {
-    size_t const result = ZSTD_getFrameHeader_internal(&(dctx->fParams), src, headerSize, dctx->format);
+    size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
     if (ZSTD_isError(result)) return result;    /* invalid header */
     if (result>0) return ERROR(srcSize_wrong);  /* headerSize too small */
     if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID))
@@ -526,6 +551,7 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
 static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
                           const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
@@ -542,6 +568,9 @@ static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
     return regenSize;
 }
 
+/* Hidden declaration for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize);
 /*! ZSTD_decodeLiteralsBlock() :
  * @return : nb of bytes read from src (< srcSize )
  *  note : symbol not declared but exposed for fullbench */
@@ -558,6 +587,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         case set_repeat:
             if (dctx->litEntropy==0) return ERROR(dictionary_corrupted);
             /* fall-through */
+
         case set_compressed:
             if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
             {   size_t lhSize, litSize, litCSize;
@@ -589,15 +619,20 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                 if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
                 if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
 
+                /* prefetch huffman table if cold */
+                if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
+                    PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
+                }
+
                 if (HUF_isError((litEncType==set_repeat) ?
                                     ( singleStream ?
                                         HUF_decompress1X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) :
                                         HUF_decompress4X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) ) :
                                     ( singleStream ?
-                                        HUF_decompress1X2_DCtx_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                         dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2) :
+                                        HUF_decompress1X1_DCtx_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                                                                         dctx->workspace, sizeof(dctx->workspace), dctx->bmi2) :
                                         HUF_decompress4X_hufOnly_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                           dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2))))
+                                                                           dctx->workspace, sizeof(dctx->workspace), dctx->bmi2))))
                     return ERROR(corruption_detected);
 
                 dctx->litPtr = dctx->litBuffer;
@@ -869,7 +904,8 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
                                  symbolEncodingType_e type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
                                  const U32* baseValue, const U32* nbAdditionalBits,
-                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable)
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
+                                 int ddictIsCold, int nbSeq)
 {
     switch(type)
     {
@@ -888,6 +924,12 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
         return 0;
     case set_repeat:
         if (!flagRepeatTable) return ERROR(corruption_detected);
+        /* prefetch FSE table if used */
+        if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
+            const void* const pStart = *DTablePtr;
+            size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
+            PREFETCH_AREA(pStart, pSize);
+        }
         return 0;
     case set_compressed :
         {   U32 tableLog;
@@ -933,6 +975,9 @@ static const U32 ML_base[MaxML+1] = {
                     67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
                     0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
 
+/* Hidden delcaration for fullbench */
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                             const void* src, size_t srcSize);
 
 size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                              const void* src, size_t srcSize)
@@ -940,25 +985,25 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
     const BYTE* const istart = (const BYTE* const)src;
     const BYTE* const iend = istart + srcSize;
     const BYTE* ip = istart;
+    int nbSeq;
     DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
 
     /* check */
     if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
 
     /* SeqHead */
-    {   int nbSeq = *ip++;
-        if (!nbSeq) { *nbSeqPtr=0; return 1; }
-        if (nbSeq > 0x7F) {
-            if (nbSeq == 0xFF) {
-                if (ip+2 > iend) return ERROR(srcSize_wrong);
-                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
-            } else {
-                if (ip >= iend) return ERROR(srcSize_wrong);
-                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
-            }
+    nbSeq = *ip++;
+    if (!nbSeq) { *nbSeqPtr=0; return 1; }
+    if (nbSeq > 0x7F) {
+        if (nbSeq == 0xFF) {
+            if (ip+2 > iend) return ERROR(srcSize_wrong);
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+        } else {
+            if (ip >= iend) return ERROR(srcSize_wrong);
+            nbSeq = ((nbSeq-0x80)<<8) + *ip++;
         }
-        *nbSeqPtr = nbSeq;
     }
+    *nbSeqPtr = nbSeq;
 
     /* FSE table descriptors */
     if (ip+4 > iend) return ERROR(srcSize_wrong); /* minimum possible size */
@@ -972,7 +1017,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       LLtype, MaxLL, LLFSELog,
                                                       ip, iend-ip,
                                                       LL_base, LL_bits,
-                                                      LL_defaultDTable, dctx->fseEntropy);
+                                                      LL_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
             ip += llhSize;
         }
@@ -981,7 +1027,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       OFtype, MaxOff, OffFSELog,
                                                       ip, iend-ip,
                                                       OF_base, OF_bits,
-                                                      OF_defaultDTable, dctx->fseEntropy);
+                                                      OF_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
             ip += ofhSize;
         }
@@ -990,12 +1037,23 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       MLtype, MaxML, MLFSELog,
                                                       ip, iend-ip,
                                                       ML_base, ML_bits,
-                                                      ML_defaultDTable, dctx->fseEntropy);
+                                                      ML_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
             ip += mlhSize;
         }
     }
 
+    /* prefetch dictionary content */
+    if (dctx->ddictIsCold) {
+        size_t const dictSize = (const char*)dctx->prefixStart - (const char*)dctx->virtualStart;
+        size_t const psmin = MIN(dictSize, (size_t)(64*nbSeq) /* heuristic */ );
+        size_t const pSize = MIN(psmin, 128 KB /* protection */ );
+        const void* const pStart = (const char*)dctx->dictEnd - pSize;
+        PREFETCH_AREA(pStart, pSize);
+        dctx->ddictIsCold = 0;
+    }
+
     return ip-istart;
 }
 
@@ -1075,7 +1133,7 @@ HINT_INLINE
 size_t ZSTD_execSequence(BYTE* op,
                          BYTE* const oend, seq_t sequence,
                          const BYTE** litPtr, const BYTE* const litLimit,
-                         const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+                         const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -1087,7 +1145,7 @@ size_t ZSTD_execSequence(BYTE* op,
     /* check */
     if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
     if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
-    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
+    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
 
     /* copy Literals */
     ZSTD_copy8(op, *litPtr);
@@ -1097,11 +1155,11 @@ size_t ZSTD_execSequence(BYTE* op,
     *litPtr = iLitEnd;   /* update for next sequence */
 
     /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
         /* offset beyond prefix -> go into extDict */
-        if (sequence.offset > (size_t)(oLitEnd - vBase))
+        if (sequence.offset > (size_t)(oLitEnd - virtualStart))
             return ERROR(corruption_detected);
-        match = dictEnd + (match - base);
+        match = dictEnd + (match - prefixStart);
         if (match + sequence.matchLength <= dictEnd) {
             memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
@@ -1111,7 +1169,7 @@ size_t ZSTD_execSequence(BYTE* op,
             memmove(oLitEnd, match, length1);
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
-            match = base;
+            match = prefixStart;
             if (op > oend_w || sequence.matchLength < MINMATCH) {
               U32 i;
               for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
@@ -1354,10 +1412,10 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* const litEnd = litPtr + dctx->litSize;
-    const BYTE* const base = (const BYTE*) (dctx->base);
-    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    DEBUGLOG(5, "ZSTD_decompressSequences");
+    DEBUGLOG(5, "ZSTD_decompressSequences_body");
 
     /* Regen sequences */
     if (nbSeq) {
@@ -1372,14 +1430,14 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
         for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
             nbSeq--;
             {   seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
-                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
                 DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
                 op += oneSeqSize;
         }   }
 
         /* check if reached exact end */
-        DEBUGLOG(5, "ZSTD_decompressSequences: after decode loop, remaining nbSeq : %i", nbSeq);
+        DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
         if (nbSeq) return ERROR(corruption_detected);
         /* save reps for next block */
         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
@@ -1498,8 +1556,8 @@ ZSTD_decompressSequencesLong_body(
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* const litEnd = litPtr + dctx->litSize;
-    const BYTE* const prefixStart = (const BYTE*) (dctx->base);
-    const BYTE* const dictStart = (const BYTE*) (dctx->vBase);
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
 
     /* Regen sequences */
@@ -1662,7 +1720,8 @@ static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
     /* isLongOffset must be true if there are long offsets.
      * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
      * We don't expect that to be the case in 64-bit mode.
-     * In block mode, window size is not known, so we have to be conservative. (note: but it could be evaluated from current-lowLimit)
+     * In block mode, window size is not known, so we have to be conservative.
+     * (note: but it could be evaluated from current-lowLimit)
      */
     ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)));
     DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
@@ -1701,8 +1760,8 @@ static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
 {
     if (dst != dctx->previousDstEnd) {   /* not contiguous */
         dctx->dictEnd = dctx->previousDstEnd;
-        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
-        dctx->base = dst;
+        dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+        dctx->prefixStart = dst;
         dctx->previousDstEnd = dst;
     }
 }
@@ -1729,10 +1788,10 @@ ZSTDLIB_API size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, siz
 }
 
 
-static size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+static size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE value, size_t length)
 {
     if (length > dstCapacity) return ERROR(dstSize_tooSmall);
-    memset(dst, byte, length);
+    memset(dst, value, length);
     return length;
 }
 
@@ -1749,7 +1808,7 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
 #endif
     if ( (srcSize >= ZSTD_skippableHeaderSize)
       && (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START ) {
-        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize);
+        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_FRAMEIDSIZE);
     } else {
         const BYTE* ip = (const BYTE*)src;
         const BYTE* const ipstart = ip;
@@ -1783,7 +1842,6 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
         if (zfh.checksumFlag) {   /* Final frame content checksum */
             if (remainingSize < 4) return ERROR(srcSize_wrong);
             ip += 4;
-            remainingSize -= 4;
         }
 
         return ip - ipstart;
@@ -1871,9 +1929,6 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
     return op-ostart;
 }
 
-static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
-static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
-
 static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                         void* dst, size_t dstCapacity,
                                   const void* src, size_t srcSize,
@@ -1881,6 +1936,9 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                   const ZSTD_DDict* ddict)
 {
     void* const dststart = dst;
+    int moreThan1Frame = 0;
+
+    DEBUGLOG(5, "ZSTD_decompressMultiFrame");
     assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
 
     if (ddict) {
@@ -1889,7 +1947,6 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
     }
 
     while (srcSize >= ZSTD_frameHeaderSize_prefix) {
-        U32 magicNumber;
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (ZSTD_isLegacy(src, srcSize)) {
@@ -1911,24 +1968,21 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
         }
 #endif
 
-        magicNumber = MEM_readLE32(src);
-        DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
-                    (U32)magicNumber, (U32)ZSTD_MAGICNUMBER);
-        if (magicNumber != ZSTD_MAGICNUMBER) {
+        {   U32 const magicNumber = MEM_readLE32(src);
+            DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
+                        (U32)magicNumber, (U32)ZSTD_MAGICNUMBER);
             if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
                 size_t skippableSize;
                 if (srcSize < ZSTD_skippableHeaderSize)
                     return ERROR(srcSize_wrong);
-                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize)
+                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_FRAMEIDSIZE)
                               + ZSTD_skippableHeaderSize;
                 if (srcSize < skippableSize) return ERROR(srcSize_wrong);
 
                 src = (const BYTE *)src + skippableSize;
                 srcSize -= skippableSize;
                 continue;
-            }
-            return ERROR(prefix_unknown);
-        }
+        }   }
 
         if (ddict) {
             /* we were called from ZSTD_decompress_usingDDict */
@@ -1942,11 +1996,25 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
 
         {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
                                                     &src, &srcSize);
+            if ( (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
+              && (moreThan1Frame==1) ) {
+                /* at least one frame successfully completed,
+                 * but following bytes are garbage :
+                 * it's more likely to be a srcSize error,
+                 * specifying more bytes than compressed size of frame(s).
+                 * This error message replaces ERROR(prefix_unknown),
+                 * which would be confusing, as the first header is actually correct.
+                 * Note that one could be unlucky, it might be a corruption error instead,
+                 * happening right at the place where we expect zstd magic bytes.
+                 * But this is _much_ less likely than a srcSize field error. */
+                return ERROR(srcSize_wrong);
+            }
             if (ZSTD_isError(res)) return res;
             /* no need to bound check, ZSTD_decompressFrame already has */
             dst = (BYTE*)dst + res;
             dstCapacity -= res;
         }
+        moreThan1Frame = 1;
     }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
 
     if (srcSize) return ERROR(srcSize_wrong); /* input not entirely consumed */
@@ -1980,6 +2048,7 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
     return regenSize;
 #else   /* stack mode */
     ZSTD_DCtx dctx;
+    ZSTD_initDCtx_internal(&dctx);
     return ZSTD_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
 #endif
 }
@@ -2031,7 +2100,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     case ZSTDds_getFrameHeaderSize :
         assert(src != NULL);
         if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
-            assert(srcSize >= ZSTD_frameIdSize);  /* to read skippable magic number */
+            assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
             if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
                 memcpy(dctx->headerBuffer, src, srcSize);
                 dctx->expected = ZSTD_skippableHeaderSize - srcSize;  /* remaining to load to get full skippable frame header */
@@ -2141,7 +2210,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         assert(src != NULL);
         assert(srcSize <= ZSTD_skippableHeaderSize);
         memcpy(dctx->headerBuffer + (ZSTD_skippableHeaderSize - srcSize), src, srcSize);   /* complete skippable header */
-        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_frameIdSize);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
         dctx->stage = ZSTDds_skipFrame;
         return 0;
 
@@ -2159,27 +2228,33 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
     dctx->dictEnd = dctx->previousDstEnd;
-    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
-    dctx->base = dict;
+    dctx->virtualStart = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+    dctx->prefixStart = dict;
     dctx->previousDstEnd = (const char*)dict + dictSize;
     return 0;
 }
 
-/* ZSTD_loadEntropy() :
- * dict : must point at beginning of a valid zstd dictionary
+/*! ZSTD_loadEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
  * @return : size of entropy tables read */
-static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
+static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy,
+                         const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
     if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
     dictPtr += 8;   /* skip header = magic + dictID */
 
