| 1 |
691 |
jeremybenn |
/* { dg-do compile } */
|
| 2 |
|
|
/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8" } */
|
| 3 |
|
|
|
| 4 |
|
|
#include <mm_malloc.h>
|
| 5 |
|
|
|
| 6 |
|
|
/* Test that the intrinsics compile with optimization. All of them
|
| 7 |
|
|
are defined as inline functions in {,x,e,p,t,s,w,a,b,i}mmintrin.h,
|
| 8 |
|
|
mm3dnow.h, fma4intrin.h, xopintrin.h, abmintrin.h, bmiintrin.h,
|
| 9 |
|
|
tbmintrin.h, lwpintrin.h, popcntintrin.h, fmaintrin.h and mm_malloc.h
|
| 10 |
|
|
that reference the proper builtin functions.
|
| 11 |
|
|
|
| 12 |
|
|
Defining away "extern" and "__inline" results in all of them being
|
| 13 |
|
|
compiled as proper functions. */
|
| 14 |
|
|
|
| 15 |
|
|
#define extern
|
| 16 |
|
|
#define __inline
|
| 17 |
|
|
|
| 18 |
|
|
/* Following intrinsics require immediate arguments. */
|
| 19 |
|
|
|
| 20 |
|
|
/* ammintrin.h */
|
| 21 |
|
|
#define __builtin_ia32_extrqi(X, I, L) __builtin_ia32_extrqi(X, 1, 1)
|
| 22 |
|
|
#define __builtin_ia32_insertqi(X, Y, I, L) __builtin_ia32_insertqi(X, Y, 1, 1)
|
| 23 |
|
|
|
| 24 |
|
|
/* wmmintrin.h */
|
| 25 |
|
|
#define __builtin_ia32_aeskeygenassist128(X, C) __builtin_ia32_aeskeygenassist128(X, 1)
|
| 26 |
|
|
#define __builtin_ia32_pclmulqdq128(X, Y, I) __builtin_ia32_pclmulqdq128(X, Y, 1)
|
| 27 |
|
|
|
| 28 |
|
|
/* smmintrin.h */
|
| 29 |
|
|
#define __builtin_ia32_roundpd(V, M) __builtin_ia32_roundpd(V, 1)
|
| 30 |
|
|
#define __builtin_ia32_roundsd(D, V, M) __builtin_ia32_roundsd(D, V, 1)
|
| 31 |
|
|
#define __builtin_ia32_roundps(V, M) __builtin_ia32_roundps(V, 1)
|
| 32 |
|
|
#define __builtin_ia32_roundss(D, V, M) __builtin_ia32_roundss(D, V, 1)
|
| 33 |
|
|
|
| 34 |
|
|
#define __builtin_ia32_pblendw128(X, Y, M) __builtin_ia32_pblendw128 (X, Y, 1)
|
| 35 |
|
|
#define __builtin_ia32_blendps(X, Y, M) __builtin_ia32_blendps(X, Y, 1)
|
| 36 |
|
|
#define __builtin_ia32_blendpd(X, Y, M) __builtin_ia32_blendpd(X, Y, 1)
|
| 37 |
|
|
#define __builtin_ia32_dpps(X, Y, M) __builtin_ia32_dpps(X, Y, 1)
|
| 38 |
|
|
#define __builtin_ia32_dppd(X, Y, M) __builtin_ia32_dppd(X, Y, 1)
|
| 39 |
|
|
#define __builtin_ia32_insertps128(D, S, N) __builtin_ia32_insertps128(D, S, 1)
|
| 40 |
|
|
#define __builtin_ia32_vec_ext_v4sf(X, N) __builtin_ia32_vec_ext_v4sf(X, 1)
|
| 41 |
|
|
#define __builtin_ia32_vec_set_v16qi(D, S, N) __builtin_ia32_vec_set_v16qi(D, S, 1)
|
| 42 |
|
|
#define __builtin_ia32_vec_set_v4si(D, S, N) __builtin_ia32_vec_set_v4si(D, S, 1)
|
| 43 |
|
|
#define __builtin_ia32_vec_set_v2di(D, S, N) __builtin_ia32_vec_set_v2di(D, S, 1)
|
| 44 |
|
|
#define __builtin_ia32_vec_ext_v16qi(X, N) __builtin_ia32_vec_ext_v16qi(X, 1)
|
| 45 |
|
|
#define __builtin_ia32_vec_ext_v4si(X, N) __builtin_ia32_vec_ext_v4si(X, 1)
|
| 46 |
|
|
