1*7330f729Sjoerg /*===------------- avx512ifmaintrin.h - IFMA intrinsics ------------------===
2*7330f729Sjoerg *
3*7330f729Sjoerg *
4*7330f729Sjoerg * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5*7330f729Sjoerg * See https://llvm.org/LICENSE.txt for license information.
6*7330f729Sjoerg * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7*7330f729Sjoerg *
8*7330f729Sjoerg *===-----------------------------------------------------------------------===
9*7330f729Sjoerg */
10*7330f729Sjoerg #ifndef __IMMINTRIN_H
11*7330f729Sjoerg #error "Never use <avx512ifmaintrin.h> directly; include <immintrin.h> instead."
12*7330f729Sjoerg #endif
13*7330f729Sjoerg
14*7330f729Sjoerg #ifndef __IFMAINTRIN_H
15*7330f729Sjoerg #define __IFMAINTRIN_H
16*7330f729Sjoerg
17*7330f729Sjoerg /* Define the default attributes for the functions in this file. */
18*7330f729Sjoerg #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma"), __min_vector_width__(512)))
19*7330f729Sjoerg
20*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_madd52hi_epu64(__m512i __X,__m512i __Y,__m512i __Z)21*7330f729Sjoerg _mm512_madd52hi_epu64 (__m512i __X, __m512i __Y, __m512i __Z)
22*7330f729Sjoerg {
23*7330f729Sjoerg return (__m512i)__builtin_ia32_vpmadd52huq512((__v8di) __X, (__v8di) __Y,
24*7330f729Sjoerg (__v8di) __Z);
25*7330f729Sjoerg }
26*7330f729Sjoerg
27*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_madd52hi_epu64(__m512i __W,__mmask8 __M,__m512i __X,__m512i __Y)28*7330f729Sjoerg _mm512_mask_madd52hi_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
29*7330f729Sjoerg {
30*7330f729Sjoerg return (__m512i)__builtin_ia32_selectq_512(__M,
31*7330f729Sjoerg (__v8di)_mm512_madd52hi_epu64(__W, __X, __Y),
32*7330f729Sjoerg (__v8di)__W);
33*7330f729Sjoerg }
34*7330f729Sjoerg
35*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_madd52hi_epu64(__mmask8 __M,__m512i __X,__m512i __Y,__m512i __Z)36*7330f729Sjoerg _mm512_maskz_madd52hi_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z)
37*7330f729Sjoerg {
38*7330f729Sjoerg return (__m512i)__builtin_ia32_selectq_512(__M,
39*7330f729Sjoerg (__v8di)_mm512_madd52hi_epu64(__X, __Y, __Z),
40*7330f729Sjoerg (__v8di)_mm512_setzero_si512());
41*7330f729Sjoerg }
42*7330f729Sjoerg
43*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_madd52lo_epu64(__m512i __X,__m512i __Y,__m512i __Z)44*7330f729Sjoerg _mm512_madd52lo_epu64 (__m512i __X, __m512i __Y, __m512i __Z)
45*7330f729Sjoerg {
46*7330f729Sjoerg return (__m512i)__builtin_ia32_vpmadd52luq512((__v8di) __X, (__v8di) __Y,
47*7330f729Sjoerg (__v8di) __Z);
48*7330f729Sjoerg }
49*7330f729Sjoerg
50*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_madd52lo_epu64(__m512i __W,__mmask8 __M,__m512i __X,__m512i __Y)51*7330f729Sjoerg _mm512_mask_madd52lo_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
52*7330f729Sjoerg {
53*7330f729Sjoerg return (__m512i)__builtin_ia32_selectq_512(__M,
54*7330f729Sjoerg (__v8di)_mm512_madd52lo_epu64(__W, __X, __Y),
55*7330f729Sjoerg (__v8di)__W);
56*7330f729Sjoerg }
57*7330f729Sjoerg
58*7330f729Sjoerg static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_madd52lo_epu64(__mmask8 __M,__m512i __X,__m512i __Y,__m512i __Z)59*7330f729Sjoerg _mm512_maskz_madd52lo_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z)
60*7330f729Sjoerg {
61*7330f729Sjoerg return (__m512i)__builtin_ia32_selectq_512(__M,
62*7330f729Sjoerg (__v8di)_mm512_madd52lo_epu64(__X, __Y, __Z),
63*7330f729Sjoerg (__v8di)_mm512_setzero_si512());
64*7330f729Sjoerg }
65*7330f729Sjoerg
66*7330f729Sjoerg #undef __DEFAULT_FN_ATTRS
67*7330f729Sjoerg
68*7330f729Sjoerg #endif
69