-
-    {   size_t const hSize = HUF_readDTableX4_wksp(
-            entropy->hufTable, dictPtr, dictEnd - dictPtr,
-            entropy->workspace, sizeof(entropy->workspace));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, OFTable) == offsetof(ZSTD_entropyDTables_t, LLTable) + sizeof(entropy->LLTable));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, MLTable) == offsetof(ZSTD_entropyDTables_t, OFTable) + sizeof(entropy->OFTable));
+    ZSTD_STATIC_ASSERT(sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable) >= HUF_DECOMPRESS_WORKSPACE_SIZE);
+    {   void* const workspace = &entropy->LLTable;   /* use fse tables as temporary workspace; implies fse tables are grouped together */
+        size_t const workspaceSize = sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable);
+        size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize);
         if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
         dictPtr += hSize;
     }
@@ -2190,7 +2265,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
         if (offcodeMaxValue > MaxOff) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->OFTable,
+        ZSTD_buildFSETable( entropy->OFTable,
                             offcodeNCount, offcodeMaxValue,
                             OF_base, OF_bits,
                             offcodeLog);
@@ -2203,7 +2278,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (matchlengthMaxValue > MaxML) return ERROR(dictionary_corrupted);
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->MLTable,
+        ZSTD_buildFSETable( entropy->MLTable,
                             matchlengthNCount, matchlengthMaxValue,
                             ML_base, ML_bits,
                             matchlengthLog);
@@ -2216,7 +2291,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (litlengthMaxValue > MaxLL) return ERROR(dictionary_corrupted);
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->LLTable,
+        ZSTD_buildFSETable( entropy->LLTable,
                             litlengthNCount, litlengthMaxValue,
                             LL_base, LL_bits,
                             litlengthLog);
@@ -2242,7 +2317,7 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
         if (magic != ZSTD_MAGIC_DICTIONARY) {
             return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
     }   }
-    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 
     /* load entropy tables */
     {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
@@ -2256,7 +2331,6 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
     return ZSTD_refDictContent(dctx, dict, dictSize);
 }
 
-/* Note : this function cannot fail */
 size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 {
     assert(dctx != NULL);
@@ -2264,8 +2338,8 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
     dctx->stage = ZSTDds_getFrameHeaderSize;
     dctx->decodedSize = 0;
     dctx->previousDstEnd = NULL;
-    dctx->base = NULL;
-    dctx->vBase = NULL;
+    dctx->prefixStart = NULL;
+    dctx->virtualStart = NULL;
     dctx->dictEnd = NULL;
     dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
     dctx->litEntropy = dctx->fseEntropy = 0;
@@ -2302,42 +2376,53 @@ struct ZSTD_DDict_s {
 
 static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
 {
+    assert(ddict != NULL);
     return ddict->dictContent;
 }
 
 static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
 {
+    assert(ddict != NULL);
     return ddict->dictSize;
 }
 
-size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    CHECK_F( ZSTD_decompressBegin(dstDCtx) );
-    if (ddict) {   /* support begin on NULL */
-        dstDCtx->dictID = ddict->dictID;
-        dstDCtx->base = ddict->dictContent;
-        dstDCtx->vBase = ddict->dictContent;
-        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
-        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
+    DEBUGLOG(4, "ZSTD_decompressBegin_usingDDict");
+    assert(dctx != NULL);
+    if (ddict) {
+        dctx->ddictIsCold = (dctx->dictEnd != (const char*)ddict->dictContent + ddict->dictSize);
+        DEBUGLOG(4, "DDict is %s",
+                    dctx->ddictIsCold ? "~cold~" : "hot!");
+    }
+    CHECK_F( ZSTD_decompressBegin(dctx) );
+    if (ddict) {   /* NULL ddict is equivalent to no dictionary */
+        dctx->dictID = ddict->dictID;
+        dctx->prefixStart = ddict->dictContent;
+        dctx->virtualStart = ddict->dictContent;
+        dctx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+        dctx->previousDstEnd = dctx->dictEnd;
         if (ddict->entropyPresent) {
-            dstDCtx->litEntropy = 1;
-            dstDCtx->fseEntropy = 1;
-            dstDCtx->LLTptr = ddict->entropy.LLTable;
-            dstDCtx->MLTptr = ddict->entropy.MLTable;
-            dstDCtx->OFTptr = ddict->entropy.OFTable;
-            dstDCtx->HUFptr = ddict->entropy.hufTable;
-            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
-            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
-            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
+            dctx->litEntropy = 1;
+            dctx->fseEntropy = 1;
+            dctx->LLTptr = ddict->entropy.LLTable;
+            dctx->MLTptr = ddict->entropy.MLTable;
+            dctx->OFTptr = ddict->entropy.OFTable;
+            dctx->HUFptr = ddict->entropy.hufTable;
+            dctx->entropy.rep[0] = ddict->entropy.rep[0];
+            dctx->entropy.rep[1] = ddict->entropy.rep[1];
+            dctx->entropy.rep[2] = ddict->entropy.rep[2];
         } else {
-            dstDCtx->litEntropy = 0;
-            dstDCtx->fseEntropy = 0;
+            dctx->litEntropy = 0;
+            dctx->fseEntropy = 0;
         }
     }
     return 0;
 }
 
-static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e dictContentType)
+static size_t
+ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict,
+                         ZSTD_dictContentType_e dictContentType)
 {
     ddict->dictID = 0;
     ddict->entropyPresent = 0;
@@ -2355,10 +2440,12 @@ static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e
             return 0;   /* pure content mode */
         }
     }
-    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_frameIdSize);
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_FRAMEIDSIZE);
 
     /* load entropy tables */
-    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
+    CHECK_E( ZSTD_loadEntropy(&ddict->entropy,
+                              ddict->dictContent, ddict->dictSize),
+             dictionary_corrupted );
     ddict->entropyPresent = 1;
     return 0;
 }
@@ -2372,6 +2459,7 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
     if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
         ddict->dictBuffer = NULL;
         ddict->dictContent = dict;
+        if (!dict) dictSize = 0;
     } else {
         void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
         ddict->dictBuffer = internalBuffer;
@@ -2396,14 +2484,15 @@ ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
     if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
 
     {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
-        if (!ddict) return NULL;
+        if (ddict == NULL) return NULL;
         ddict->cMem = customMem;
-
-        if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, dictLoadMethod, dictContentType) )) {
-            ZSTD_freeDDict(ddict);
-            return NULL;
-        }
-
+        {   size_t const initResult = ZSTD_initDDict_internal(ddict,
+                                            dict, dictSize,
+                                            dictLoadMethod, dictContentType);
+            if (ZSTD_isError(initResult)) {
+                ZSTD_freeDDict(ddict);
+                return NULL;
+        }   }
         return ddict;
     }
 }
@@ -2430,23 +2519,25 @@ ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize
 
 
 const ZSTD_DDict* ZSTD_initStaticDDict(
-                                void* workspace, size_t workspaceSize,
+                                void* sBuffer, size_t sBufferSize,
                                 const void* dict, size_t dictSize,
                                 ZSTD_dictLoadMethod_e dictLoadMethod,
                                 ZSTD_dictContentType_e dictContentType)
 {
-    size_t const neededSpace =
-            sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
-    ZSTD_DDict* const ddict = (ZSTD_DDict*)workspace;
-    assert(workspace != NULL);
+    size_t const neededSpace = sizeof(ZSTD_DDict)
+                             + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)sBuffer;
+    assert(sBuffer != NULL);
     assert(dict != NULL);
-    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
-    if (workspaceSize < neededSpace) return NULL;
+    if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
+    if (sBufferSize < neededSpace) return NULL;
     if (dictLoadMethod == ZSTD_dlm_byCopy) {
         memcpy(ddict+1, dict, dictSize);  /* local copy */
         dict = ddict+1;
     }
-    if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, ZSTD_dlm_byRef, dictContentType) ))
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType) ))
         return NULL;
     return ddict;
 }
@@ -2484,7 +2575,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
 {
     if (dictSize < 8) return 0;
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
-    return MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+    return MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 }
 
 /*! ZSTD_getDictID_fromDDict() :
@@ -2560,12 +2651,15 @@ size_t ZSTD_freeDStream(ZSTD_DStream* zds)
 }
 
 
-/* *** Initialization *** */
+/* ***  Initialization  *** */
 
 size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
 size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
 
-size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
+                                   const void* dict, size_t dictSize,
+                                         ZSTD_dictLoadMethod_e dictLoadMethod,
+                                         ZSTD_dictContentType_e dictContentType)
 {
     if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
     ZSTD_freeDDict(dctx->ddictLocal);
@@ -2607,6 +2701,7 @@ size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t di
 {
     DEBUGLOG(4, "ZSTD_initDStream_usingDict");
     zds->streamStage = zdss_init;
+    zds->noForwardProgress = 0;
     CHECK_F( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) );
     return ZSTD_frameHeaderSize_prefix;
 }
@@ -2618,13 +2713,6 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
     return ZSTD_initDStream_usingDict(zds, NULL, 0);
 }
 
-size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
-{
-    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
-    dctx->ddict = ddict;
-    return 0;
-}
-
 /* ZSTD_initDStream_usingDDict() :
  * ddict will just be referenced, and must outlive decompression session
  * this function cannot fail */
@@ -2663,6 +2751,13 @@ size_t ZSTD_setDStreamParameter(ZSTD_DStream* dctx,
     return 0;
 }
 
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->ddict = ddict;
+    return 0;
+}
+
 size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
 {
     if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
@@ -2767,7 +2862,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     return hint;
             }   }
 #endif
-            {   size_t const hSize = ZSTD_getFrameHeader_internal(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+            {   size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
                 DEBUGLOG(5, "header size : %u", (U32)hSize);
                 if (ZSTD_isError(hSize)) {
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
@@ -2828,7 +2923,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             CHECK_F(ZSTD_decompressBegin_usingDDict(zds, zds->ddict));
 
             if ((MEM_readLE32(zds->headerBuffer) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
-                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_frameIdSize);
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
                 zds->stage = ZSTDds_skipFrame;
             } else {
                 CHECK_F(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
@@ -2947,8 +3042,18 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
     }   }
 
     /* result */
-    input->pos += (size_t)(ip-istart);
-    output->pos += (size_t)(op-ostart);
+    input->pos = (size_t)(ip - (const char*)(input->src));
+    output->pos = (size_t)(op - (char*)(output->dst));
+    if ((ip==istart) && (op==ostart)) {  /* no forward progress */
+        zds->noForwardProgress ++;
+        if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
+            if (op==oend) return ERROR(dstSize_tooSmall);
+            if (ip==iend) return ERROR(srcSize_wrong);
+            assert(0);
+        }
+    } else {
+        zds->noForwardProgress = 0;
+    }
     {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
         if (!nextSrcSizeHint) {   /* frame fully decoded */
             if (zds->outEnd == zds->outStart) {  /* output fully flushed */
diff --git a/lib/dictBuilder/cover.c b/lib/dictBuilder/cover.c
index b5a3957a9b96..6b4af69d29c5 100644
--- a/lib/dictBuilder/cover.c
+++ b/lib/dictBuilder/cover.c
@@ -29,6 +29,7 @@
 #include "mem.h" /* read */
 #include "pool.h"
 #include "threading.h"
+#include "cover.h"
 #include "zstd_internal.h" /* includes zstd.h */
 #ifndef ZDICT_STATIC_LINKING_ONLY
 #define ZDICT_STATIC_LINKING_ONLY
@@ -39,6 +40,7 @@
 *  Constants
 ***************************************/
 #define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
+#define DEFAULT_SPLITPOINT 1.0
 
 /*-*************************************
 *  Console display
@@ -184,7 +186,7 @@ static void COVER_map_remove(COVER_map_t *map, U32 key) {
 }
 
 /**
- * Destroyes a map that is inited with COVER_map_init().
+ * Destroys a map that is inited with COVER_map_init().
  */
 static void COVER_map_destroy(COVER_map_t *map) {
   if (map->data) {
@@ -203,6 +205,8 @@ typedef struct {
   size_t *offsets;
   const size_t *samplesSizes;
   size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
   U32 *suffix;
   size_t suffixSize;
   U32 *freqs;
@@ -220,9 +224,9 @@ static COVER_ctx_t *g_ctx = NULL;
 /**
  * Returns the sum of the sample sizes.
  */
-static size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
   size_t sum = 0;
-  size_t i;
+  unsigned i;
   for (i = 0; i < nbSamples; ++i) {
     sum += samplesSizes[i];
   }
@@ -377,14 +381,6 @@ static void COVER_group(COVER_ctx_t *ctx, const void *group,
   ctx->suffix[dmerId] = freq;
 }
 
-/**
- * A segment is a range in the source as well as the score of the segment.
- */
-typedef struct {
-  U32 begin;
-  U32 end;
-  U32 score;
-} COVER_segment_t;
 
 /**
  * Selects the best segment in an epoch.
@@ -494,6 +490,10 @@ static int COVER_checkParameters(ZDICT_cover_params_t parameters,
   if (parameters.d > parameters.k) {
     return 0;
   }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1){
+    return 0;
+  }
   return 1;
 }
 
@@ -531,9 +531,14 @@ static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
  */
 static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
                           const size_t *samplesSizes, unsigned nbSamples,
-                          unsigned d) {
+                          unsigned d, double splitPoint) {
   const BYTE *const samples = (const BYTE *)samplesBuffer;
   const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+  /* Split samples into testing and training sets */
+  const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+  const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+  const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+  const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
   /* Checks */
   if (totalSamplesSize < MAX(d, sizeof(U64)) ||
       totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
@@ -541,15 +546,29 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
                  (U32)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
     return 0;
   }
+  /* Check if there are at least 5 training samples */
+  if (nbTrainSamples < 5) {
+    DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid.", nbTrainSamples);
+    return 0;
+  }
+  /* Check if there's testing sample */
+  if (nbTestSamples < 1) {
+    DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.", nbTestSamples);
+    return 0;
+  }
   /* Zero the context */
   memset(ctx, 0, sizeof(*ctx));
-  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbSamples,
-               (U32)totalSamplesSize);
+  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+               (U32)trainingSamplesSize);
+  DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+               (U32)testSamplesSize);
   ctx->samples = samples;
   ctx->samplesSizes = samplesSizes;
   ctx->nbSamples = nbSamples;
+  ctx->nbTrainSamples = nbTrainSamples;
+  ctx->nbTestSamples = nbTestSamples;
   /* Partial suffix array */
-  ctx->suffixSize = totalSamplesSize - MAX(d, sizeof(U64)) + 1;
+  ctx->suffixSize = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
   ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
   /* Maps index to the dmerID */
   ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
@@ -563,7 +582,7 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
   ctx->freqs = NULL;
   ctx->d = d;
 