#define __builtin_ia32_vec_ext_v2di(X, N) __builtin_ia32_vec_ext_v2di(X, 1)
|
| 47 |
|
|
#define __builtin_ia32_mpsadbw128(X, Y, M) __builtin_ia32_mpsadbw128(X, Y, 1)
|
| 48 |
|
|
#define __builtin_ia32_pcmpistrm128(X, Y, M) \
|
| 49 |
|
|
__builtin_ia32_pcmpistrm128(X, Y, 1)
|
| 50 |
|
|
#define __builtin_ia32_pcmpistri128(X, Y, M) \
|
| 51 |
|
|
__builtin_ia32_pcmpistri128(X, Y, 1)
|
| 52 |
|
|
#define __builtin_ia32_pcmpestrm128(X, LX, Y, LY, M) \
|
| 53 |
|
|
__builtin_ia32_pcmpestrm128(X, LX, Y, LY, 1)
|
| 54 |
|
|
#define __builtin_ia32_pcmpestri128(X, LX, Y, LY, M) \
|
| 55 |
|
|
__builtin_ia32_pcmpestri128(X, LX, Y, LY, 1)
|
| 56 |
|
|
#define __builtin_ia32_pcmpistria128(X, Y, M) \
|
| 57 |
|
|
__builtin_ia32_pcmpistria128(X, Y, 1)
|
| 58 |
|
|
#define __builtin_ia32_pcmpistric128(X, Y, M) \
|
| 59 |
|
|
__builtin_ia32_pcmpistric128(X, Y, 1)
|
| 60 |
|
|
#define __builtin_ia32_pcmpistrio128(X, Y, M) \
|
| 61 |
|
|
__builtin_ia32_pcmpistrio128(X, Y, 1)
|
| 62 |
|
|
#define __builtin_ia32_pcmpistris128(X, Y, M) \
|
| 63 |
|
|
__builtin_ia32_pcmpistris128(X, Y, 1)
|
| 64 |
|
|
#define __builtin_ia32_pcmpistriz128(X, Y, M) \
|
| 65 |
|
|
__builtin_ia32_pcmpistriz128(X, Y, 1)
|
| 66 |
|
|
#define __builtin_ia32_pcmpestria128(X, LX, Y, LY, M) \
|
| 67 |
|
|
__builtin_ia32_pcmpestria128(X, LX, Y, LY, 1)
|
| 68 |
|
|
#define __builtin_ia32_pcmpestric128(X, LX, Y, LY, M) \
|
| 69 |
|
|
__builtin_ia32_pcmpestric128(X, LX, Y, LY, 1)
|
| 70 |
|
|
#define __builtin_ia32_pcmpestrio128(X, LX, Y, LY, M) \
|
| 71 |
|
|
__builtin_ia32_pcmpestrio128(X, LX, Y, LY, 1)
|
| 72 |
|
|
#define __builtin_ia32_pcmpestris128(X, LX, Y, LY, M) \
|
| 73 |
|
|
__builtin_ia32_pcmpestris128(X, LX, Y, LY, 1)
|
| 74 |
|
|
#define __builtin_ia32_pcmpestriz128(X, LX, Y, LY, M) \
|
| 75 |
|
|
__builtin_ia32_pcmpestriz128(X, LX, Y, LY, 1)
|
| 76 |
|
|
|
| 77 |
|
|
/* tmmintrin.h */
|
| 78 |
|
|
#define __builtin_ia32_palignr128(X, Y, N) __builtin_ia32_palignr128(X, Y, 8)
|
| 79 |
|
|
#define __builtin_ia32_palignr(X, Y, N) __builtin_ia32_palignr(X, Y, 8)
|
| 80 |
|
|
|
| 81 |
|
|
/* emmintrin.h */
|
| 82 |
|
|
#define __builtin_ia32_psrldqi128(A, B) __builtin_ia32_psrldqi128(A, 8)
|
| 83 |
|
|
#define __builtin_ia32_pslldqi128(A, B) __builtin_ia32_pslldqi128(A, 8)
|
| 84 |
|
|
#define __builtin_ia32_pshufhw(A, N) __builtin_ia32_pshufhw(A, 0)
|
| 85 |
|
|
#define __builtin_ia32_pshuflw(A, N) __builtin_ia32_pshuflw(A, 0)
|
| 86 |
|
|
#define __builtin_ia32_pshufd(A, N) __builtin_ia32_pshufd(A, 0)
|
| 87 |
|
|
#define __builtin_ia32_vec_set_v8hi(A, D, N) \
|
| 88 |
|
|
__builtin_ia32_vec_set_v8hi(A, D, 0)
|
| 89 |
|
|
#define __builtin_ia32_vec_ext_v8hi(A, N) __builtin_ia32_vec_ext_v8hi(A, 0)
|
| 90 |
|
|
#define __builtin_ia32_shufpd(A, B, N) __builtin_ia32_shufpd(A, B, 0)
|
| 91 |
|
|
|
| 92 |
|
|
