clang 22.0.0git
avx512ifmavlintrin.h
Go to the documentation of this file.
1/*===------------- avx512ifmavlintrin.h - IFMA intrinsics ------------------===
2 *
3 *
4 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5 * See https://llvm.org/LICENSE.txt for license information.
6 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 *
8 *===-----------------------------------------------------------------------===
9 */
10#ifndef __IMMINTRIN_H
11#error \
12 "Never use <avx512ifmavlintrin.h> directly; include <immintrin.h> instead."
13#endif
14
15#ifndef __IFMAVLINTRIN_H
16#define __IFMAVLINTRIN_H
17
18/* Define the default attributes for the functions in this file. */
19#if defined(__cplusplus) && (__cplusplus >= 201103L)
20#define __DEFAULT_FN_ATTRS128 \
21 constexpr __attribute__((__always_inline__, __nodebug__, \
22 __target__("avx512ifma,avx512vl"), \
23 __min_vector_width__(128)))
24#define __DEFAULT_FN_ATTRS256 \
25 constexpr __attribute__((__always_inline__, __nodebug__, \
26 __target__("avx512ifma,avx512vl"), \
27 __min_vector_width__(256)))
28#else
29#define __DEFAULT_FN_ATTRS128 \
30 __attribute__((__always_inline__, __nodebug__, \
31 __target__("avx512ifma,avx512vl"), \
32 __min_vector_width__(128)))
33#define __DEFAULT_FN_ATTRS256 \
34 __attribute__((__always_inline__, __nodebug__, \
35 __target__("avx512ifma,avx512vl"), \
36 __min_vector_width__(256)))
37
38#endif
39
40#if !(defined(__AVXIFMA__) || defined(__AVX512IFMA__))
41#define _mm_madd52hi_epu64(X, Y, Z) \
42 ((__m128i)__builtin_ia32_vpmadd52huq128((__v2di)(X), (__v2di)(Y), \
43 (__v2di)(Z)))
44
45#define _mm256_madd52hi_epu64(X, Y, Z) \
46 ((__m256i)__builtin_ia32_vpmadd52huq256((__v4di)(X), (__v4di)(Y), \
47 (__v4di)(Z)))
48
49#define _mm_madd52lo_epu64(X, Y, Z) \
50 ((__m128i)__builtin_ia32_vpmadd52luq128((__v2di)(X), (__v2di)(Y), \
51 (__v2di)(Z)))
52
53#define _mm256_madd52lo_epu64(X, Y, Z) \
54 ((__m256i)__builtin_ia32_vpmadd52luq256((__v4di)(X), (__v4di)(Y), \
55 (__v4di)(Z)))
56#endif
57
58#if defined(__AVX512IFMA__)
59static __inline__ __m128i __DEFAULT_FN_ATTRS128
60_mm_madd52hi_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
61 return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di)__X, (__v2di)__Y,
62 (__v2di)__Z);
63}
64
65static __inline__ __m256i __DEFAULT_FN_ATTRS256
66_mm256_madd52hi_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
67 return (__m256i)__builtin_ia32_vpmadd52huq256((__v4di)__X, (__v4di)__Y,
68 (__v4di)__Z);
69}
70
71static __inline__ __m128i __DEFAULT_FN_ATTRS128
72_mm_madd52lo_epu64(__m128i __X, __m128i __Y, __m128i __Z) {
73 return (__m128i)__builtin_ia32_vpmadd52luq128((__v2di)__X, (__v2di)__Y,
74 (__v2di)__Z);
75}
76
77static __inline__ __m256i __DEFAULT_FN_ATTRS256
78_mm256_madd52lo_epu64(__m256i __X, __m256i __Y, __m256i __Z) {
79 return (__m256i)__builtin_ia32_vpmadd52luq256((__v4di)__X, (__v4di)__Y,
80 (__v4di)__Z);
81}
82#endif
83
84static __inline__ __m128i __DEFAULT_FN_ATTRS128
85_mm_mask_madd52hi_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) {
86 return (__m128i)__builtin_ia32_selectq_128(
87 __M, (__v2di)__builtin_ia32_vpmadd52huq128(__W, __X, __Y), (__v2di)__W);
88}
89
90static __inline__ __m128i __DEFAULT_FN_ATTRS128
91_mm_maskz_madd52hi_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) {
92 return (__m128i)__builtin_ia32_selectq_128(
93 __M, (__v2di)__builtin_ia32_vpmadd52huq128(__X, __Y, __Z),
94 (__v2di)_mm_setzero_si128());
95}
96
98 __m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) {
99 return (__m256i)__builtin_ia32_selectq_256(
100 __M, (__v4di)__builtin_ia32_vpmadd52huq256(__W, __X, __Y), (__v4di)__W);
101}
102
104 __mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) {
105 return (__m256i)__builtin_ia32_selectq_256(
106 __M, (__v4di)__builtin_ia32_vpmadd52huq256(__X, __Y, __Z),
107 (__v4di)_mm256_setzero_si256());
108}
109
110static __inline__ __m128i __DEFAULT_FN_ATTRS128
111_mm_mask_madd52lo_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) {
112 return (__m128i)__builtin_ia32_selectq_128(
113 __M, (__v2di)__builtin_ia32_vpmadd52luq128(__W, __X, __Y), (__v2di)__W);
114}
115
116static __inline__ __m128i __DEFAULT_FN_ATTRS128
117_mm_maskz_madd52lo_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) {
118 return (__m128i)__builtin_ia32_selectq_128(
119 __M, (__v2di)__builtin_ia32_vpmadd52luq128(__X, __Y, __Z),
120 (__v2di)_mm_setzero_si128());
121}
122
124 __m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) {
125 return (__m256i)__builtin_ia32_selectq_256(
126 __M, (__v4di)__builtin_ia32_vpmadd52luq256(__W, __X, __Y), (__v4di)__W);
127}
128
130 __mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) {
131 return (__m256i)__builtin_ia32_selectq_256(
132 __M, (__v4di)__builtin_ia32_vpmadd52luq256(__X, __Y, __Z),
133 (__v4di)_mm256_setzero_si256());
134}
135
136#undef __DEFAULT_FN_ATTRS128
137#undef __DEFAULT_FN_ATTRS256
138
139#endif
#define __DEFAULT_FN_ATTRS128
#define __DEFAULT_FN_ATTRS256
unsigned char __mmask8
#define _mm256_madd52lo_epu64(X, Y, Z)
static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_mask_madd52hi_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_madd52lo_epu64(__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
#define _mm_madd52hi_epu64(X, Y, Z)
static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_maskz_madd52lo_epu64(__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_mask_madd52lo_epu64(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
#define _mm_madd52lo_epu64(X, Y, Z)
static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_maskz_madd52lo_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
#define _mm256_madd52hi_epu64(X, Y, Z)
static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_maskz_madd52hi_epu64(__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_maskz_madd52hi_epu64(__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_madd52hi_epu64(__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
static __inline __m256i __DEFAULT_FN_ATTRS_CONSTEXPR _mm256_setzero_si256(void)
Constructs a 256-bit integer vector initialized to zero.
Definition avxintrin.h:4316
static __inline__ __m128i __DEFAULT_FN_ATTRS_CONSTEXPR _mm_setzero_si128(void)
Creates a 128-bit integer vector initialized to zero.
Definition emmintrin.h:3878
__inline unsigned int unsigned int __Y
Definition bmi2intrin.h:19