-  /* Fill offsets from the samlesSizes */
+  /* Fill offsets from the samplesSizes */
   {
     U32 i;
     ctx->offsets[0] = 0;
@@ -581,10 +600,17 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
     for (i = 0; i < ctx->suffixSize; ++i) {
       ctx->suffix[i] = i;
     }
-    /* qsort doesn't take an opaque pointer, so pass as a global */
+    /* qsort doesn't take an opaque pointer, so pass as a global.
+     * On OpenBSD qsort() is not guaranteed to be stable, their mergesort() is.
+     */
     g_ctx = ctx;
+#if defined(__OpenBSD__)
+    mergesort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+#else
     qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
           (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+#endif
   }
   DISPLAYLEVEL(2, "Computing frequencies\n");
   /* For each dmer group (group of positions with the same first d bytes):
@@ -613,7 +639,7 @@ static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
   /* Divide the data up into epochs of equal size.
    * We will select at least one segment from each epoch.
    */
-  const U32 epochs = (U32)(dictBufferCapacity / parameters.k);
+  const U32 epochs = MAX(1, (U32)(dictBufferCapacity / parameters.k / 4));
   const U32 epochSize = (U32)(ctx->suffixSize / epochs);
   size_t epoch;
   DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
@@ -658,7 +684,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   BYTE* const dict = (BYTE*)dictBuffer;
   COVER_ctx_t ctx;
   COVER_map_t activeDmers;
-
+  parameters.splitPoint = 1.0;
   /* Initialize global data */
   g_displayLevel = parameters.zParams.notificationLevel;
   /* Checks */
@@ -677,7 +703,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   }
   /* Initialize context and activeDmers */
   if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
-                      parameters.d)) {
+                      parameters.d, parameters.splitPoint)) {
     return ERROR(GENERIC);
   }
   if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
@@ -704,28 +730,65 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   }
 }
 
-/**
- * COVER_best_t is used for two purposes:
- * 1. Synchronizing threads.
- * 2. Saving the best parameters and dictionary.
- *
- * All of the methods except COVER_best_init() are thread safe if zstd is
- * compiled with multithreaded support.
- */
-typedef struct COVER_best_s {
-  ZSTD_pthread_mutex_t mutex;
-  ZSTD_pthread_cond_t cond;
-  size_t liveJobs;
-  void *dict;
-  size_t dictSize;
-  ZDICT_cover_params_t parameters;
-  size_t compressedSize;
-} COVER_best_t;
+
+
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                    const size_t *samplesSizes, const BYTE *samples,
+                                    size_t *offsets,
+                                    size_t nbTrainSamples, size_t nbSamples,
+                                    BYTE *const dict, size_t dictBufferCapacity) {
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Pointers */
+  ZSTD_CCtx *cctx;
+  ZSTD_CDict *cdict;
+  void *dst;
+  /* Local variables */
+  size_t dstCapacity;
+  size_t i;
+  /* Allocate dst with enough space to compress the maximum sized sample */
+  {
+    size_t maxSampleSize = 0;
+    i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+    for (; i < nbSamples; ++i) {
+      maxSampleSize = MAX(samplesSizes[i], maxSampleSize);
+    }
+    dstCapacity = ZSTD_compressBound(maxSampleSize);
+    dst = malloc(dstCapacity);
+  }
+  /* Create the cctx and cdict */
+  cctx = ZSTD_createCCtx();
+  cdict = ZSTD_createCDict(dict, dictBufferCapacity,
+                           parameters.zParams.compressionLevel);
+  if (!dst || !cctx || !cdict) {
+    goto _compressCleanup;
+  }
+  /* Compress each sample and sum their sizes (or error) */
+  totalCompressedSize = dictBufferCapacity;
+  i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+  for (; i < nbSamples; ++i) {
+    const size_t size = ZSTD_compress_usingCDict(
+        cctx, dst, dstCapacity, samples + offsets[i],
+        samplesSizes[i], cdict);
+    if (ZSTD_isError(size)) {
+      totalCompressedSize = ERROR(GENERIC);
+      goto _compressCleanup;
+    }
+    totalCompressedSize += size;
+  }
+_compressCleanup:
+  ZSTD_freeCCtx(cctx);
+  ZSTD_freeCDict(cdict);
+  if (dst) {
+    free(dst);
+  }
+  return totalCompressedSize;
+}
+
 
 /**
  * Initialize the `COVER_best_t`.
  */
-static void COVER_best_init(COVER_best_t *best) {
+void COVER_best_init(COVER_best_t *best) {
   if (best==NULL) return; /* compatible with init on NULL */
   (void)ZSTD_pthread_mutex_init(&best->mutex, NULL);
   (void)ZSTD_pthread_cond_init(&best->cond, NULL);
@@ -739,7 +802,7 @@ static void COVER_best_init(COVER_best_t *best) {
 /**
  * Wait until liveJobs == 0.
  */
-static void COVER_best_wait(COVER_best_t *best) {
+void COVER_best_wait(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -753,7 +816,7 @@ static void COVER_best_wait(COVER_best_t *best) {
 /**
  * Call COVER_best_wait() and then destroy the COVER_best_t.
  */
-static void COVER_best_destroy(COVER_best_t *best) {
+void COVER_best_destroy(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -769,7 +832,7 @@ static void COVER_best_destroy(COVER_best_t *best) {
  * Called when a thread is about to be launched.
  * Increments liveJobs.
  */
-static void COVER_best_start(COVER_best_t *best) {
+void COVER_best_start(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -783,7 +846,7 @@ static void COVER_best_start(COVER_best_t *best) {
  * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
  * If this dictionary is the best so far save it and its parameters.
  */
-static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
                               ZDICT_cover_params_t parameters, void *dict,
                               size_t dictSize) {
   if (!best) {
@@ -814,10 +877,10 @@ static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
       best->parameters = parameters;
       best->compressedSize = compressedSize;
     }
-    ZSTD_pthread_mutex_unlock(&best->mutex);
     if (liveJobs == 0) {
       ZSTD_pthread_cond_broadcast(&best->cond);
     }
+    ZSTD_pthread_mutex_unlock(&best->mutex);
   }
 }
 
@@ -832,7 +895,7 @@ typedef struct COVER_tryParameters_data_s {
 } COVER_tryParameters_data_t;
 
 /**
- * Tries a set of parameters and upates the COVER_best_t with the results.
+ * Tries a set of parameters and updates the COVER_best_t with the results.
  * This function is thread safe if zstd is compiled with multithreaded support.
  * It takes its parameters as an *OWNING* opaque pointer to support threading.
  */
@@ -863,7 +926,7 @@ static void COVER_tryParameters(void *opaque) {
                                               dictBufferCapacity, parameters);
     dictBufferCapacity = ZDICT_finalizeDictionary(
         dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples,
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples,
         parameters.zParams);
     if (ZDICT_isError(dictBufferCapacity)) {
       DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
@@ -871,49 +934,10 @@ static void COVER_tryParameters(void *opaque) {
     }
   }
   /* Check total compressed size */
-  {
-    /* Pointers */
-    ZSTD_CCtx *cctx;
-    ZSTD_CDict *cdict;
-    void *dst;
-    /* Local variables */
-    size_t dstCapacity;
-    size_t i;
-    /* Allocate dst with enough space to compress the maximum sized sample */
-    {
-      size_t maxSampleSize = 0;
-      for (i = 0; i < ctx->nbSamples; ++i) {
-        maxSampleSize = MAX(ctx->samplesSizes[i], maxSampleSize);
-      }
-      dstCapacity = ZSTD_compressBound(maxSampleSize);
-      dst = malloc(dstCapacity);
-    }
-    /* Create the cctx and cdict */
-    cctx = ZSTD_createCCtx();
-    cdict = ZSTD_createCDict(dict, dictBufferCapacity,
-                             parameters.zParams.compressionLevel);
-    if (!dst || !cctx || !cdict) {
-      goto _compressCleanup;
-    }
-    /* Compress each sample and sum their sizes (or error) */
-    totalCompressedSize = dictBufferCapacity;
-    for (i = 0; i < ctx->nbSamples; ++i) {
-      const size_t size = ZSTD_compress_usingCDict(
-          cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
-          ctx->samplesSizes[i], cdict);
-      if (ZSTD_isError(size)) {
-        totalCompressedSize = ERROR(GENERIC);
-        goto _compressCleanup;
-      }
-      totalCompressedSize += size;
-    }
-  _compressCleanup:
-    ZSTD_freeCCtx(cctx);
-    ZSTD_freeCDict(cdict);
-    if (dst) {
-      free(dst);
-    }
-  }
+  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
+                                                       ctx->samples, ctx->offsets,
+                                                       ctx->nbTrainSamples, ctx->nbSamples,
+                                                       dict, dictBufferCapacity);
 
 _cleanup:
   COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
@@ -934,6 +958,8 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
     ZDICT_cover_params_t *parameters) {
   /* constants */
   const unsigned nbThreads = parameters->nbThreads;
+  const double splitPoint =
+      parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
   const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
   const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
   const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
@@ -951,6 +977,10 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
   POOL_ctx *pool = NULL;
 
