diff options
author | Dimitry Andric <dim@FreeBSD.org> | 2019-01-20 14:02:54 +0000 |
---|---|---|
committer | Dimitry Andric <dim@FreeBSD.org> | 2019-01-20 14:02:54 +0000 |
commit | d5ea6fa648f8835a44adfb322b788e615d77cb71 (patch) | |
tree | 570aa90958a58b9d6a71fa8594ee0ad8d2a18f2c /contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h | |
parent | d9484dd61cc151c4f34c31e07f693fefa66316b5 (diff) | |
parent | 676fbe8105eeb6ff4bb2ed261cb212fcfdbe7b63 (diff) |
Merge clang trunk r351319, resolve conflicts, and update FREEBSD-Xlist.
Notes
Notes:
svn path=/projects/clang800-import/; revision=343214
Diffstat (limited to 'contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h')
-rw-r--r-- | contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h | 302 |
1 files changed, 216 insertions, 86 deletions
diff --git a/contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h b/contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h index 8a00b3afa9d5..6e6c293af22e 100644 --- a/contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h +++ b/contrib/llvm/tools/clang/lib/Headers/avx512dqintrin.h @@ -29,180 +29,309 @@ #define __AVX512DQINTRIN_H /* Define the default attributes for the functions in this file. */ -#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512dq"), __min_vector_width__(512))) +#define __DEFAULT_FN_ATTRS512 __attribute__((__always_inline__, __nodebug__, __target__("avx512dq"), __min_vector_width__(512))) +#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512dq"))) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline __mmask8 __DEFAULT_FN_ATTRS +_knot_mask8(__mmask8 __M) +{ + return __builtin_ia32_knotqi(__M); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kand_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_kandqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kandn_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_kandnqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kor_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_korqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kxnor_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_kxnorqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kxor_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_kxorqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_kortestc_mask8_u8(__mmask8 __A, __mmask8 __B) +{ + return (unsigned char)__builtin_ia32_kortestcqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_kortestz_mask8_u8(__mmask8 __A, __mmask8 __B) +{ + return (unsigned char)__builtin_ia32_kortestzqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_kortest_mask8_u8(__mmask8 __A, __mmask8 __B, unsigned char *__C) { + *__C = (unsigned char)__builtin_ia32_kortestcqi(__A, __B); + return (unsigned char)__builtin_ia32_kortestzqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktestc_mask8_u8(__mmask8 __A, __mmask8 __B) +{ + return (unsigned char)__builtin_ia32_ktestcqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktestz_mask8_u8(__mmask8 __A, __mmask8 __B) +{ + return (unsigned char)__builtin_ia32_ktestzqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktest_mask8_u8(__mmask8 __A, __mmask8 __B, unsigned char *__C) { + *__C = (unsigned char)__builtin_ia32_ktestcqi(__A, __B); + return (unsigned char)__builtin_ia32_ktestzqi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktestc_mask16_u8(__mmask16 __A, __mmask16 __B) +{ + return (unsigned char)__builtin_ia32_ktestchi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktestz_mask16_u8(__mmask16 __A, __mmask16 __B) +{ + return (unsigned char)__builtin_ia32_ktestzhi(__A, __B); +} + +static __inline__ unsigned char __DEFAULT_FN_ATTRS +_ktest_mask16_u8(__mmask16 __A, __mmask16 __B, unsigned char *__C) { + *__C = (unsigned char)__builtin_ia32_ktestchi(__A, __B); + return (unsigned char)__builtin_ia32_ktestzhi(__A, __B); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_kadd_mask8(__mmask8 __A, __mmask8 __B) +{ + return (__mmask8)__builtin_ia32_kaddqi((__mmask8)__A, (__mmask8)__B); +} + +static __inline__ __mmask16 __DEFAULT_FN_ATTRS +_kadd_mask16(__mmask16 __A, __mmask16 __B) +{ + return (__mmask16)__builtin_ia32_kaddhi((__mmask16)__A, (__mmask16)__B); +} + +#define _kshiftli_mask8(A, I) \ + (__mmask8)__builtin_ia32_kshiftliqi((__mmask8)(A), (unsigned int)(I)) + +#define _kshiftri_mask8(A, I) \ + (__mmask8)__builtin_ia32_kshiftriqi((__mmask8)(A), (unsigned int)(I)) + +static __inline__ unsigned int __DEFAULT_FN_ATTRS +_cvtmask8_u32(__mmask8 __A) { + return (unsigned int)__builtin_ia32_kmovb((__mmask8)__A); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_cvtu32_mask8(unsigned int __A) { + return (__mmask8)__builtin_ia32_kmovb((__mmask8)__A); +} + +static __inline__ __mmask8 __DEFAULT_FN_ATTRS +_load_mask8(__mmask8 *__A) { + return (__mmask8)__builtin_ia32_kmovb(*(__mmask8 *)__A); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_store_mask8(__mmask8 *__A, __mmask8 __B) { + *(__mmask8 *)__A = __builtin_ia32_kmovb((__mmask8)__B); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mullo_epi64 (__m512i __A, __m512i __B) { return (__m512i) ((__v8du) __A * (__v8du) __B); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_mullo_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m512i __B) { return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, (__v8di)_mm512_mullo_epi64(__A, __B), (__v8di)__W); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_mullo_epi64(__mmask8 __U, __m512i __A, __m512i __B) { return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, (__v8di)_mm512_mullo_epi64(__A, __B), (__v8di)_mm512_setzero_si512()); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_xor_pd(__m512d __A, __m512d __B) { return (__m512d)((__v8du)__A ^ (__v8du)__B); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_xor_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_xor_pd(__A, __B), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_xor_pd(__mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_xor_pd(__A, __B), (__v8df)_mm512_setzero_pd()); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_xor_ps (__m512 __A, __m512 __B) { return (__m512)((__v16su)__A ^ (__v16su)__B); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_xor_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_xor_ps(__A, __B), (__v16sf)__W); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_xor_ps(__mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_xor_ps(__A, __B), (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_or_pd(__m512d __A, __m512d __B) { return (__m512d)((__v8du)__A | (__v8du)__B); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_or_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_or_pd(__A, __B), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_or_pd(__mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_or_pd(__A, __B), (__v8df)_mm512_setzero_pd()); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_or_ps(__m512 __A, __m512 __B) { return (__m512)((__v16su)__A | (__v16su)__B); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_or_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_or_ps(__A, __B), (__v16sf)__W); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_or_ps(__mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_or_ps(__A, __B), (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_and_pd(__m512d __A, __m512d __B) { return (__m512d)((__v8du)__A & (__v8du)__B); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_and_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_and_pd(__A, __B), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_and_pd(__mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_and_pd(__A, __B), (__v8df)_mm512_setzero_pd()); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_and_ps(__m512 __A, __m512 __B) { return (__m512)((__v16su)__A & (__v16su)__B); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_and_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_and_ps(__A, __B), (__v16sf)__W); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_and_ps(__mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_and_ps(__A, __B), (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_andnot_pd(__m512d __A, __m512d __B) { return (__m512d)(~(__v8du)__A & (__v8du)__B); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_andnot_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_andnot_pd(__A, __B), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_andnot_pd(__mmask8 __U, __m512d __A, __m512d __B) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_andnot_pd(__A, __B), (__v8df)_mm512_setzero_pd()); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_andnot_ps(__m512 __A, __m512 __B) { return (__m512)(~(__v16su)__A & (__v16su)__B); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_andnot_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_andnot_ps(__A, __B), (__v16sf)__W); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_andnot_ps(__mmask16 __U, __m512 __A, __m512 __B) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, (__v16sf)_mm512_andnot_ps(__A, __B), (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvtpd_epi64 (__m512d __A) { return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -210,7 +339,7 @@ _mm512_cvtpd_epi64 (__m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvtpd_epi64 (__m512i __W, __mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A, (__v8di) __W, @@ -218,7 +347,7 @@ _mm512_mask_cvtpd_epi64 (__m512i __W, __mmask8 __U, __m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtpd_epi64 (__mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvtpd2qq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -241,7 +370,7 @@ _mm512_maskz_cvtpd_epi64 (__mmask8 __U, __m512d __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvtpd_epu64 (__m512d __A) { return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -249,7 +378,7 @@ _mm512_cvtpd_epu64 (__m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvtpd_epu64 (__m512i __W, __mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A, (__v8di) __W, @@ -257,7 +386,7 @@ _mm512_mask_cvtpd_epu64 (__m512i __W, __mmask8 __U, __m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtpd_epu64 (__mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvtpd2uqq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -280,7 +409,7 @@ _mm512_maskz_cvtpd_epu64 (__mmask8 __U, __m512d __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvtps_epi64 (__m256 __A) { return (__m512i) __builtin_ia32_cvtps2qq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -288,7 +417,7 @@ _mm512_cvtps_epi64 (__m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvtps_epi64 (__m512i __W, __mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvtps2qq512_mask ((__v8sf) __A, (__v8di) __W, @@ -296,7 +425,7 @@ _mm512_mask_cvtps_epi64 (__m512i __W, __mmask8 __U, __m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtps_epi64 (__mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvtps2qq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -319,7 +448,7 @@ _mm512_maskz_cvtps_epi64 (__mmask8 __U, __m256 __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvtps_epu64 (__m256 __A) { return (__m512i) __builtin_ia32_cvtps2uqq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -327,7 +456,7 @@ _mm512_cvtps_epu64 (__m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvtps_epu64 (__m512i __W, __mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvtps2uqq512_mask ((__v8sf) __A, (__v8di) __W, @@ -335,7 +464,7 @@ _mm512_mask_cvtps_epu64 (__m512i __W, __mmask8 __U, __m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtps_epu64 (__mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvtps2uqq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -359,19 +488,19 @@ _mm512_maskz_cvtps_epu64 (__mmask8 __U, __m256 __A) { (__mmask8)(U), (int)(R)) -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_cvtepi64_pd (__m512i __A) { return (__m512d)__builtin_convertvector((__v8di)__A, __v8df); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_cvtepi64_pd (__m512d __W, __mmask8 __U, __m512i __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_cvtepi64_pd(__A), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtepi64_pd (__mmask8 __U, __m512i __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_cvtepi64_pd(__A), @@ -393,7 +522,7 @@ _mm512_maskz_cvtepi64_pd (__mmask8 __U, __m512i __A) { (__v8df)_mm512_setzero_pd(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_cvtepi64_ps (__m512i __A) { return (__m256) __builtin_ia32_cvtqq2ps512_mask ((__v8di) __A, (__v8sf) _mm256_setzero_ps(), @@ -401,7 +530,7 @@ _mm512_cvtepi64_ps (__m512i __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_mask_cvtepi64_ps (__m256 __W, __mmask8 __U, __m512i __A) { return (__m256) __builtin_ia32_cvtqq2ps512_mask ((__v8di) __A, (__v8sf) __W, @@ -409,7 +538,7 @@ _mm512_mask_cvtepi64_ps (__m256 __W, __mmask8 __U, __m512i __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtepi64_ps (__mmask8 __U, __m512i __A) { return (__m256) __builtin_ia32_cvtqq2ps512_mask ((__v8di) __A, (__v8sf) _mm256_setzero_ps(), @@ -433,7 +562,7 @@ _mm512_maskz_cvtepi64_ps (__mmask8 __U, __m512i __A) { (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvttpd_epi64 (__m512d __A) { return (__m512i) __builtin_ia32_cvttpd2qq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -441,7 +570,7 @@ _mm512_cvttpd_epi64 (__m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvttpd_epi64 (__m512i __W, __mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvttpd2qq512_mask ((__v8df) __A, (__v8di) __W, @@ -449,7 +578,7 @@ _mm512_mask_cvttpd_epi64 (__m512i __W, __mmask8 __U, __m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvttpd_epi64 (__mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvttpd2qq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -472,7 +601,7 @@ _mm512_maskz_cvttpd_epi64 (__mmask8 __U, __m512d __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvttpd_epu64 (__m512d __A) { return (__m512i) __builtin_ia32_cvttpd2uqq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -480,7 +609,7 @@ _mm512_cvttpd_epu64 (__m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvttpd_epu64 (__m512i __W, __mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvttpd2uqq512_mask ((__v8df) __A, (__v8di) __W, @@ -488,7 +617,7 @@ _mm512_mask_cvttpd_epu64 (__m512i __W, __mmask8 __U, __m512d __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvttpd_epu64 (__mmask8 __U, __m512d __A) { return (__m512i) __builtin_ia32_cvttpd2uqq512_mask ((__v8df) __A, (__v8di) _mm512_setzero_si512(), @@ -511,7 +640,7 @@ _mm512_maskz_cvttpd_epu64 (__mmask8 __U, __m512d __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvttps_epi64 (__m256 __A) { return (__m512i) __builtin_ia32_cvttps2qq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -519,7 +648,7 @@ _mm512_cvttps_epi64 (__m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvttps_epi64 (__m512i __W, __mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvttps2qq512_mask ((__v8sf) __A, (__v8di) __W, @@ -527,7 +656,7 @@ _mm512_mask_cvttps_epi64 (__m512i __W, __mmask8 __U, __m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvttps_epi64 (__mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvttps2qq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -550,7 +679,7 @@ _mm512_maskz_cvttps_epi64 (__mmask8 __U, __m256 __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_cvttps_epu64 (__m256 __A) { return (__m512i) __builtin_ia32_cvttps2uqq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -558,7 +687,7 @@ _mm512_cvttps_epu64 (__m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_cvttps_epu64 (__m512i __W, __mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvttps2uqq512_mask ((__v8sf) __A, (__v8di) __W, @@ -566,7 +695,7 @@ _mm512_mask_cvttps_epu64 (__m512i __W, __mmask8 __U, __m256 __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_cvttps_epu64 (__mmask8 __U, __m256 __A) { return (__m512i) __builtin_ia32_cvttps2uqq512_mask ((__v8sf) __A, (__v8di) _mm512_setzero_si512(), @@ -589,19 +718,19 @@ _mm512_maskz_cvttps_epu64 (__mmask8 __U, __m256 __A) { (__v8di)_mm512_setzero_si512(), \ (__mmask8)(U), (int)(R)) -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_cvtepu64_pd (__m512i __A) { return (__m512d)__builtin_convertvector((__v8du)__A, __v8df); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_cvtepu64_pd (__m512d __W, __mmask8 __U, __m512i __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_cvtepu64_pd(__A), (__v8df)__W); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtepu64_pd (__mmask8 __U, __m512i __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, (__v8df)_mm512_cvtepu64_pd(__A), @@ -625,7 +754,7 @@ _mm512_maskz_cvtepu64_pd (__mmask8 __U, __m512i __A) { (__mmask8)(U), (int)(R)) -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_cvtepu64_ps (__m512i __A) { return (__m256) __builtin_ia32_cvtuqq2ps512_mask ((__v8di) __A, (__v8sf) _mm256_setzero_ps(), @@ -633,7 +762,7 @@ _mm512_cvtepu64_ps (__m512i __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_mask_cvtepu64_ps (__m256 __W, __mmask8 __U, __m512i __A) { return (__m256) __builtin_ia32_cvtuqq2ps512_mask ((__v8di) __A, (__v8sf) __W, @@ -641,7 +770,7 @@ _mm512_mask_cvtepu64_ps (__m256 __W, __mmask8 __U, __m512i __A) { _MM_FROUND_CUR_DIRECTION); } -static __inline__ __m256 __DEFAULT_FN_ATTRS +static __inline__ __m256 __DEFAULT_FN_ATTRS512 _mm512_maskz_cvtepu64_ps (__mmask8 __U, __m512i __A) { return (__m256) __builtin_ia32_cvtuqq2ps512_mask ((__v8di) __A, (__v8sf) _mm256_setzero_ps(), @@ -935,32 +1064,32 @@ _mm512_maskz_cvtepu64_ps (__mmask8 __U, __m512i __A) { (__v2df)_mm_setzero_pd(), \ (__mmask8)(U), (int)(C), (int)(R)) -static __inline__ __mmask16 __DEFAULT_FN_ATTRS +static __inline__ __mmask16 __DEFAULT_FN_ATTRS512 _mm512_movepi32_mask (__m512i __A) { return (__mmask16) __builtin_ia32_cvtd2mask512 ((__v16si) __A); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_movm_epi32 (__mmask16 __A) { return (__m512i) __builtin_ia32_cvtmask2d512 (__A); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_movm_epi64 (__mmask8 __A) { return (__m512i) __builtin_ia32_cvtmask2q512 (__A); } -static __inline__ __mmask8 __DEFAULT_FN_ATTRS +static __inline__ __mmask8 __DEFAULT_FN_ATTRS512 _mm512_movepi64_mask (__m512i __A) { return (__mmask8) __builtin_ia32_cvtq2mask512 ((__v8di) __A); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_broadcast_f32x2 (__m128 __A) { return (__m512)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A, @@ -968,7 +1097,7 @@ _mm512_broadcast_f32x2 (__m128 __A) 0, 1, 0, 1, 0, 1, 0, 1); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_f32x2 (__m512 __O, __mmask16 __M, __m128 __A) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, @@ -976,7 +1105,7 @@ _mm512_mask_broadcast_f32x2 (__m512 __O, __mmask16 __M, __m128 __A) (__v16sf)__O); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, @@ -984,7 +1113,7 @@ _mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A) (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_broadcast_f32x8(__m256 __A) { return (__m512)__builtin_shufflevector((__v8sf)__A, (__v8sf)__A, @@ -992,7 +1121,7 @@ _mm512_broadcast_f32x8(__m256 __A) 0, 1, 2, 3, 4, 5, 6, 7); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, @@ -1000,7 +1129,7 @@ _mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A) (__v16sf)__O); } -static __inline__ __m512 __DEFAULT_FN_ATTRS +static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A) { return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, @@ -1008,14 +1137,14 @@ _mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A) (__v16sf)_mm512_setzero_ps()); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_broadcast_f64x2(__m128d __A) { return (__m512d)__builtin_shufflevector((__v2df)__A, (__v2df)__A, 0, 1, 0, 1, 0, 1, 0, 1); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M, @@ -1023,7 +1152,7 @@ _mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A) (__v8df)__O); } -static __inline__ __m512d __DEFAULT_FN_ATTRS +static __inline__ __m512d __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) { return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M, @@ -1031,7 +1160,7 @@ _mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) (__v8df)_mm512_setzero_pd()); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_broadcast_i32x2 (__m128i __A) { return (__m512i)__builtin_shufflevector((__v4si)__A, (__v4si)__A, @@ -1039,7 +1168,7 @@ _mm512_broadcast_i32x2 (__m128i __A) 0, 1, 0, 1, 0, 1, 0, 1); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_i32x2 (__m512i __O, __mmask16 __M, __m128i __A) { return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, @@ -1047,7 +1176,7 @@ _mm512_mask_broadcast_i32x2 (__m512i __O, __mmask16 __M, __m128i __A) (__v16si)__O); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A) { return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, @@ -1055,7 +1184,7 @@ _mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A) (__v16si)_mm512_setzero_si512()); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_broadcast_i32x8(__m256i __A) { return (__m512i)__builtin_shufflevector((__v8si)__A, (__v8si)__A, @@ -1063,7 +1192,7 @@ _mm512_broadcast_i32x8(__m256i __A) 0, 1, 2, 3, 4, 5, 6, 7); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A) { return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, @@ -1071,7 +1200,7 @@ _mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A) (__v16si)__O); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A) { return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, @@ -1079,14 +1208,14 @@ _mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A) (__v16si)_mm512_setzero_si512()); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_broadcast_i64x2(__m128i __A) { return (__m512i)__builtin_shufflevector((__v2di)__A, (__v2di)__A, 0, 1, 0, 1, 0, 1, 0, 1); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A) { return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M, @@ -1094,7 +1223,7 @@ _mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A) (__v8di)__O); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS512 _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) { return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M, @@ -1256,6 +1385,7 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) (__mmask8)__builtin_ia32_fpclassss_mask((__v4sf)(__m128)(A), (int)(imm), \ (__mmask8)(U)) +#undef __DEFAULT_FN_ATTRS512 #undef __DEFAULT_FN_ATTRS #endif |