/* xmmintrin.h */
|
| 93 |
|
|
#define __builtin_prefetch(P, A, I) __builtin_prefetch(P, A, _MM_HINT_NTA)
|
| 94 |
|
|
#define __builtin_ia32_pshufw(A, N) __builtin_ia32_pshufw(A, 0)
|
| 95 |
|
|
#define __builtin_ia32_vec_set_v4hi(A, D, N) \
|
| 96 |
|
|
__builtin_ia32_vec_set_v4hi(A, D, 0)
|
| 97 |
|
|
#define __builtin_ia32_vec_ext_v4hi(A, N) __builtin_ia32_vec_ext_v4hi(A, 0)
|
| 98 |
|
|
#define __builtin_ia32_shufps(A, B, N) __builtin_ia32_shufps(A, B, 0)
|
| 99 |
|
|
|
| 100 |
|
|
/* immintrin.h */
|
| 101 |
|
|
#define __builtin_ia32_blendpd256(X, Y, M) __builtin_ia32_blendpd256(X, Y, 1)
|
| 102 |
|
|
#define __builtin_ia32_blendps256(X, Y, M) __builtin_ia32_blendps256(X, Y, 1)
|
| 103 |
|
|
#define __builtin_ia32_dpps256(X, Y, M) __builtin_ia32_dpps256(X, Y, 1)
|
| 104 |
|
|
#define __builtin_ia32_shufpd256(X, Y, M) __builtin_ia32_shufpd256(X, Y, 1)
|
| 105 |
|
|
#define __builtin_ia32_shufps256(X, Y, M) __builtin_ia32_shufps256(X, Y, 1)
|
| 106 |
|
|
#define __builtin_ia32_cmpsd(X, Y, O) __builtin_ia32_cmpsd(X, Y, 1)
|
| 107 |
|
|
#define __builtin_ia32_cmpss(X, Y, O) __builtin_ia32_cmpss(X, Y, 1)
|
| 108 |
|
|
#define __builtin_ia32_cmppd(X, Y, O) __builtin_ia32_cmppd(X, Y, 1)
|
| 109 |
|
|
#define __builtin_ia32_cmpps(X, Y, O) __builtin_ia32_cmpps(X, Y, 1)
|
| 110 |
|
|
#define __builtin_ia32_cmppd256(X, Y, O) __builtin_ia32_cmppd256(X, Y, 1)
|
| 111 |
|
|
#define __builtin_ia32_cmpps256(X, Y, O) __builtin_ia32_cmpps256(X, Y, 1)
|
| 112 |
|
|
#define __builtin_ia32_vextractf128_pd256(X, N) __builtin_ia32_vextractf128_pd256(X, 1)
|
| 113 |
|
|
#define __builtin_ia32_vextractf128_ps256(X, N) __builtin_ia32_vextractf128_ps256(X, 1)
|
| 114 |
|
|
#define __builtin_ia32_vextractf128_si256(X, N) __builtin_ia32_vextractf128_si256(X, 1)
|
| 115 |
|
|
#define __builtin_ia32_vpermilpd(X, N) __builtin_ia32_vpermilpd(X, 1)
|
| 116 |
|
|
#define __builtin_ia32_vpermilpd256(X, N) __builtin_ia32_vpermilpd256(X, 1)
|
| 117 |
|
|
#define __builtin_ia32_vpermilps(X, N) __builtin_ia32_vpermilps(X, 1)
|
| 118 |
|
|
#define __builtin_ia32_vpermilps256(X, N) __builtin_ia32_vpermilps256(X, 1)
|
| 119 |
|
|
#define __builtin_ia32_vpermil2pd(X, Y, C, I) __builtin_ia32_vpermil2pd(X, Y, C, 1)
|
| 120 |
|
|
#define __builtin_ia32_vpermil2pd256(X, Y, C, I) __builtin_ia32_vpermil2pd256(X, Y, C, 1)
|
| 121 |
|
|
#define __builtin_ia32_vpermil2ps(X, Y, C, I) __builtin_ia32_vpermil2ps(X, Y, C, 1)
|
| 122 |
|
|
#define __builtin_ia32_vpermil2ps256(X, Y, C, I) __builtin_ia32_vpermil2ps256(X, Y, C, 1)
|
| 123 |
|
|
#define __builtin_ia32_vperm2f128_pd256(X, Y, C) __builtin_ia32_vperm2f128_pd256(X, Y, 1)
|
| 124 |
|
|
#define __builtin_ia32_vperm2f128_ps256(X, Y, C) __builtin_ia32_vperm2f128_ps256(X, Y, 1)
|
| 125 |
|
|
#define __builtin_ia32_vperm2f128_si256(X, Y, C) __builtin_ia32_vperm2f128_si256(X, Y, 1)