   /* Checks */
+  if (splitPoint <= 0 || splitPoint > 1) {
+    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
+    return ERROR(GENERIC);
+  }
   if (kMinK < kMaxD || kMaxK < kMinK) {
     LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
     return ERROR(GENERIC);
@@ -981,7 +1011,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
     /* Initialize the context for this value of d */
     COVER_ctx_t ctx;
     LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
-    if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d)) {
+    if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint)) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
       COVER_best_destroy(&best);
       POOL_free(pool);
@@ -1006,6 +1036,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
       data->parameters = *parameters;
       data->parameters.k = k;
       data->parameters.d = d;
+      data->parameters.splitPoint = splitPoint;
       data->parameters.steps = kSteps;
       data->parameters.zParams.notificationLevel = g_displayLevel;
       /* Check the parameters */
diff --git a/lib/dictBuilder/cover.h b/lib/dictBuilder/cover.h
new file mode 100644
index 000000000000..82e2e1cea43c
--- /dev/null
+++ b/lib/dictBuilder/cover.h
@@ -0,0 +1,83 @@
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+/**
+ * COVER_best_t is used for two purposes:
+ * 1. Synchronizing threads.
+ * 2. Saving the best parameters and dictionary.
+ *
+ * All of the methods except COVER_best_init() are thread safe if zstd is
+ * compiled with multithreaded support.
+ */
+typedef struct COVER_best_s {
+  ZSTD_pthread_mutex_t mutex;
+  ZSTD_pthread_cond_t cond;
+  size_t liveJobs;
+  void *dict;
+  size_t dictSize;
+  ZDICT_cover_params_t parameters;
+  size_t compressedSize;
+} COVER_best_t;
+
+/**
+ * A segment is a range in the source as well as the score of the segment.
+ */
+typedef struct {
+  U32 begin;
+  U32 end;
+  U32 score;
+} COVER_segment_t;
+
+/**
+ *  Checks total compressed size of a dictionary
+ */
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                      const size_t *samplesSizes, const BYTE *samples,
+                                      size_t *offsets,
+                                      size_t nbTrainSamples, size_t nbSamples,
+                                      BYTE *const dict, size_t dictBufferCapacity);
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) ;
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+void COVER_best_init(COVER_best_t *best);
+
+/**
+ * Wait until liveJobs == 0.
+ */
+void COVER_best_wait(COVER_best_t *best);
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+void COVER_best_destroy(COVER_best_t *best);
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+void COVER_best_start(COVER_best_t *best);
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+                       ZDICT_cover_params_t parameters, void *dict,
+                       size_t dictSize);
diff --git a/lib/dictBuilder/divsufsort.c b/lib/dictBuilder/divsufsort.c
index 60cceb088321..ead9220442b2 100644
--- a/lib/dictBuilder/divsufsort.c
+++ b/lib/dictBuilder/divsufsort.c
@@ -1637,7 +1637,7 @@ construct_SA(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else {
           assert(((s == 0) && (T[s] == c1)) || (s < 0));
@@ -1701,7 +1701,7 @@ construct_BWT(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else if(s != 0) {
           *j = ~s;
@@ -1785,7 +1785,7 @@ construct_BWT_indexes(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else if(s != 0) {
           *j = ~s;
diff --git a/lib/dictBuilder/fastcover.c b/lib/dictBuilder/fastcover.c
new file mode 100644
index 000000000000..dfee45743413
--- /dev/null
+++ b/lib/dictBuilder/fastcover.c
@@ -0,0 +1,728 @@
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "cover.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
+#define FASTCOVER_MAX_F 31
+#define FASTCOVER_MAX_ACCEL 10
+#define DEFAULT_SPLITPOINT 0.75
+#define DEFAULT_F 20
+#define DEFAULT_ACCEL 1
+
+
+/*-*************************************
+*  Console display
+***************************************/
+static int g_displayLevel = 2;
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+
+
+/*-*************************************
+* Hash Functions
+***************************************/
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+
+/**
+ * Hash the d-byte value pointed to by p and mod 2^f
+ */
+static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
+  if (d == 6) {
+    return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
+  }
+  return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
+}
+
+
+/*-*************************************
+* Acceleration
+***************************************/
+typedef struct {
+  unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
+  unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
+} FASTCOVER_accel_t;
+
+
+static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
+  { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
+  { 100, 0 },   /* accel = 1 */
+  { 50, 1 },   /* accel = 2 */
+  { 34, 2 },   /* accel = 3 */
+  { 25, 3 },   /* accel = 4 */
+  { 20, 4 },   /* accel = 5 */
+  { 17, 5 },   /* accel = 6 */
+  { 14, 6 },   /* accel = 7 */
+  { 13, 7 },   /* accel = 8 */
+  { 11, 8 },   /* accel = 9 */
+  { 10, 9 },   /* accel = 10 */
+};
+
+
+/*-*************************************
+* Context
+***************************************/
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
+  size_t nbDmers;
+  U32 *freqs;
+  unsigned d;
+  unsigned f;
+  FASTCOVER_accel_t accelParams;
+} FASTCOVER_ctx_t;
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of all dmers with hash value d.
+ * Let S_i be hash value of the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer with hash value d is in the dictionay we set F(d) = 0.
+ */
+static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
+                                              U32 *freqs, U32 begin, U32 end,
+                                              ZDICT_cover_params_t parameters,
+                                              U16* segmentFreqs) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 f = ctx->f;
+  const U32 dmersInK = k - d + 1;
+
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+
+  /* Reset the activeDmers in the segment */
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* Get hash value of current dmer */
+    const size_t index = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
+
+    /* Add frequency of this index to score if this is the first occurence of index in active segment */
+    if (segmentFreqs[index] == 0) {
+      activeSegment.score += freqs[index];
+    }
+    /* Increment end of segment and segmentFreqs*/
+    activeSegment.end += 1;
+    segmentFreqs[index] += 1;
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      /* Get hash value of the dmer to be eliminated from active segment */
+      const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+      segmentFreqs[delIndex] -= 1;
+      /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
+      if (segmentFreqs[delIndex] == 0) {
+        activeSegment.score -= freqs[delIndex];
+      }
+      /* Increment start of segment */
+      activeSegment.begin += 1;
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+
+  /* Zero out rest of segmentFreqs array */
+  while (activeSegment.begin < end) {
+    const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+    segmentFreqs[delIndex] -= 1;
+    activeSegment.begin += 1;
+  }
+
+  {
+    /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
+      freqs[i] = 0;
+    }
+  }
+
+  return bestSegment;
+}
+
+
+static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
+                                     size_t maxDictSize, unsigned f,
+                                     unsigned accel) {
+  /* k, d, and f are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* d has to be 6 or 8 */
+  if (parameters.d != 6 && parameters.d != 8) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  /* 0 < f <= FASTCOVER_MAX_F*/
+  if (f > FASTCOVER_MAX_F || f == 0) {
+    return 0;
+  }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
+    return 0;
+  }
+  /* 0 < accel <= 10 */
+  if (accel > 10 || accel == 0) {
+    return 0;
+  }
+  return 1;
+}
+
+
+/**
+ * Clean up a context initialized with `FASTCOVER_ctx_init()`.
+ */
+static void
+FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
+{
+    if (!ctx) return;
+
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+}
+
+
+/**
+ * Calculate for frequency of hash value of each dmer in ctx->samples
+ */
+static void
+FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
+{
+    const unsigned f = ctx->f;
+    const unsigned d = ctx->d;
+    const unsigned skip = ctx->accelParams.skip;
+    const unsigned readLength = MAX(d, 8);
+    size_t i;
+    assert(ctx->nbTrainSamples >= 5);
+    assert(ctx->nbTrainSamples <= ctx->nbSamples);
+    for (i = 0; i < ctx->nbTrainSamples; i++) {
+        size_t start = ctx->offsets[i];  /* start of current dmer */
+        size_t const currSampleEnd = ctx->offsets[i+1];
+        while (start + readLength <= currSampleEnd) {
+            const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
+            freqs[dmerIndex]++;
+            start = start + skip + 1;
+        }
+    }
+}
+
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can used multiple
+ * times.
+ * Returns 1 on success or zero on error.
+ * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
+ */
+static int
+FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
+                   const void* samplesBuffer,
+                   const size_t* samplesSizes, unsigned nbSamples,
+                   unsigned d, double splitPoint, unsigned f,
+                   FASTCOVER_accel_t accelParams)
+{
+    const BYTE* const samples = (const BYTE*)samplesBuffer;
+    const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+    /* Split samples into testing and training sets */
+    const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+    const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+    const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+    const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
+
+    /* Checks */
+    if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+        totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
+        DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                    (U32)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
+        return 0;
+    }
+
+    /* Check if there are at least 5 training samples */
+    if (nbTrainSamples < 5) {
+        DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
+        return 0;
+    }
+
+    /* Check if there's testing sample */
+    if (nbTestSamples < 1) {
+        DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
+        return 0;
+    }
+
+    /* Zero the context */
+    memset(ctx, 0, sizeof(*ctx));
+    DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+                    (U32)trainingSamplesSize);
+    DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+                    (U32)testSamplesSize);
+
+    ctx->samples = samples;
+    ctx->samplesSizes = samplesSizes;
+    ctx->nbSamples = nbSamples;
+    ctx->nbTrainSamples = nbTrainSamples;
+    ctx->nbTestSamples = nbTestSamples;
+    ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
+    ctx->d = d;
+    ctx->f = f;
+    ctx->accelParams = accelParams;
+
+    /* The offsets of each file */
+    ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
+    if (ctx->offsets == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return 0;
+    }
+
+    /* Fill offsets from the samplesSizes */
+    {   U32 i;
+        ctx->offsets[0] = 0;
+        assert(nbSamples >= 5);
+        for (i = 1; i <= nbSamples; ++i) {
+            ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+        }
+    }
+
+    /* Initialize frequency array of size 2^f */
+    ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
+    if (ctx->freqs == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return 0;
+    }
+
+    DISPLAYLEVEL(2, "Computing frequencies\n");
+    FASTCOVER_computeFrequency(ctx->freqs, ctx);
+
+    return 1;
+}
+
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t
+FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
+                          U32* freqs,
+                          void* dictBuffer, size_t dictBufferCapacity,
+                          ZDICT_cover_params_t parameters,
+                          U16* segmentFreqs)
+{
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data up into epochs of equal size.
+   * We will select at least one segment from each epoch.
+   */
+  const U32 epochs = MAX(1, (U32)(dictBufferCapacity / parameters.k));
+  const U32 epochSize = (U32)(ctx->nbDmers / epochs);
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
+               epochSize);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs) {
+    const U32 epochBegin = (U32)(epoch * epochSize);
+    const U32 epochEnd = epochBegin + epochSize;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = FASTCOVER_selectSegment(
+        ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
+
+    /* If the segment covers no dmers, then we are out of content */
+    if (segment.score == 0) {
+      break;
+    }
+
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (U32)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+
+/**
+ * Parameters for FASTCOVER_tryParameters().
+ */
+typedef struct FASTCOVER_tryParameters_data_s {
+    const FASTCOVER_ctx_t* ctx;
+    COVER_best_t* best;
+    size_t dictBufferCapacity;
+    ZDICT_cover_params_t parameters;
+} FASTCOVER_tryParameters_data_t;
+
+
+/**
+ * Tries a set of parameters and updates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void FASTCOVER_tryParameters(void *opaque)
+{
+  /* Save parameters as local variables */
+  FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
+  const FASTCOVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Initialize array to keep track of frequency of dmer within activeSegment */
+  U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
+  /* Allocate space for hash table, dict, and freqs */
+  BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
+  if (!segmentFreqs || !dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
+  /* Build the dictionary */
+  { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
+                                                  parameters, segmentFreqs);
+    const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
+    dictBufferCapacity = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, nbFinalizeSamples, parameters.zParams);
+    if (ZDICT_isError(dictBufferCapacity)) {
+      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+      goto _cleanup;
+    }
+  }
+  /* Check total compressed size */
+  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
+                                                       ctx->samples, ctx->offsets,
+                                                       ctx->nbTrainSamples, ctx->nbSamples,
+                                                       dict, dictBufferCapacity);
+_cleanup:
+  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
+                    dictBufferCapacity);
+  free(data);
+  free(segmentFreqs);
+  free(dict);
+  free(freqs);
+}
+
+
+static void
+FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
+                               ZDICT_cover_params_t* coverParams)
+{
+    coverParams->k = fastCoverParams.k;
+    coverParams->d = fastCoverParams.d;
+    coverParams->steps = fastCoverParams.steps;
+    coverParams->nbThreads = fastCoverParams.nbThreads;
+    coverParams->splitPoint = fastCoverParams.splitPoint;
+    coverParams->zParams = fastCoverParams.zParams;
+}
+
+
+static void
+FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
+                                   ZDICT_fastCover_params_t* fastCoverParams,
+                                   unsigned f, unsigned accel)
+{
+    fastCoverParams->k = coverParams.k;
+    fastCoverParams->d = coverParams.d;
+    fastCoverParams->steps = coverParams.steps;
+    fastCoverParams->nbThreads = coverParams.nbThreads;
+    fastCoverParams->splitPoint = coverParams.splitPoint;
+    fastCoverParams->f = f;
+    fastCoverParams->accel = accel;
+    fastCoverParams->zParams = coverParams.zParams;
+}
+
+
+ZDICTLIB_API size_t
+ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
+                                const void* samplesBuffer,
+                                const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_fastCover_params_t parameters)
+{
+    BYTE* const dict = (BYTE*)dictBuffer;
+    FASTCOVER_ctx_t ctx;
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* Initialize global data */
+    g_displayLevel = parameters.zParams.notificationLevel;
+    /* Assign splitPoint and f if not provided */
+    parameters.splitPoint = 1.0;
+    parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
+    parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
+    /* Convert to cover parameter */
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(parameters, &coverParams);
+    /* Checks */
+    if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
+                                   parameters.accel)) {
+      DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+      return ERROR(GENERIC);
+    }
+    if (nbSamples == 0) {
+      DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
+      return ERROR(GENERIC);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    /* Assign corresponding FASTCOVER_accel_t to accelParams*/
+    accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
+    /* Initialize context */
+    if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                            coverParams.d, parameters.splitPoint, parameters.f,
+                            accelParams)) {
+      DISPLAYLEVEL(1, "Failed to initialize context\n");
+      return ERROR(GENERIC);
+    }
+    /* Build the dictionary */
+    DISPLAYLEVEL(2, "Building dictionary\n");
+    {
+      /* Initialize array to keep track of frequency of dmer within activeSegment */
+      U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
+      const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
+                                                dictBufferCapacity, coverParams, segmentFreqs);
+      const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
+      const size_t dictionarySize = ZDICT_finalizeDictionary(
+          dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+          samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
+      if (!ZSTD_isError(dictionarySize)) {
+          DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                      (U32)dictionarySize);
+      }
+      FASTCOVER_ctx_destroy(&ctx);
+      free(segmentFreqs);
+      return dictionarySize;
+    }
+}
+
+
+ZDICTLIB_API size_t
+ZDICT_optimizeTrainFromBuffer_fastCover(
+                    void* dictBuffer, size_t dictBufferCapacity,
+                    const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters)
+{
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* constants */
+    const unsigned nbThreads = parameters->nbThreads;
+    const double splitPoint =
+        parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
+    const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+    const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+    const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+    const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+    const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+    const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+    const unsigned kIterations =
+        (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+    const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
+    const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
+    /* Local variables */
+    const int displayLevel = parameters->zParams.notificationLevel;
+    unsigned iteration = 1;
+    unsigned d;
+    unsigned k;
+    COVER_best_t best;
+    POOL_ctx *pool = NULL;
+    /* Checks */
+    if (splitPoint <= 0 || splitPoint > 1) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
+      return ERROR(GENERIC);
+    }
+    if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
+      return ERROR(GENERIC);
+    }
+    if (kMinK < kMaxD || kMaxK < kMinK) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
+      return ERROR(GENERIC);
+    }
+    if (nbSamples == 0) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
+      return ERROR(GENERIC);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    if (nbThreads > 1) {
+      pool = POOL_create(nbThreads, 1);
+      if (!pool) {
+        return ERROR(memory_allocation);
+      }
+    }
+    /* Initialization */
+    COVER_best_init(&best);
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(*parameters, &coverParams);
+    accelParams = FASTCOVER_defaultAccelParameters[accel];
+    /* Turn down global display level to clean up display at level 2 and below */
+    g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+    /* Loop through d first because each new value needs a new context */
+    LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                      kIterations);
+    for (d = kMinD; d <= kMaxD; d += 2) {
+      /* Initialize the context for this value of d */
+      FASTCOVER_ctx_t ctx;
+      LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+      if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams)) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return ERROR(GENERIC);
+      }
+      /* Loop through k reusing the same context */
+      for (k = kMinK; k <= kMaxK; k += kStepSize) {
+        /* Prepare the arguments */
+        FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
+            sizeof(FASTCOVER_tryParameters_data_t));
+        LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+        if (!data) {
+          LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+          COVER_best_destroy(&best);
+          FASTCOVER_ctx_destroy(&ctx);
+          POOL_free(pool);
+          return ERROR(GENERIC);
+        }
+        data->ctx = &ctx;
+        data->best = &best;
+        data->dictBufferCapacity = dictBufferCapacity;
+        data->parameters = coverParams;
+        data->parameters.k = k;
+        data->parameters.d = d;
+        data->parameters.splitPoint = splitPoint;
+        data->parameters.steps = kSteps;
+        data->parameters.zParams.notificationLevel = g_displayLevel;
+        /* Check the parameters */
+        if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
+                                       data->ctx->f, accel)) {
+          DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+          free(data);
+          continue;
+        }
+        /* Call the function and pass ownership of data to it */
+        COVER_best_start(&best);
+        if (pool) {
+          POOL_add(pool, &FASTCOVER_tryParameters, data);
+        } else {
+          FASTCOVER_tryParameters(data);
+        }
+        /* Print status */
+        LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                           (U32)((iteration * 100) / kIterations));
+        ++iteration;
+      }
+      COVER_best_wait(&best);
+      FASTCOVER_ctx_destroy(&ctx);
+    }
+    LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+    /* Fill the output buffer and parameters with output of the best parameters */
+    {
+      const size_t dictSize = best.dictSize;
+      if (ZSTD_isError(best.compressedSize)) {
+        const size_t compressedSize = best.compressedSize;
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return compressedSize;
+      }
+      FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
+      memcpy(dictBuffer, best.dict, dictSize);
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return dictSize;
+    }
+
+}
diff --git a/lib/dictBuilder/zdict.c b/lib/dictBuilder/zdict.c
index 7d24e4991812..2964b69fff04 100644
--- a/lib/dictBuilder/zdict.c
+++ b/lib/dictBuilder/zdict.c
@@ -293,7 +293,7 @@ static dictItem ZDICT_analyzePos(
             refinedEnd = refinedStart + selectedCount;
         }
 
