summaryrefslogtreecommitdiff
path: root/lib/Headers/avx512fintrin.h
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Headers/avx512fintrin.h')
-rw-r--r--lib/Headers/avx512fintrin.h189
1 files changed, 117 insertions, 72 deletions
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index e6a7217c8967..d8535f765889 100644
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -4229,6 +4229,18 @@ _mm512_maskz_cvtpd_epu32 (__mmask8 __U, __m512d __A)
_MM_FROUND_CUR_DIRECTION);
}
+static __inline__ double __DEFAULT_FN_ATTRS
+_mm512_cvtsd_f64(__m512d __a)
+{
+ return __a[0];
+}
+
+static __inline__ float __DEFAULT_FN_ATTRS
+_mm512_cvtss_f32(__m512 __a)
+{
+ return __a[0];
+}
+
/* Unpack and Interleave */
static __inline __m512d __DEFAULT_FN_ATTRS
@@ -4540,7 +4552,7 @@ _mm512_maskz_loadu_pd(__mmask8 __U, void const *__P)
}
static __inline __m512d __DEFAULT_FN_ATTRS
-_mm512_loadu_pd(double const *__p)
+_mm512_loadu_pd(void const *__p)
{
struct __loadu_pd {
__m512d __v;
@@ -4549,7 +4561,7 @@ _mm512_loadu_pd(double const *__p)
}
static __inline __m512 __DEFAULT_FN_ATTRS
-_mm512_loadu_ps(float const *__p)
+_mm512_loadu_ps(void const *__p)
{
struct __loadu_ps {
__m512 __v;
@@ -4558,7 +4570,7 @@ _mm512_loadu_ps(float const *__p)
}
static __inline __m512 __DEFAULT_FN_ATTRS
-_mm512_load_ps(float const *__p)
+_mm512_load_ps(void const *__p)
{
return (__m512) __builtin_ia32_loadaps512_mask ((const __v16sf *)__p,
(__v16sf)
@@ -4584,7 +4596,7 @@ _mm512_maskz_load_ps(__mmask16 __U, void const *__P)
}
static __inline __m512d __DEFAULT_FN_ATTRS
-_mm512_load_pd(double const *__p)
+_mm512_load_pd(void const *__p)
{
return (__m512d) __builtin_ia32_loadapd512_mask ((const __v8df *)__p,
(__v8df)
@@ -7278,107 +7290,97 @@ _mm_maskz_sqrt_ss (__mmask8 __U, __m128 __A, __m128 __B)
(__mmask8)(U), (int)(R)); })
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_broadcast_f32x4 (__m128 __A)
+_mm512_broadcast_f32x4(__m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf)
- _mm512_undefined_ps (),
- (__mmask16) -1);
+ return (__m512)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f32x4 (__m512 __O, __mmask16 __M, __m128 __A)
+_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, __m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf) __O,
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+ (__v16sf)_mm512_broadcast_f32x4(__A),
+ (__v16sf)__O);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f32x4 (__mmask16 __M, __m128 __A)
+_mm512_maskz_broadcast_f32x4(__mmask16 __M, __m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf)
- _mm512_setzero_ps (),
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+ (__v16sf)_mm512_broadcast_f32x4(__A),
+ (__v16sf)_mm512_setzero_ps());
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_broadcast_f64x4 (__m256d __A)
+_mm512_broadcast_f64x4(__m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df)
- _mm512_undefined_pd (),
- (__mmask8) -1);
+ return (__m512d)__builtin_shufflevector((__v4df)__A, (__v4df)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f64x4 (__m512d __O, __mmask8 __M, __m256d __A)
+_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, __m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df) __O,
- __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x4(__A),
+ (__v8df)__O);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f64x4 (__mmask8 __M, __m256d __A)
+_mm512_maskz_broadcast_f64x4(__mmask8 __M, __m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df)
- _mm512_setzero_pd (),
- __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x4(__A),
+ (__v8df)_mm512_setzero_pd());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i32x4 (__m128i __A)
+_mm512_broadcast_i32x4(__m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si)
- _mm512_undefined_epi32 (),
- (__mmask16) -1);
+ return (__m512i)__builtin_shufflevector((__v4si)__A, (__v4si)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i32x4 (__m512i __O, __mmask16 __M, __m128i __A)
+_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si) __O,
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+ (__v16si)_mm512_broadcast_i32x4(__A),
+ (__v16si)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i32x4 (__mmask16 __M, __m128i __A)
+_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si)
- _mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+ (__v16si)_mm512_broadcast_i32x4(__A),
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i64x4 (__m256i __A)
+_mm512_broadcast_i64x4(__m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di)
- _mm512_undefined_epi32 (),
- (__mmask8) -1);
+ return (__m512i)__builtin_shufflevector((__v4di)__A, (__v4di)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i64x4 (__m512i __O, __mmask8 __M, __m256i __A)
+_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di) __O,
- __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x4(__A),
+ (__v8di)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i64x4 (__mmask8 __M, __m256i __A)
+_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di)
- _mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x4(__A),
+ (__v8di)_mm512_setzero_si512());
