1*67e74705SXin Li /*===------------- avx512cdintrin.h - AVX512CD intrinsics ------------------===
2*67e74705SXin Li *
3*67e74705SXin Li *
4*67e74705SXin Li * Permission is hereby granted, free of charge, to any person obtaining a copy
5*67e74705SXin Li * of this software and associated documentation files (the "Software"), to deal
6*67e74705SXin Li * in the Software without restriction, including without limitation the rights
7*67e74705SXin Li * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
8*67e74705SXin Li * copies of the Software, and to permit persons to whom the Software is
9*67e74705SXin Li * furnished to do so, subject to the following conditions:
10*67e74705SXin Li *
11*67e74705SXin Li * The above copyright notice and this permission notice shall be included in
12*67e74705SXin Li * all copies or substantial portions of the Software.
13*67e74705SXin Li *
14*67e74705SXin Li * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15*67e74705SXin Li * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16*67e74705SXin Li * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
17*67e74705SXin Li * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18*67e74705SXin Li * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
19*67e74705SXin Li * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
20*67e74705SXin Li * THE SOFTWARE.
21*67e74705SXin Li *
22*67e74705SXin Li *===-----------------------------------------------------------------------===
23*67e74705SXin Li */
24*67e74705SXin Li #ifndef __IMMINTRIN_H
25*67e74705SXin Li #error "Never use <avx512cdintrin.h> directly; include <immintrin.h> instead."
26*67e74705SXin Li #endif
27*67e74705SXin Li
28*67e74705SXin Li #ifndef __AVX512CDINTRIN_H
29*67e74705SXin Li #define __AVX512CDINTRIN_H
30*67e74705SXin Li
31*67e74705SXin Li /* Define the default attributes for the functions in this file. */
32*67e74705SXin Li #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512cd")))
33*67e74705SXin Li
34*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_conflict_epi64(__m512i __A)35*67e74705SXin Li _mm512_conflict_epi64 (__m512i __A)
36*67e74705SXin Li {
37*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
38*67e74705SXin Li (__v8di) _mm512_setzero_si512 (),
39*67e74705SXin Li (__mmask8) -1);
40*67e74705SXin Li }
41*67e74705SXin Li
42*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_conflict_epi64(__m512i __W,__mmask8 __U,__m512i __A)43*67e74705SXin Li _mm512_mask_conflict_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
44*67e74705SXin Li {
45*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
46*67e74705SXin Li (__v8di) __W,
47*67e74705SXin Li (__mmask8) __U);
48*67e74705SXin Li }
49*67e74705SXin Li
50*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_conflict_epi64(__mmask8 __U,__m512i __A)51*67e74705SXin Li _mm512_maskz_conflict_epi64 (__mmask8 __U, __m512i __A)
52*67e74705SXin Li {
53*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
54*67e74705SXin Li (__v8di) _mm512_setzero_si512 (),
55*67e74705SXin Li (__mmask8) __U);
56*67e74705SXin Li }
57*67e74705SXin Li
58*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_conflict_epi32(__m512i __A)59*67e74705SXin Li _mm512_conflict_epi32 (__m512i __A)
60*67e74705SXin Li {
61*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
62*67e74705SXin Li (__v16si) _mm512_setzero_si512 (),
63*67e74705SXin Li (__mmask16) -1);
64*67e74705SXin Li }
65*67e74705SXin Li
66*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_conflict_epi32(__m512i __W,__mmask16 __U,__m512i __A)67*67e74705SXin Li _mm512_mask_conflict_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
68*67e74705SXin Li {
69*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
70*67e74705SXin Li (__v16si) __W,
71*67e74705SXin Li (__mmask16) __U);
72*67e74705SXin Li }
73*67e74705SXin Li
74*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_conflict_epi32(__mmask16 __U,__m512i __A)75*67e74705SXin Li _mm512_maskz_conflict_epi32 (__mmask16 __U, __m512i __A)
76*67e74705SXin Li {
77*67e74705SXin Li return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
78*67e74705SXin Li (__v16si) _mm512_setzero_si512 (),
79*67e74705SXin Li (__mmask16) __U);
80*67e74705SXin Li }
81*67e74705SXin Li
82*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_lzcnt_epi32(__m512i __A)83*67e74705SXin Li _mm512_lzcnt_epi32 (__m512i __A)
84*67e74705SXin Li {
85*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
86*67e74705SXin Li (__v16si) _mm512_setzero_si512 (),
87*67e74705SXin Li (__mmask16) -1);
88*67e74705SXin Li }
89*67e74705SXin Li
90*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_lzcnt_epi32(__m512i __W,__mmask16 __U,__m512i __A)91*67e74705SXin Li _mm512_mask_lzcnt_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
92*67e74705SXin Li {
93*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
94*67e74705SXin Li (__v16si) __W,
95*67e74705SXin Li (__mmask16) __U);
96*67e74705SXin Li }
97*67e74705SXin Li
98*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_lzcnt_epi32(__mmask16 __U,__m512i __A)99*67e74705SXin Li _mm512_maskz_lzcnt_epi32 (__mmask16 __U, __m512i __A)
100*67e74705SXin Li {
101*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
102*67e74705SXin Li (__v16si) _mm512_setzero_si512 (),
103*67e74705SXin Li (__mmask16) __U);
104*67e74705SXin Li }
105*67e74705SXin Li
106*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_lzcnt_epi64(__m512i __A)107*67e74705SXin Li _mm512_lzcnt_epi64 (__m512i __A)
108*67e74705SXin Li {
109*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
110*67e74705SXin Li (__v8di) _mm512_setzero_si512 (),
111*67e74705SXin Li (__mmask8) -1);
112*67e74705SXin Li }
113*67e74705SXin Li
114*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_lzcnt_epi64(__m512i __W,__mmask8 __U,__m512i __A)115*67e74705SXin Li _mm512_mask_lzcnt_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
116*67e74705SXin Li {
117*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
118*67e74705SXin Li (__v8di) __W,
119*67e74705SXin Li (__mmask8) __U);
120*67e74705SXin Li }
121*67e74705SXin Li
122*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_lzcnt_epi64(__mmask8 __U,__m512i __A)123*67e74705SXin Li _mm512_maskz_lzcnt_epi64 (__mmask8 __U, __m512i __A)
124*67e74705SXin Li {
125*67e74705SXin Li return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
126*67e74705SXin Li (__v8di) _mm512_setzero_si512 (),
127*67e74705SXin Li (__mmask8) __U);
128*67e74705SXin Li }
129*67e74705SXin Li
130*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_broadcastmb_epi64(__mmask8 __A)131*67e74705SXin Li _mm512_broadcastmb_epi64 (__mmask8 __A)
132*67e74705SXin Li {
133*67e74705SXin Li return (__m512i) __builtin_ia32_broadcastmb512 (__A);
134*67e74705SXin Li }
135*67e74705SXin Li
136*67e74705SXin Li static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_broadcastmw_epi32(__mmask16 __A)137*67e74705SXin Li _mm512_broadcastmw_epi32 (__mmask16 __A)
138*67e74705SXin Li {
139*67e74705SXin Li return (__m512i) __builtin_ia32_broadcastmw512 (__A);
140*67e74705SXin Li }
141*67e74705SXin Li
142*67e74705SXin Li #undef __DEFAULT_FN_ATTRS
143*67e74705SXin Li
144*67e74705SXin Li #endif
145