-        /* evaluate gain based on new ref */
+        /* evaluate gain based on new dict */
         start = refinedStart;
         pos = suffix[refinedStart];
         end = start;
@@ -341,7 +341,7 @@ static dictItem ZDICT_analyzePos(
         for (i=MINMATCHLENGTH; i<=(int)maxLength; i++)
             savings[i] = savings[i-1] + (lengthList[i] * (i-3));
 
-        DISPLAYLEVEL(4, "Selected ref at position %u, of length %u : saves %u (ratio: %.2f)  \n",
+        DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f)  \n",
                      (U32)pos, (U32)maxLength, savings[maxLength], (double)savings[maxLength] / maxLength);
 
         solution.pos = (U32)pos;
@@ -581,7 +581,7 @@ static void ZDICT_fillNoise(void* buffer, size_t length)
 
 typedef struct
 {
-    ZSTD_CCtx* ref;    /* contains reference to dictionary */
+    ZSTD_CDict* dict;    /* dictionary */
     ZSTD_CCtx* zc;     /* working context */
     void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
 } EStats_ress_t;
@@ -597,8 +597,9 @@ static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
     size_t cSize;
 
     if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
-    {   size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
-        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
+    {   size_t const errorCode = ZSTD_compressBegin_usingCDict(esr.zc, esr.dict);
+        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_compressBegin_usingCDict failed \n"); return; }
+
     }
     cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
     if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
@@ -697,7 +698,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     short litLengthNCount[MaxLL+1];
     U32 repOffset[MAXREPOFFSET];
     offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
-    EStats_ress_t esr;
+    EStats_ress_t esr = { NULL, NULL, NULL };
     ZSTD_parameters params;
     U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
     size_t pos = 0, errorCode;
@@ -708,14 +709,6 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
 
     /* init */
     DEBUGLOG(4, "ZDICT_analyzeEntropy");
-    esr.ref = ZSTD_createCCtx();
-    esr.zc = ZSTD_createCCtx();
-    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
-    if (!esr.ref || !esr.zc || !esr.workPlace) {
-        eSize = ERROR(memory_allocation);
-        DISPLAYLEVEL(1, "Not enough memory \n");
-        goto _cleanup;
-    }
     if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
     for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
     for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
@@ -724,14 +717,17 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     memset(repOffset, 0, sizeof(repOffset));
     repOffset[1] = repOffset[4] = repOffset[8] = 1;
     memset(bestRepOffset, 0, sizeof(bestRepOffset));
-    if (compressionLevel<=0) compressionLevel = g_compressionLevel_default;
+    if (compressionLevel==0) compressionLevel = g_compressionLevel_default;
     params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
-    {   size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
-        if (ZSTD_isError(beginResult)) {
-            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced() failed : %s \n", ZSTD_getErrorName(beginResult));
-            eSize = ERROR(GENERIC);
-            goto _cleanup;
-    }   }
+
+    esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
+    esr.zc = ZSTD_createCCtx();
+    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
+    if (!esr.dict || !esr.zc || !esr.workPlace) {
+        eSize = ERROR(memory_allocation);
+        DISPLAYLEVEL(1, "Not enough memory \n");
+        goto _cleanup;
+    }
 
     /* collect stats on all samples */
     for (u=0; u<nbFiles; u++) {
@@ -856,7 +852,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     eSize += 12;
 
 _cleanup:
-    ZSTD_freeCCtx(esr.ref);
+    ZSTD_freeCDict(esr.dict);
     ZSTD_freeCCtx(esr.zc);
     free(esr.workPlace);
 
@@ -867,13 +863,13 @@ _cleanup:
 
 size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
                           const void* customDictContent, size_t dictContentSize,
-                          const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                          ZDICT_params_t params)
+                          const void* samplesBuffer, const size_t* samplesSizes,
+                          unsigned nbSamples, ZDICT_params_t params)
 {
     size_t hSize;
 #define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
     BYTE header[HBUFFSIZE];
-    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    int const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
 
     /* check conditions */
@@ -914,11 +910,12 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
 }
 
 
-size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                                 ZDICT_params_t params)
+static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
+        void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+        ZDICT_params_t params)
 {
-    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    int const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
     size_t hSize = 8;
 
@@ -947,7 +944,11 @@ size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictCo
     return MIN(dictBufferCapacity, hSize+dictContentSize);
 }
 
-
+/* Hidden declaration for dbio.c */
+size_t ZDICT_trainFromBuffer_unsafe_legacy(
+                            void* dictBuffer, size_t maxDictSize,
+                            const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                            ZDICT_legacy_params_t params);
 /*! ZDICT_trainFromBuffer_unsafe_legacy() :
 *   Warning : `samplesBuffer` must be followed by noisy guard band.
 *   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
@@ -991,8 +992,10 @@ size_t ZDICT_trainFromBuffer_unsafe_legacy(
             U32 const pos = dictList[u].pos;
             U32 const length = dictList[u].length;
             U32 const printedLength = MIN(40, length);
-            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize))
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize)) {
+                free(dictList);
                 return ERROR(GENERIC);   /* should never happen */
+            }
             DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
                          u, length, pos, dictList[u].savings);
             ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
@@ -1082,17 +1085,17 @@ size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
 size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
 {
-    ZDICT_cover_params_t params;
+    ZDICT_fastCover_params_t params;
     DEBUGLOG(3, "ZDICT_trainFromBuffer");
     memset(&params, 0, sizeof(params));
     params.d = 8;
     params.steps = 4;
     /* Default to level 6 since no compression level information is available */
-    params.zParams.compressionLevel = 6;
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-    params.zParams.notificationLevel = ZSTD_DEBUG;
+    params.zParams.compressionLevel = 3;
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
+    params.zParams.notificationLevel = DEBUGLEVEL;
 #endif
-    return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
+    return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity,
                                                samplesBuffer, samplesSizes, nbSamples,
                                                &params);
 }
diff --git a/lib/dictBuilder/zdict.h b/lib/dictBuilder/zdict.h
index ad459c2d7d53..d57d59f01e72 100644
--- a/lib/dictBuilder/zdict.h
+++ b/lib/dictBuilder/zdict.h
@@ -39,7 +39,8 @@ extern "C" {
 
 /*! ZDICT_trainFromBuffer():
  *  Train a dictionary from an array of samples.
- *  Redirect towards ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
+ *  Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4,
+ *  f=20, and accel=1.
  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
  *  The resulting dictionary will be saved into `dictBuffer`.
@@ -52,7 +53,8 @@ extern "C" {
  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
  */
 ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                                    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+                                    const void* samplesBuffer,
+                                    const size_t* samplesSizes, unsigned nbSamples);
 
 
 /*======   Helper functions   ======*/
@@ -84,11 +86,22 @@ typedef struct {
 typedef struct {
     unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
     unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
-    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
     ZDICT_params_t zParams;
 } ZDICT_cover_params_t;
 
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned f;                  /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */
+    unsigned accel;              /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
+    ZDICT_params_t zParams;
+} ZDICT_fastCover_params_t;
 
 /*! ZDICT_trainFromBuffer_cover():
  *  Train a dictionary from an array of samples using the COVER algorithm.
@@ -115,9 +128,9 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
  * dictionary constructed with those parameters is stored in `dictBuffer`.
  *
  * All of the parameters d, k, steps are optional.
- * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
  * if steps is zero it defaults to its default value.
- * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
  *
  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
  *           or an error code, which can be tested with ZDICT_isError().
@@ -129,6 +142,48 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
           ZDICT_cover_params_t* parameters);
 
+/*! ZDICT_trainFromBuffer_fastCover():
+ *  Train a dictionary from an array of samples using a modified version of COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  d and k are required.
+ *  All other parameters are optional, will use default values if not provided
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer_fastCover() requires about 1 bytes of memory for each input byte and additionally another 6 * 2^f bytes of memory .
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
+                    size_t dictBufferCapacity, const void *samplesBuffer,
+                    const size_t *samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t parameters);
+
+/*! ZDICT_optimizeTrainFromBuffer_fastCover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations (specifically, k and d combinations)
+ * and picks the best parameters. `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ * All of the parameters d, k, steps, f, and accel are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
+ * If f is zero, default value of 20 is used.
+ * If accel is zero, default value of 1 is used.
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *           or an error code, which can be tested with ZDICT_isError().
+ *           On success `*parameters` contains the parameters selected.
+ * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 1 byte of memory for each input byte and additionally another 6 * 2^f bytes of memory for each thread.
+ */
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
+                    size_t dictBufferCapacity, const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters);
+
 /*! ZDICT_finalizeDictionary():
  * Given a custom content as a basis for dictionary, and a set of samples,
  * finalize dictionary by adding headers and statistics.
diff --git a/lib/legacy/zstd_v01.c b/lib/legacy/zstd_v01.c
index ae1cb2ce5aa0..c007e7ceb518 100644
--- a/lib/legacy/zstd_v01.c
+++ b/lib/legacy/zstd_v01.c
@@ -668,11 +668,17 @@ static size_t FSE_initDStream(FSE_DStream_t* bitD, const void* srcBuffer, size_t
         switch(srcSize)
         {
             case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
             case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
             case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
             case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
             case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
             case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
             default:;
         }
         contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
@@ -1458,7 +1464,7 @@ unsigned ZSTDv01_isError(size_t code) { return ERR_isError(code); }
 *   Decompression code
 **************************************************************/
 
-size_t ZSTDv01_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+static size_t ZSTDv01_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
 {
     const BYTE* const in = (const BYTE* const)src;
     BYTE headerFlags;
@@ -1511,7 +1517,7 @@ static size_t ZSTD_decompressLiterals(void* ctx,
 }
 
 
-size_t ZSTDv01_decodeLiteralsBlock(void* ctx,
+static size_t ZSTDv01_decodeLiteralsBlock(void* ctx,
                                 void* dst, size_t maxDstSize,
                           const BYTE** litStart, size_t* litSize,
                           const void* src, size_t srcSize)
@@ -1563,7 +1569,7 @@ size_t ZSTDv01_decodeLiteralsBlock(void* ctx,
 }
 
 
-size_t ZSTDv01_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+static size_t ZSTDv01_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
                          FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
                          const void* src, size_t srcSize)
 {
diff --git a/lib/legacy/zstd_v02.c b/lib/legacy/zstd_v02.c
index 8bc0eceeda8f..c09ef8cff23e 100644
--- a/lib/legacy/zstd_v02.c
+++ b/lib/legacy/zstd_v02.c
@@ -399,11 +399,17 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         switch(srcSize)
         {
             case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
             case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
             case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
             case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
             case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
             case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
             default:;
         }
         contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
diff --git a/lib/legacy/zstd_v03.c b/lib/legacy/zstd_v03.c
index 54445af577ed..0c4cdf6888a7 100644
--- a/lib/legacy/zstd_v03.c
+++ b/lib/legacy/zstd_v03.c
@@ -402,11 +402,17 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         switch(srcSize)
         {
             case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
             case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
             case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
             case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
             case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
             case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
             default:;
         }
         contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
diff --git a/lib/legacy/zstd_v04.c b/lib/legacy/zstd_v04.c
index fb6d1d4b15c8..e852bb91116a 100644
--- a/lib/legacy/zstd_v04.c
+++ b/lib/legacy/zstd_v04.c
@@ -9,14 +9,19 @@
  */
 
 
-/*- Dependencies -*/
+ /******************************************
+ *  Includes
+ ******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
 #include "zstd_v04.h"
 #include "error_private.h"
 
 
 /* ******************************************************************
-   mem.h
-****************************************************************** */
+ *   mem.h
+ *******************************************************************/
 #ifndef MEM_H_MODULE
 #define MEM_H_MODULE
 
@@ -24,12 +29,6 @@
 extern "C" {
 #endif
 
-/******************************************
-*  Includes
-******************************************/
-#include <stddef.h>    /* size_t, ptrdiff_t */
-#include <string.h>    /* memcpy */
-
 
 /******************************************
 *  Compiler-specific
@@ -75,38 +74,9 @@ extern "C" {
 /*-*************************************
 *  Debug
 ***************************************/
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-#  include <assert.h>
-#else
-#  ifndef assert
-#    define assert(condition) ((void)0)
-#  endif
-#endif
-
-#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
-
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
-#  include <stdio.h>
-extern int g_debuglog_enable;
-/* recommended values for ZSTD_DEBUG display levels :
- * 1 : no display, enables assert() only
- * 2 : reserved for currently active debug path
- * 3 : events once per object lifetime (CCtx, CDict, etc.)
- * 4 : events once per frame
- * 5 : events once per block
- * 6 : events once per sequence (*very* verbose) */
-#  define RAWLOG(l, ...) {                                      \
-                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
-                    fprintf(stderr, __VA_ARGS__);               \
-            }   }
-#  define DEBUGLOG(l, ...) {                                    \
-                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
-                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
-                    fprintf(stderr, " \n");                     \
-            }   }
-#else
-#  define RAWLOG(l, ...)      {}    /* disabled */
-#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#include "debug.h"
+#ifndef assert
+#  define assert(condition) ((void)0)
 #endif
 
 
@@ -266,14 +236,6 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
 #ifndef ZSTD_STATIC_H
 #define ZSTD_STATIC_H
 
-/* The objects defined into this file shall be considered experimental.
- * They are not considered stable, as their prototype may change in the future.
- * You can use them for tests, provide feedback, or if you can endure risks of future changes.
- */
-
-#if defined (__cplusplus)
-extern "C" {
-#endif
 
 /* *************************************
 *  Types
@@ -360,9 +322,6 @@ static size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t maxDstS
 */
 
 
-#if defined (__cplusplus)
-}
-#endif
 
 
 #endif  /* ZSTD_STATIC_H */
@@ -375,10 +334,6 @@ static size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t maxDstS
 #ifndef ZSTD_CCOMMON_H_MODULE
 #define ZSTD_CCOMMON_H_MODULE
 
-#if defined (__cplusplus)
-extern "C" {
-#endif
-
 /* *************************************
 *  Common macros
 ***************************************/
@@ -450,10 +405,6 @@ static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
 }
 
 
-#if defined (__cplusplus)
-}
-#endif
-
 
 /* ******************************************************************
    FSE : Finite State Entropy coder
@@ -1142,6 +1093,7 @@ static size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, un
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
     /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSE_DECODE_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
     DTableH.tableLog = (U16)tableLog;
     for (s=0; s<=maxSymbolValue; s++)
     {
@@ -2991,7 +2943,7 @@ static size_t ZSTD_execSequence(BYTE* op,
     }
     else
     {
-        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8, but must be signed */
     }
     return sequenceLength;
 }
@@ -3670,8 +3622,3 @@ size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDs
 