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
@@ -7860,12 +7862,12 @@ _mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
3 + ((imm) & 0x3) * 4); })
#define _mm512_mask_extracti32x4_epi32(W, U, A, imm) __extension__ ({ \
- (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, \
+ (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
(__v4si)_mm512_extracti32x4_epi32((A), (imm)), \
- (__v4si)__W); })
+ (__v4si)(W)); })
#define _mm512_maskz_extracti32x4_epi32(U, A, imm) __extension__ ({ \
- (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, \
+ (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
(__v4si)_mm512_extracti32x4_epi32((A), (imm)), \
(__v4si)_mm_setzero_si128()); })
@@ -7878,12 +7880,12 @@ _mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
((imm) & 1) ? 7 : 3); })
#define _mm512_mask_extracti64x4_epi64(W, U, A, imm) __extension__ ({ \
- (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
(__v4di)_mm512_extracti64x4_epi64((A), (imm)), \
- (__v4di)__W); })
+ (__v4di)(W)); })
#define _mm512_maskz_extracti64x4_epi64(U, A, imm) __extension__ ({ \
- (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
(__v4di)_mm512_extracti64x4_epi64((A), (imm)), \
(__v4di)_mm256_setzero_si256()); })
@@ -8159,11 +8161,11 @@ _mm512_maskz_getexp_ps (__mmask16 __U, __m512 __A)
(__v8di)(__m512i)(index), (__mmask8)-1, \
(int)(scale)); })
-#define _mm512_mask_i64gather_ps( __v1_old, __mask, __index,\
- __addr, __scale) __extension__({\
-__builtin_ia32_gatherdiv16sf ((__v8sf) __v1_old,\
- __addr,(__v8di) __index, __mask, __scale);\
-})
+#define _mm512_mask_i64gather_ps(v1_old, mask, index, addr, scale) __extension__({\
+ (__m256)__builtin_ia32_gatherdiv16sf((__v8sf)(__m256)(v1_old),\
+ (float const *)(addr), \
+ (__v8di)(__m512i)(index), \
+ (__mmask8)(mask), (int)(scale)); })
#define _mm512_i64gather_epi32(index, addr, scale) __extension__ ({\
(__m256i)__builtin_ia32_gatherdiv16si((__v8si)_mm256_undefined_ps(), \
@@ -8858,6 +8860,8 @@ _mm512_permutexvar_epi32 (__m512i __X, __m512i __Y)
(__mmask16) -1);
}
+#define _mm512_permutevar_epi32 _mm512_permutexvar_epi32
+
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X,
__m512i __Y)
@@ -8868,6 +8872,8 @@ _mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X,
__M);
}
+#define _mm512_mask_permutevar_epi32 _mm512_mask_permutexvar_epi32
+
static __inline__ __mmask16 __DEFAULT_FN_ATTRS
_mm512_kand (__mmask16 __A, __mmask16 __B)
{
@@ -8925,7 +8931,7 @@ _mm512_stream_si512 (__m512i * __P, __m512i __A)
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_stream_load_si512 (void *__P)
{
- return __builtin_ia32_movntdqa512 ((__v8di *)__P);
+ return (__m512i) __builtin_nontemporal_load((const __v8di *)__P);
}
static __inline__ void __DEFAULT_FN_ATTRS
@@ -9635,6 +9641,45 @@ _mm512_mask_set1_epi64 (__m512i __O, __mmask8 __M, long long __A)
}
#endif
+static __inline __m512i __DEFAULT_FN_ATTRS
+_mm512_set_epi8 (char __e63, char __e62, char __e61, char __e60, char __e59,
+ char __e58, char __e57, char __e56, char __e55, char __e54, char __e53,
+ char __e52, char __e51, char __e50, char __e49, char __e48, char __e47,
+ char __e46, char __e45, char __e44, char __e43, char __e42, char __e41,
+ char __e40, char __e39, char __e38, char __e37, char __e36, char __e35,
+ char __e34, char __e33, char __e32, char __e31, char __e30, char __e29,
+ char __e28, char __e27, char __e26, char __e25, char __e24, char __e23,
+ char __e22, char __e21, char __e20, char __e19, char __e18, char __e17,
+ char __e16, char __e15, char __e14, char __e13, char __e12, char __e11,
+ char __e10, char __e9, char __e8, char __e7, char __e6, char __e5,
+ char __e4, char __e3, char __e2, char __e1, char __e0) {
+
+ return __extension__ (__m512i)(__v64qi)
+ {__e0, __e1, __e2, __e3, __e4, __e5, __e6, __e7,
+ __e8, __e9, __e10, __e11, __e12, __e13, __e14, __e15,
+ __e16, __e17, __e18, __e19, __e20, __e21, __e22, __e23,
+ __e24, __e25, __e26, __e27, __e28, __e29, __e30, __e31,
+ __e32, __e33, __e34, __e35, __e36, __e37, __e38, __e39,
+ __e40, __e41, __e42, __e43, __e44, __e45, __e46, __e47,
+ __e48, __e49, __e50, __e51, __e52, __e53, __e54, __e55,
+ __e56, __e57, __e58, __e59, __e60, __e61, __e62, __e63};
+}
+
+static __inline __m512i __DEFAULT_FN_ATTRS
+_mm512_set_epi16(short __e31, short __e30, short __e29, short __e28,
+ short __e27, short __e26, short __e25, short __e24, short __e23,
+ short __e22, short __e21, short __e20, short __e19, short __e18,
+ short __e17, short __e16, short __e15, short __e14, short __e13,
+ short __e12, short __e11, short __e10, short __e9, short __e8,
+ short __e7, short __e6, short __e5, short __e4, short __e3,
+ short __e2, short __e1, short __e0) {
+ return __extension__ (__m512i)(__v32hi)
+ {__e0, __e1, __e2, __e3, __e4, __e5, __e6, __e7,
+ __e8, __e9, __e10, __e11, __e12, __e13, __e14, __e15,
+ __e16, __e17, __e18, __e19, __e20, __e21, __e22, __e23,
+ __e24, __e25, __e26, __e27, __e28, __e29, __e30, __e31 };
+}
+
static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_set_epi32 (int __A, int __B, int __C, int __D,
int __E, int __F, int __G, int __H,
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-11 14:53:06 +0000