 ZSTD_DCtx* ZSTDv04_createDCtx(void) { return ZSTD_createDCtx(); }
 size_t ZSTDv04_freeDCtx(ZSTD_DCtx* dctx) { return ZSTD_freeDCtx(dctx); }
-
-size_t ZSTDv04_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize)
-{
-    return ZSTD_getFrameParams(params, src, srcSize);
-}
diff --git a/lib/legacy/zstd_v05.c b/lib/legacy/zstd_v05.c
index a5e1b1ffc8ce..a1580a271eea 100644
--- a/lib/legacy/zstd_v05.c
+++ b/lib/legacy/zstd_v05.c
@@ -1224,6 +1224,7 @@ size_t FSEv05_buildDTable(FSEv05_DTable* dt, const short* normalizedCounter, uns
     if (tableLog > FSEv05_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
     /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSEv05_FUNCTION_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
     DTableH.tableLog = (U16)tableLog;
     for (s=0; s<=maxSymbolValue; s++) {
         if (normalizedCounter[s]==-1) {
@@ -2658,6 +2659,7 @@ struct ZSTDv05_DCtx_s
     BYTE headerBuffer[ZSTDv05_frameHeaderSize_max];
 };  /* typedef'd to ZSTDv05_DCtx within "zstd_static.h" */
 
+size_t ZSTDv05_sizeofDCtx (void); /* Hidden declaration */
 size_t ZSTDv05_sizeofDCtx (void) { return sizeof(ZSTDv05_DCtx); }
 
 size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx)
@@ -2822,7 +2824,7 @@ static size_t ZSTDv05_decodeFrameHeader_Part2(ZSTDv05_DCtx* zc, const void* src,
 }
 
 
-size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+static size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
 {
     const BYTE* const in = (const BYTE* const)src;
     BYTE headerFlags;
@@ -2845,6 +2847,7 @@ size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t*
 
 static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
@@ -2853,8 +2856,8 @@ static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src
 
 /*! ZSTDv05_decodeLiteralsBlock() :
     @return : nb of bytes read from src (< srcSize ) */
-size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
-                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+static size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
+                                    const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
 {
     const BYTE* const istart = (const BYTE*) src;
 
@@ -2988,7 +2991,7 @@ size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
 }
 
 
-size_t ZSTDv05_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+static size_t ZSTDv05_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
                          FSEv05_DTable* DTableLL, FSEv05_DTable* DTableML, FSEv05_DTable* DTableOffb,
                          const void* src, size_t srcSize, U32 flagStaticTable)
 {
@@ -3297,11 +3300,11 @@ static size_t ZSTDv05_decompressSequences(
     BYTE* const ostart = (BYTE* const)dst;
     BYTE* op = ostart;
     BYTE* const oend = ostart + maxDstSize;
-    size_t errorCode, dumpsLength;
+    size_t errorCode, dumpsLength=0;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* const litEnd = litPtr + dctx->litSize;
-    int nbSeq;
-    const BYTE* dumps;
+    int nbSeq=0;
+    const BYTE* dumps = NULL;
     U32* DTableLL = dctx->LLTable;
     U32* DTableML = dctx->MLTable;
     U32* DTableOffb = dctx->OffTable;
@@ -3410,10 +3413,10 @@ static size_t ZSTDv05_decompress_continueDCtx(ZSTDv05_DCtx* dctx,
     BYTE* const oend = ostart + maxDstSize;
     size_t remainingSize = srcSize;
     blockProperties_t blockProperties;
+    memset(&blockProperties, 0, sizeof(blockProperties));
 
     /* Frame Header */
-    {
-        size_t frameHeaderSize;
+    {   size_t frameHeaderSize;
         if (srcSize < ZSTDv05_frameHeaderSize_min+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
         frameHeaderSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
         if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
diff --git a/lib/legacy/zstd_v06.c b/lib/legacy/zstd_v06.c
index 8b068b3e5464..60d8d6fd9a9b 100644
--- a/lib/legacy/zstd_v06.c
+++ b/lib/legacy/zstd_v06.c
@@ -1250,9 +1250,7 @@ const char* FSEv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
 /* **************************************************************
 *  HUF Error Management
 ****************************************************************/
-unsigned HUFv06_isError(size_t code) { return ERR_isError(code); }
-
-const char* HUFv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+static unsigned HUFv06_isError(size_t code) { return ERR_isError(code); }
 
 
 /*-**************************************************************
@@ -2823,7 +2821,8 @@ struct ZSTDv06_DCtx_s
     BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
 };  /* typedef'd to ZSTDv06_DCtx within "zstd_static.h" */
 
-size_t ZSTDv06_sizeofDCtx (void) { return sizeof(ZSTDv06_DCtx); }   /* non published interface */
+size_t ZSTDv06_sizeofDCtx (void); /* Hidden declaration */
+size_t ZSTDv06_sizeofDCtx (void) { return sizeof(ZSTDv06_DCtx); }
 
 size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx)
 {
@@ -3022,7 +3021,7 @@ typedef struct
 
 /*! ZSTDv06_getcBlockSize() :
 *   Provides the size of compressed block from block header `src` */
-size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+static size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
 {
     const BYTE* const in = (const BYTE* const)src;
     U32 cSize;
@@ -3041,6 +3040,7 @@ size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t*
 
 static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
@@ -3049,7 +3049,7 @@ static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* sr
 
 /*! ZSTDv06_decodeLiteralsBlock() :
     @return : nb of bytes read from src (< srcSize ) */
-size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
+static size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
                           const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
 {
     const BYTE* const istart = (const BYTE*) src;
@@ -3183,7 +3183,7 @@ size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
     @return : nb bytes read from src,
               or an error code if it fails, testable with ZSTDv06_isError()
 */
-size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLog,
+static size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
                                  const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
 {
@@ -3213,7 +3213,7 @@ size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLo
 }
 
 
-size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
+static size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
                              FSEv06_DTable* DTableLL, FSEv06_DTable* DTableML, FSEv06_DTable* DTableOffb, U32 flagRepeatTable,
                              const void* src, size_t srcSize)
 {
@@ -3358,7 +3358,7 @@ static void ZSTDv06_decodeSequence(seq_t* seq, seqState_t* seqState)
 }
 
 
-size_t ZSTDv06_execSequence(BYTE* op,
+static size_t ZSTDv06_execSequence(BYTE* op,
                                 BYTE* const oend, seq_t sequence,
                                 const BYTE** litPtr, const BYTE* const litLimit,
                                 const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
@@ -4006,7 +4006,7 @@ size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
                     if (ZSTDv06_isError(hSize)) return hSize;
                     if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
                         memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
-                        zbd->lhSize += iend-ip; ip = iend; notDone = 0;
+                        zbd->lhSize += iend-ip;
                         *dstCapacityPtr = 0;
                         return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
diff --git a/lib/legacy/zstd_v07.c b/lib/legacy/zstd_v07.c
index 70b170f0f154..c7bb7a529813 100644
--- a/lib/legacy/zstd_v07.c
+++ b/lib/legacy/zstd_v07.c
@@ -2628,7 +2628,7 @@ const char* ZBUFFv07_getErrorName(size_t errorCode) { return ERR_getErrorName(er
 
 
 
-void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
+static void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
 {
     void* address = malloc(size);
     (void)opaque;
@@ -2636,7 +2636,7 @@ void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
     return address;
 }
 
-void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
+static void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
 {
     (void)opaque;
     /* if (address) printf("free %p opaque=%p \n", address, opaque); */
@@ -3150,10 +3150,10 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
     const BYTE* ip = (const BYTE*)src;
 
     if (srcSize < ZSTDv07_frameHeaderSize_min) return ZSTDv07_frameHeaderSize_min;
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
     if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
         if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
             if (srcSize < ZSTDv07_skippableHeaderSize) return ZSTDv07_skippableHeaderSize; /* magic number + skippable frame length */
-            memset(fparamsPtr, 0, sizeof(*fparamsPtr));
             fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
             fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
             return 0;
@@ -3175,11 +3175,13 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
         U32 windowSize = 0;
         U32 dictID = 0;
         U64 frameContentSize = 0;
-        if ((fhdByte & 0x08) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits, which must be zero */
+        if ((fhdByte & 0x08) != 0)   /* reserved bits, which must be zero */
+            return ERROR(frameParameter_unsupported);
         if (!directMode) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTDv07_WINDOWLOG_ABSOLUTEMIN;
-            if (windowLog > ZSTDv07_WINDOWLOG_MAX) return ERROR(frameParameter_unsupported);
+            if (windowLog > ZSTDv07_WINDOWLOG_MAX)
+                return ERROR(frameParameter_unsupported);
             windowSize = (1U << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
@@ -3201,7 +3203,8 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
             case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
         }
         if (!windowSize) windowSize = (U32)frameContentSize;
-        if (windowSize > windowSizeMax) return ERROR(frameParameter_unsupported);
+        if (windowSize > windowSizeMax)
+            return ERROR(frameParameter_unsupported);
         fparamsPtr->frameContentSize = frameContentSize;
         fparamsPtr->windowSize = windowSize;
         fparamsPtr->dictID = dictID;
@@ -3220,11 +3223,10 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
                    - frame header not completely provided (`srcSize` too small) */
 unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize)
 {
-    {   ZSTDv07_frameParams fparams;
-        size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
-        if (frResult!=0) return 0;
-        return fparams.frameContentSize;
-    }
+    ZSTDv07_frameParams fparams;
+    size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
+    if (frResult!=0) return 0;
+    return fparams.frameContentSize;
 }
 
 
@@ -3248,7 +3250,7 @@ typedef struct
 
 /*! ZSTDv07_getcBlockSize() :
 *   Provides the size of compressed block from block header `src` */
-size_t ZSTDv07_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+static size_t ZSTDv07_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
 {
     const BYTE* const in = (const BYTE* const)src;
     U32 cSize;
@@ -3275,7 +3277,7 @@ static size_t ZSTDv07_copyRawBlock(void* dst, size_t dstCapacity, const void* sr
 
 /*! ZSTDv07_decodeLiteralsBlock() :
     @return : nb of bytes read from src (< srcSize ) */
-size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
+static size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
                           const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
 {
     const BYTE* const istart = (const BYTE*) src;
@@ -3409,7 +3411,7 @@ size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
     @return : nb bytes read from src,
               or an error code if it fails, testable with ZSTDv07_isError()
 */
-size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLog,
+static size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
                                  const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
 {
@@ -3439,7 +3441,7 @@ size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLo
 }
 
 
-size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
+static size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
                              FSEv07_DTable* DTableLL, FSEv07_DTable* DTableML, FSEv07_DTable* DTableOffb, U32 flagRepeatTable,
                              const void* src, size_t srcSize)
 {
@@ -3771,7 +3773,7 @@ ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockS
 }
 
 
-size_t ZSTDv07_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+static size_t ZSTDv07_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
 {
     if (length > dstCapacity) return ERROR(dstSize_tooSmall);
     memset(dst, byte, length);
@@ -3851,7 +3853,7 @@ static size_t ZSTDv07_decompressFrame(ZSTDv07_DCtx* dctx,
 *   It avoids reloading the dictionary each time.
 *   `preparedDCtx` must have been properly initialized using ZSTDv07_decompressBegin_usingDict().
 *   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
-size_t ZSTDv07_decompress_usingPreparedDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* refDCtx,
+static size_t ZSTDv07_decompress_usingPreparedDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* refDCtx,
                                          void* dst, size_t dstCapacity,
                                    const void* src, size_t srcSize)
 {
@@ -4146,7 +4148,7 @@ struct ZSTDv07_DDict_s {
     ZSTDv07_DCtx* refContext;
 };  /* typedef'd tp ZSTDv07_CDict within zstd.h */
 
-ZSTDv07_DDict* ZSTDv07_createDDict_advanced(const void* dict, size_t dictSize, ZSTDv07_customMem customMem)
+static ZSTDv07_DDict* ZSTDv07_createDDict_advanced(const void* dict, size_t dictSize, ZSTDv07_customMem customMem)
 {
     if (!customMem.customAlloc && !customMem.customFree)
         customMem = defaultCustomMem;
diff --git a/lib/zstd.h b/lib/zstd.h
index 6405da602e8f..f2af4ac8c429 100644
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -35,31 +35,43 @@ extern "C" {
 #endif
 
 
-/*******************************************************************************************************
+/*******************************************************************************
   Introduction
 
-  zstd, short for Zstandard, is a fast lossless compression algorithm,
-  targeting real-time compression scenarios at zlib-level and better compression ratios.
-  The zstd compression library provides in-memory compression and decompression functions.
-  The library supports compression levels from 1 up to ZSTD_maxCLevel() which is currently 22.
-  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
+  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting
+  real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression
+  functions.
+
+  The library supports regular compression levels from 1 up to ZSTD_maxCLevel(),
+  which is currently 22. Levels >= 20, labeled `--ultra`, should be used with
+  caution, as they require more memory. The library also offers negative
+  compression levels, which extend the range of speed vs. ratio preferences.
+  The lower the level, the faster the speed (at the cost of compression).
+
   Compression can be done in:
     - a single step (described as Simple API)
     - a single step, reusing a context (described as Explicit context)
     - unbounded multiple steps (described as Streaming compression)
-  The compression ratio achievable on small data can be highly improved using a dictionary in:
+
+  The compression ratio achievable on small data can be highly improved using
+  a dictionary. Dictionary compression can be performed in:
     - a single step (described as Simple dictionary API)
-    - a single step, reusing a dictionary (described as Bulk-processing dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing
+      dictionary API)
 
-  Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
-  Advanced experimental APIs shall never be used with a dynamic library.
-  They are not "stable", their definition may change in the future. Only static linking is allowed.
-*********************************************************************************************************/
+  Advanced experimental functions can be accessed using
+  `#define ZSTD_STATIC_LINKING_ONLY` before including zstd.h.
+
+  Advanced experimental APIs should never be used with a dynamically-linked
+  library. They are not "stable"; their definitions or signatures may change in
+  the future. Only static linking is allowed.
+*******************************************************************************/
 
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
 #define ZSTD_VERSION_MINOR    3
-#define ZSTD_VERSION_RELEASE  4
+#define ZSTD_VERSION_RELEASE  7
 
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
 ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< useful to check dll version */
@@ -68,8 +80,14 @@ ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< useful to check dll versio
 #define ZSTD_QUOTE(str) #str
 #define ZSTD_EXPAND_AND_QUOTE(str) ZSTD_QUOTE(str)
 #define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
-ZSTDLIB_API const char* ZSTD_versionString(void);   /* added in v1.3.0 */
+ZSTDLIB_API const char* ZSTD_versionString(void);   /* v1.3.0+ */
 
+/***************************************
+*  Default constant
+***************************************/
+#ifndef ZSTD_CLEVEL_DEFAULT
+#  define ZSTD_CLEVEL_DEFAULT 3
+#endif
 
 /***************************************
 *  Simple API
@@ -96,7 +114,7 @@ ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
  *  `src` should point to the start of a ZSTD encoded frame.
  *  `srcSize` must be at least as large as the frame header.
  *            hint : any size >= `ZSTD_frameHeaderSize_max` is large enough.
- *  @return : - decompressed size of the frame in `src`, if known
+ *  @return : - decompressed size of `src` frame content, if known
  *            - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
  *            - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small)
  *   note 1 : a 0 return value means the frame is valid but "empty".
@@ -106,7 +124,8 @@ ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
  *            Optionally, application can rely on some implicit limit,
  *            as ZSTD_decompress() only needs an upper bound of decompressed size.
  *            (For example, data could be necessarily cut into blocks <= 16 KB).
- *   note 3 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size is always present when compression is completed using single-pass functions,
+ *            such as ZSTD_compress(), ZSTD_compressCCtx() ZSTD_compress_usingDict() or ZSTD_compress_usingCDict().
  *   note 4 : decompressed size can be very large (64-bits value),
  *            potentially larger than what local system can handle as a single memory segment.
  *            In which case, it's necessary to use streaming mode to decompress data.
@@ -123,8 +142,7 @@ ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t
  *  Both functions work the same way, but ZSTD_getDecompressedSize() blends
  *  "empty", "unknown" and "error" results to the same return value (0),
  *  while ZSTD_getFrameContentSize() gives them separate return values.
- * `src` is the start of a zstd compressed frame.
- * @return : content size to be decompressed, as a 64-bits value _if known and not empty_, 0 otherwise. */
+ * @return : decompressed size of `src` frame content _if known and not empty_, 0 otherwise. */
 ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
 
 
@@ -205,7 +223,8 @@ typedef struct ZSTD_CDict_s ZSTD_CDict;
  *  When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
  *  ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
  *  ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
- *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict */
+ *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict
+ *  Note : A ZSTD_CDict can be created with an empty dictionary, but it is inefficient for small data. */
 ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
                                          int compressionLevel);
 
@@ -217,7 +236,9 @@ ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
  *  Compression using a digested Dictionary.
  *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
  *  Note that compression level is decided during dictionary creation.
- *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
+ *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no)
+ *  Note : ZSTD_compress_usingCDict() can be used with a ZSTD_CDict created from an empty dictionary.
+ *         But it is inefficient for small data, and it is recommended to use ZSTD_compressCCtx(). */
 ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
                                             void* dst, size_t dstCapacity,
                                       const void* src, size_t srcSize,
@@ -272,39 +293,44 @@ typedef struct ZSTD_outBuffer_s {
 *  since it will play nicer with system's memory, by re-using already allocated memory.
 *  Use one separate ZSTD_CStream per thread for parallel execution.
 *
-*  Start a new compression by initializing ZSTD_CStream.
+*  Start a new compression by initializing ZSTD_CStream context.
 *  Use ZSTD_initCStream() to start a new compression operation.
-*  Use ZSTD_initCStream_usingDict() or ZSTD_initCStream_usingCDict() for a compression which requires a dictionary (experimental section)
+*  Use variants ZSTD_initCStream_usingDict() or ZSTD_initCStream_usingCDict() for streaming with dictionary (experimental section)
 *
-*  Use ZSTD_compressStream() repetitively to consume input stream.
-*  The function will automatically update both `pos` fields.
-*  Note that it may not consume the entire input, in which case `pos < size`,
-*  and it's up to the caller to present again remaining data.
+*  Use ZSTD_compressStream() as many times as necessary to consume input stream.
+*  The function will automatically update both `pos` fields within `input` and `output`.
+*  Note that the function may not consume the entire input,
+*  for example, because the output buffer is already full,
+*  in which case `input.pos < input.size`.
+*  The caller must check if input has been entirely consumed.
+*  If not, the caller must make some room to receive more compressed data,
+*  typically by emptying output buffer, or allocating a new output buffer,
+*  and then present again remaining input data.
 *  @return : a size hint, preferred nb of bytes to use as input for next function call
 *            or an error code, which can be tested using ZSTD_isError().
 *            Note 1 : it's just a hint, to help latency a little, any other value will work fine.
 *            Note 2 : size hint is guaranteed to be <= ZSTD_CStreamInSize()
 *
-*  At any moment, it's possible to flush whatever data remains within internal buffer, using ZSTD_flushStream().
-*  `output->pos` will be updated.
-*  Note that some content might still be left within internal buffer if `output->size` is too small.
-*  @return : nb of bytes still present within internal buffer (0 if it's empty)
+*  At any moment, it's possible to flush whatever data might remain stuck within internal buffer,
+*  using ZSTD_flushStream(). `output->pos` will be updated.
+*  Note that, if `output->size` is too small, a single invocation of ZSTD_flushStream() might not be enough (return code > 0).
+*  In which case, make some room to receive more compressed data, and call again ZSTD_flushStream().
+*  @return : 0 if internal buffers are entirely flushed,
+*            >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
 *            or an error code, which can be tested using ZSTD_isError().
 *
 *  ZSTD_endStream() instructs to finish a frame.
 *  It will perform a flush and write frame epilogue.
 *  The epilogue is required for decoders to consider a frame completed.
-*  ZSTD_endStream() may not be able to flush full data if `output->size` is too small.
-*  In which case, call again ZSTD_endStream() to complete the flush.
+*  flush() operation is the same, and follows same rules as ZSTD_flushStream().
 *  @return : 0 if frame fully completed and fully flushed,
-             or >0 if some data is still present within internal buffer
-                  (value is minimum size estimation for remaining data to flush, but it could be more)
+*            >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
 *            or an error code, which can be tested using ZSTD_isError().
 *
 * *******************************************************************/
 
 typedef ZSTD_CCtx ZSTD_CStream;  /**< CCtx and CStream are now effectively same object (>= v1.3.0) */
-                                 /* Continue to distinguish them for compatibility with versions <= v1.2.0 */
+                                 /* Continue to distinguish them for compatibility with older versions <= v1.2.0 */
 /*===== ZSTD_CStream management functions =====*/
 ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
 ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs);
@@ -335,15 +361,21 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output
 *  The function will update both `pos` fields.
 *  If `input.pos < input.size`, some input has not been consumed.
 *  It's up to the caller to present again remaining data.
+*  The function tries to flush all data decoded immediately, repecting buffer sizes.
 *  If `output.pos < output.size`, decoder has flushed everything it could.
-*  @return : 0 when a frame is completely decoded and fully flushed,
-*            an error code, which can be tested using ZSTD_isError(),
-*            any other value > 0, which means there is still some decoding to do to complete current frame.
-*            The return value is a suggested next input size (a hint to improve latency) that will never load more than the current frame.
+*  But if `output.pos == output.size`, there is no such guarantee,
+*  it's likely that some decoded data was not flushed and still remains within internal buffers.
+*  In which case, call ZSTD_decompressStream() again to flush whatever remains in the buffer.
+*  When no additional input is provided, amount of data flushed is necessarily <= ZSTD_BLOCKSIZE_MAX.
+* @return : 0 when a frame is completely decoded and fully flushed,
+*        or an error code, which can be tested using ZSTD_isError(),
+*        or any other value > 0, which means there is still some decoding or flushing to do to complete current frame :
+*                                the return value is a suggested next input size (a hint for better latency)
+*                                that will never load more than the current frame.
 * *******************************************************************************/
 
 typedef ZSTD_DCtx ZSTD_DStream;  /**< DCtx and DStream are now effectively same object (>= v1.3.0) */
-                                 /* For compatibility with versions <= v1.2.0, continue to consider them separated. */
+                                 /* For compatibility with versions <= v1.2.0, prefer differentiating them. */
 /*===== ZSTD_DStream management functions =====*/
 ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
 ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);
@@ -359,21 +391,28 @@ ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output
 
 
 
+
+#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
+#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
+
 /****************************************************************************************
- * START OF ADVANCED AND EXPERIMENTAL FUNCTIONS
+ *   ADVANCED AND EXPERIMENTAL FUNCTIONS
+ ****************************************************************************************
  * The definitions in this section are considered experimental.
  * They should never be used with a dynamic library, as prototypes may change in the future.
  * They are provided for advanced scenarios.
  * Use them only in association with static linking.
  * ***************************************************************************************/
 
-#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
-#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
+ZSTDLIB_API int ZSTD_minCLevel(void);  /*!< minimum negative compression level allowed */
 
-/* --- Constants ---*/
-#define ZSTD_MAGICNUMBER            0xFD2FB528   /* >= v0.8.0 */
+/* ---  Constants  ---*/
+#define ZSTD_MAGICNUMBER            0xFD2FB528   /* v0.8+ */
+#define ZSTD_MAGIC_DICTIONARY       0xEC30A437   /* v0.7+ */
 #define ZSTD_MAGIC_SKIPPABLE_START  0x184D2A50U
-#define ZSTD_MAGIC_DICTIONARY       0xEC30A437   /* >= v0.7.0 */
+
+#define ZSTD_BLOCKSIZELOG_MAX 17
+#define ZSTD_BLOCKSIZE_MAX   (1<<ZSTD_BLOCKSIZELOG_MAX)   /* define, for static allocation */
 
 #define ZSTD_WINDOWLOG_MAX_32   30
 #define ZSTD_WINDOWLOG_MAX_64   31
@@ -390,9 +429,10 @@ ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output
 #define ZSTD_SEARCHLOG_MIN       1
 #define ZSTD_SEARCHLENGTH_MAX    7   /* only for ZSTD_fast, other strategies are limited to 6 */
 #define ZSTD_SEARCHLENGTH_MIN    3   /* only for ZSTD_btopt, other strategies are limited to 4 */
-#define ZSTD_TARGETLENGTH_MIN    1   /* only used by btopt, btultra and btfast */
-#define ZSTD_LDM_MINMATCH_MIN    4
+#define ZSTD_TARGETLENGTH_MAX  ZSTD_BLOCKSIZE_MAX
+#define ZSTD_TARGETLENGTH_MIN    0   /* note : comparing this constant to an unsigned results in a tautological test */
 #define ZSTD_LDM_MINMATCH_MAX 4096
+#define ZSTD_LDM_MINMATCH_MIN    4
 #define ZSTD_LDM_BUCKETSIZELOG_MAX 8
 
 #define ZSTD_FRAMEHEADERSIZE_PREFIX 5   /* minimum input size to know frame header size */
@@ -404,7 +444,8 @@ static const size_t ZSTD_frameHeaderSize_max = ZSTD_FRAMEHEADERSIZE_MAX;
 static const size_t ZSTD_skippableHeaderSize = 8;  /* magic number + skippable frame length */
 
 
-/*--- Advanced types ---*/
+
+/* ---  Advanced types  --- */
 typedef enum { ZSTD_fast=1, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2,
                ZSTD_btlazy2, ZSTD_btopt, ZSTD_btultra } ZSTD_strategy;   /* from faster to stronger */
 
@@ -480,9 +521,9 @@ ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize)
 ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
 
 /*! ZSTD_frameHeaderSize() :
-*   `src` should point to the start of a ZSTD frame
-*   `srcSize` must be >= ZSTD_frameHeaderSize_prefix.
-*   @return : size of the Frame Header */
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
 ZSTDLIB_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
 
 
@@ -711,29 +752,48 @@ ZSTDLIB_API size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs, const
 
 /*! ZSTD_resetCStream() :
  *  start a new compression job, using same parameters from previous job.
- *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place..
+ *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place.
  *  Note that zcs must be init at least once before using ZSTD_resetCStream().
  *  If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
  *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
  *  For the time being, pledgedSrcSize==0 is interpreted as "srcSize unknown" for compatibility with older programs,
  *  but it will change to mean "empty" in future version, so use macro ZSTD_CONTENTSIZE_UNKNOWN instead.
- * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ * @return : 0, or an error code (which can be tested using ZSTD_isError())
+ */
 ZSTDLIB_API size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
 
 
 typedef struct {
-    unsigned long long ingested;
-    unsigned long long consumed;
-    unsigned long long produced;
+    unsigned long long ingested;   /* nb input bytes read and buffered */
+    unsigned long long consumed;   /* nb input bytes actually compressed */
+    unsigned long long produced;   /* nb of compressed bytes generated and buffered */
+    unsigned long long flushed;    /* nb of compressed bytes flushed : not provided; can be tracked from caller side */
+    unsigned currentJobID;         /* MT only : latest started job nb */
+    unsigned nbActiveWorkers;      /* MT only : nb of workers actively compressing at probe time */
 } ZSTD_frameProgression;
 
-/* ZSTD_getFrameProgression():
+/* ZSTD_getFrameProgression() :
  * tells how much data has been ingested (read from input)
  * consumed (input actually compressed) and produced (output) for current frame.
- * Therefore, (ingested - consumed) is amount of input data buffered internally, not yet compressed.
- * Can report progression inside worker threads (multi-threading and non-blocking mode).
+ * Note : (ingested - consumed) is amount of input data buffered internally, not yet compressed.
+ * Aggregates progression inside active worker threads.
  */
-ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+ZSTDLIB_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+
+/*! ZSTD_toFlushNow() :
+ *  Tell how many bytes are ready to be flushed immediately.
+ *  Useful for multithreading scenarios (nbWorkers >= 1).
+ *  Probe the oldest active job, defined as oldest job not yet entirely flushed,
+ *  and check its output buffer.
+ * @return : amount of data stored in oldest job and ready to be flushed immediately.
+ *  if @return == 0, it means either :
+ *  + there is no active job (could be checked with ZSTD_frameProgression()), or
+ *  + oldest job is still actively compressing data,
+ *    but everything it has produced has also been flushed so far,
+ *    therefore flushing speed is currently limited by production speed of oldest job
+ *    irrespective of the speed of concurrent newer jobs.
+ */
+ZSTDLIB_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
 
 
 
@@ -880,6 +940,11 @@ typedef struct {
     unsigned dictID;
     unsigned checksumFlag;
 } ZSTD_frameHeader;
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or requires larger `srcSize`.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
 ZSTDLIB_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
 ZSTDLIB_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
 
@@ -901,23 +966,15 @@ ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 /**       New advanced API (experimental)       */
 /* ============================================ */
 
-/* notes on API design :
- *   In this proposal, parameters are pushed one by one into an existing context,
- *   and then applied on all subsequent compression jobs.
- *   When no parameter is ever provided, CCtx is created with compression level ZSTD_CLEVEL_DEFAULT.
+/* API design :
+ *   In this advanced API, parameters are pushed one by one into an existing context,
+ *   using ZSTD_CCtx_set*() functions.
+ *   Pushed parameters are sticky : they are applied to next job, and any subsequent job.
+ *   It's possible to reset parameters to "default" using ZSTD_CCtx_reset().
+ *   Important : "sticky" parameters only work with `ZSTD_compress_generic()` !
+ *               For any other entry point, "sticky" parameters are ignored !
  *
  *   This API is intended to replace all others advanced / experimental API entry points.
- *   But it stands a reasonable chance to become "stable", after a reasonable testing period.
- */
-
-/* note on naming convention :
- *   Initially, the API favored names like ZSTD_setCCtxParameter() .
- *   In this proposal, convention is changed towards ZSTD_CCtx_setParameter() .
- *   The main driver is that it identifies more clearly the target object type.
- *   It feels clearer when considering multiple targets :
- *   ZSTD_CDict_setParameter() (rather than ZSTD_setCDictParameter())
- *   ZSTD_CCtxParams_setParameter()  (rather than ZSTD_setCCtxParamsParameter() )
- *   etc...
  */
 
 /* note on enum design :
@@ -947,7 +1004,7 @@ typedef enum {
     /* compression parameters */
     ZSTD_p_compressionLevel=100, /* Update all compression parameters according to pre-defined cLevel table
                               * Default level is ZSTD_CLEVEL_DEFAULT==3.
-                              * Special: value 0 means "do not change cLevel".
+                              * Special: value 0 means default, which is controlled by ZSTD_CLEVEL_DEFAULT.
                               * Note 1 : it's possible to pass a negative compression level by casting it to unsigned type.
                               * Note 2 : setting a level sets all default values of other compression parameters.
                               * Note 3 : setting compressionLevel automatically updates ZSTD_p_compressLiterals. */
@@ -956,16 +1013,19 @@ typedef enum {
                               * Special: value 0 means "use default windowLog".
                               * Note: Using a window size greater than ZSTD_MAXWINDOWSIZE_DEFAULT (default: 2^27)
                               *       requires explicitly allowing such window size during decompression stage. */
-    ZSTD_p_hashLog,          /* Size of the probe table, as a power of 2.
+    ZSTD_p_hashLog,          /* Size of the initial probe table, as a power of 2.
                               * Resulting table size is (1 << (hashLog+2)).
                               * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX.
                               * Larger tables improve compression ratio of strategies <= dFast,
                               * and improve speed of strategies > dFast.
                               * Special: value 0 means "use default hashLog". */
-    ZSTD_p_chainLog,         /* Size of the full-search table, as a power of 2.
+    ZSTD_p_chainLog,         /* Size of the multi-probe search table, as a power of 2.
                               * Resulting table size is (1 << (chainLog+2)).
+                              * Must be clamped between ZSTD_CHAINLOG_MIN and ZSTD_CHAINLOG_MAX.
                               * Larger tables result in better and slower compression.
                               * This parameter is useless when using "fast" strategy.
+                              * Note it's still useful when using "dfast" strategy,
+                              * in which case it defines a secondary probe table.
                               * Special: value 0 means "use default chainLog". */
     ZSTD_p_searchLog,        /* Number of search attempts, as a power of 2.
                               * More attempts result in better and slower compression.
@@ -1047,27 +1107,52 @@ typedef enum {
     /* experimental parameters - no stability guaranteed                   */
     /* =================================================================== */
 
-    ZSTD_p_compressLiterals=1000, /* control huffman compression of literals (enabled) by default.
-                              * disabling it improves speed and decreases compression ratio by a large amount.
-                              * note : this setting is automatically updated when changing compression level.
-                              *        positive compression levels set ZSTD_p_compressLiterals to 1.
-                              *        negative compression levels set ZSTD_p_compressLiterals to 0. */
-
     ZSTD_p_forceMaxWindow=1100, /* Force back-reference distances to remain < windowSize,
                               * even when referencing into Dictionary content (default:0) */
+    ZSTD_p_forceAttachDict,  /* ZSTD supports usage of a CDict in-place
+                              * (avoiding having to copy the compression tables
+                              * from the CDict into the working context). Using
+                              * a CDict in this way saves an initial setup step,
+                              * but comes at the cost of more work per byte of
+                              * input. ZSTD has a simple internal heuristic that
+                              * guesses which strategy will be faster. You can
+                              * use this flag to override that guess.
+                              *
+                              * Note that the by-reference, in-place strategy is
+                              * only used when reusing a compression context
+                              * with compatible compression parameters. (If
+                              * incompatible / uninitialized, the working
+                              * context needs to be cleared anyways, which is
+                              * about as expensive as overwriting it with the
+                              * dictionary context, so there's no savings in
+                              * using the CDict by-ref.)
+                              *
+                              * Values greater than 0 force attaching the dict.
+                              * Values less than 0 force copying the dict.
+                              * 0 selects the default heuristic-guided behavior.
+                              */
 
 } ZSTD_cParameter;
 
 
 /*! ZSTD_CCtx_setParameter() :
  *  Set one compression parameter, selected by enum ZSTD_cParameter.
- *  Setting a parameter is generally only possible during frame initialization (before starting compression),
- *  except for a few exceptions which can be updated during compression: compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
- *  Note : when `value` is an enum, cast it to unsigned for proper type checking.
- *  @result : informational value (typically, value being set clamped correctly),
+ *  Setting a parameter is generally only possible during frame initialization (before starting compression).
+ *  Exception : when using multi-threading mode (nbThreads >= 1),
+ *              following parameters can be updated _during_ compression (within same frame):
+ *              => compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
+ *              new parameters will be active on next job, or after a flush().
+ *  Note : when `value` type is not unsigned (int, or enum), cast it to unsigned for proper type checking.
+ *  @result : informational value (typically, value being set, correctly clamped),
  *            or an error code (which can be tested with ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value);
 
+/*! ZSTD_CCtx_getParameter() :
+ * Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_getParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned* value);
+
 /*! ZSTD_CCtx_setPledgedSrcSize() :
  *  Total input data size to be compressed as a single frame.
  *  This value will be controlled at the end, and result in error if not respected.
@@ -1114,36 +1199,55 @@ ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
 
 /*! ZSTD_CCtx_refPrefix() :
  *  Reference a prefix (single-usage dictionary) for next compression job.
- *  Decompression need same prefix to properly regenerate data.
- *  Prefix is **only used once**. Tables are discarded at end of compression job.
- *  Subsequent compression jobs will be done without prefix (if none is explicitly referenced).
- *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_CDict instead.
+ *  Decompression will need same prefix to properly regenerate data.
+ *  Compressing with a prefix is similar in outcome as performing a diff and compressing it,
+ *  but performs much faster, especially during decompression (compression speed is tunable with compression level).
+ *  Note that prefix is **only used once**. Tables are discarded at end of compression job (ZSTD_e_end).
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
- *  Note 1 : Prefix buffer is referenced. It must outlive compression job.
- *  Note 2 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+ *  Note 1 : Prefix buffer is referenced. It **must** outlive compression job.
+ *           Its contain must remain unmodified up to end of compression (ZSTD_e_end).
+ *  Note 2 : If the intention is to diff some large src data blob with some prior version of itself,
+ *           ensure that the window size is large enough to contain the entire source.
+ *           See ZSTD_p_windowLog.
+ *  Note 3 : Referencing a prefix involves building tables, which are dependent on compression parameters.
  *           It's a CPU consuming operation, with non-negligible impact on latency.
- *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *           If there is a need to use same prefix multiple times, consider loadDictionary instead.
+ *  Note 4 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
  *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode. */
-ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize);
-ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
+                                       const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx,
+                                       const void* prefix, size_t prefixSize,
+                                       ZSTD_dictContentType_e dictContentType);
 
 /*! ZSTD_CCtx_reset() :
  *  Return a CCtx to clean state.
  *  Useful after an error, or to interrupt an ongoing compression job and start a new one.
  *  Any internal data not yet flushed is cancelled.
- *  Dictionary (if any) is dropped.
- *  All parameters are back to default values.
- *  It's possible to modify compression parameters after a reset.
+ *  The parameters and dictionary are kept unchanged, to reset them use ZSTD_CCtx_resetParameters().
  */
 ZSTDLIB_API void ZSTD_CCtx_reset(ZSTD_CCtx* cctx);
 
+/*! ZSTD_CCtx_resetParameters() :
+ *  All parameters are back to default values (compression level is ZSTD_CLEVEL_DEFAULT).
+ *  Dictionary (if any) is dropped.
+ *  Resetting parameters is only possible during frame initialization (before starting compression).
+ *  To reset the context use ZSTD_CCtx_reset().
+ *  @return 0 or an error code (which can be checked with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_resetParameters(ZSTD_CCtx* cctx);
+
 
 
 typedef enum {
-    ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal conditions */
-    ZSTD_e_flush,      /* flush any data provided so far - frame will continue, future data can still reference previous data for better compression */
-    ZSTD_e_end         /* flush any remaining data and close current frame. Any additional data starts a new frame. */
+    ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal compression ratio */
+    ZSTD_e_flush,      /* flush any data provided so far,
+                        * it creates (at least) one new block, that can be decoded immediately on reception;
+                        * frame will continue: any future data can still reference previously compressed data, improving compression. */
+    ZSTD_e_end         /* flush any remaining data and close current frame.
+                        * any additional data starts a new frame.
+                        * each frame is independent (does not reference any content from previous frame). */
 } ZSTD_EndDirective;
 
 /*! ZSTD_compress_generic() :
@@ -1235,6 +1339,13 @@ ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, Z
  */
 ZSTDLIB_API size_t ZSTD_CCtxParam_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value);
 
+/*! ZSTD_CCtxParam_getParameter() :
+ * Similar to ZSTD_CCtx_getParameter.
+ * Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParam_getParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned* value);
+
 /*! ZSTD_CCtx_setParametersUsingCCtxParams() :
  *  Apply a set of ZSTD_CCtx_params to the compression context.
  *  This can be done even after compression is started,
@@ -1246,10 +1357,13 @@ ZSTDLIB_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
         ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
 
 
-/*===   Advanced parameters for decompression API  ===*/
+/* ==================================== */
+/*===   Advanced decompression API   ===*/
+/* ==================================== */
 
-/* The following parameters must be set after creating a ZSTD_DCtx* (or ZSTD_DStream*) object,
- * but before starting decompression of a frame.
+/* The following API works the same way as the advanced compression API :
+ * a context is created, parameters are pushed into it one by one,
+ * then the context can be used to decompress data using an interface similar to the straming API.
  */
 
 /*! ZSTD_DCtx_loadDictionary() :
@@ -1286,17 +1400,25 @@ ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
 
 /*! ZSTD_DCtx_refPrefix() :
  *  Reference a prefix (single-usage dictionary) for next compression job.
- *  Prefix is **only used once**. It must be explicitly referenced before each frame.
- *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_DDict instead.
+ *  This is the reverse operation of ZSTD_CCtx_refPrefix(),
+ *  and must use the same prefix as the one used during compression.
+ *  Prefix is **only used once**. Reference is discarded at end of frame.
+ *  End of frame is reached when ZSTD_DCtx_decompress_generic() returns 0.
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Note 1 : Adding any prefix (including NULL) invalidates any previously set prefix or dictionary
- *  Note 2 : Prefix buffer is referenced. It must outlive compression job.
+ *  Note 2 : Prefix buffer is referenced. It **must** outlive decompression job.
+ *           Prefix buffer must remain unmodified up to the end of frame,
+ *           reached when ZSTD_DCtx_decompress_generic() returns 0.
  *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
  *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode.
  *  Note 4 : Referencing a raw content prefix has almost no cpu nor memory cost.
+ *           A fulldict prefix is more costly though.
  */
-ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize);
-ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx,
+                                    const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx,
+                                    const void* prefix, size_t prefixSize,
+                                    ZSTD_dictContentType_e dictContentType);
 
 
 /*! ZSTD_DCtx_setMaxWindowSize() :
@@ -1318,6 +1440,13 @@ ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowS
 ZSTDLIB_API size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
 
 
+/*! ZSTD_getFrameHeader_advanced() :
+ *  same as ZSTD_getFrameHeader(),
+ *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
+ZSTDLIB_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr,
+                        const void* src, size_t srcSize, ZSTD_format_e format);
+
+
 /*! ZSTD_decompress_generic() :
  *  Behave the same as ZSTD_decompressStream.
  *  Decompression parameters cannot be changed once decompression is started.
@@ -1383,8 +1512,6 @@ ZSTDLIB_API void ZSTD_DCtx_reset(ZSTD_DCtx* dctx);
         Use ZSTD_insertBlock() for such a case.
 */
 
-#define ZSTD_BLOCKSIZELOG_MAX 17
-#define ZSTD_BLOCKSIZE_MAX   (1<<ZSTD_BLOCKSIZELOG_MAX)   /* define, for static allocation */
 /*=====   Raw zstd block functions  =====*/
 ZSTDLIB_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
 ZSTDLIB_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);