|
| 126 |
|
|
#define __builtin_ia32_vinsertf128_pd256(X, Y, C) __builtin_ia32_vinsertf128_pd256(X, Y, 1)
|
| 127 |
|
|
#define __builtin_ia32_vinsertf128_ps256(X, Y, C) __builtin_ia32_vinsertf128_ps256(X, Y, 1)
|
| 128 |
|
|
#define __builtin_ia32_vinsertf128_si256(X, Y, C) __builtin_ia32_vinsertf128_si256(X, Y, 1)
|
| 129 |
|
|
#define __builtin_ia32_roundpd256(V, M) __builtin_ia32_roundpd256(V, 1)
|
| 130 |
|
|
#define __builtin_ia32_roundps256(V, M) __builtin_ia32_roundps256(V, 1)
|
| 131 |
|
|
#define __builtin_ia32_vcvtps2ph(A, I) __builtin_ia32_vcvtps2ph(A, 1)
|
| 132 |
|
|
#define __builtin_ia32_vcvtps2ph256(A, I) __builtin_ia32_vcvtps2ph256(A, 1)
|
| 133 |
|
|
|
| 134 |
|
|
/* xopintrin.h */
|
| 135 |
|
|
#define __builtin_ia32_vprotbi(A, B) __builtin_ia32_vprotbi(A,1)
|
| 136 |
|
|
#define __builtin_ia32_vprotwi(A, B) __builtin_ia32_vprotwi(A,1)
|
| 137 |
|
|
#define __builtin_ia32_vprotdi(A, B) __builtin_ia32_vprotdi(A,1)
|
| 138 |
|
|
#define __builtin_ia32_vprotqi(A, B) __builtin_ia32_vprotqi(A,1)
|
| 139 |
|
|
|
| 140 |
|
|
/* lwpintrin.h */
|
| 141 |
|
|
#define __builtin_ia32_lwpval32(D2, D1, F) __builtin_ia32_lwpval32 (D2, D1, 1)
|
| 142 |
|
|
#define __builtin_ia32_lwpval64(D2, D1, F) __builtin_ia32_lwpval64 (D2, D1, 1)
|
| 143 |
|
|
#define __builtin_ia32_lwpins32(D2, D1, F) __builtin_ia32_lwpins32 (D2, D1, 1)
|
| 144 |
|
|
#define __builtin_ia32_lwpins64(D2, D1, F) __builtin_ia32_lwpins64 (D2, D1, 1)
|
| 145 |
|
|
|
| 146 |
|
|
/* tbmintrin.h */
|
| 147 |
|
|
#define __builtin_ia32_bextri_u32(X, Y) __builtin_ia32_bextr_u32 (X, 1)
|
| 148 |
|
|
#define __builtin_ia32_bextri_u64(X, Y) __builtin_ia32_bextr_u64 (X, 1)
|
| 149 |
|
|
|
| 150 |
|
|
/* avx2intrin.h */
|
| 151 |
|
|
#define __builtin_ia32_mpsadbw256(X, Y, Z) __builtin_ia32_mpsadbw256 (X, Y, 1)
|
| 152 |
|
|
#define __builtin_ia32_palignr256(X, Y, Z) __builtin_ia32_palignr256 (X, Y, 8)
|
| 153 |
|
|
#define __builtin_ia32_pblendw256(X, Y, Z) __builtin_ia32_pblendw256 (X, Y, 1)
|
| 154 |
|
|
#define __builtin_ia32_pshufd256(X, Y) __builtin_ia32_pshufd256(X, 1)
|
| 155 |
|
|
#define __builtin_ia32_pshufhw256(X, Y) __builtin_ia32_pshufhw256(X, 1)
|
| 156 |
|
|
#define __builtin_ia32_pshuflw256(X, Y) __builtin_ia32_pshuflw256(X, 1)
|
| 157 |
|
|
#define __builtin_ia32_pslldqi256(X, Y) __builtin_ia32_pslldqi256(X, 8)
|
| 158 |
|
|
#define __builtin_ia32_psrldqi256(X, Y) __builtin_ia32_psrldqi256(X, 8)
|
| 159 |
|
|
#define __builtin_ia32_pblendd128(X, Y, Z) __builtin_ia32_pblendd128(X, Y, 1)
|
| 160 |
|
|
#define __builtin_ia32_pblendd256(X, Y, Z) __builtin_ia32_pblendd256(X, Y, 1)
|
| 161 |
|
|
#define __builtin_ia32_permdf256(X, Y) __builtin_ia32_permdf256(X, 1)
|
| 162 |
|
|
#define __builtin_ia32_permdi256(X, Y) __builtin_ia32_permdi256(X, 1)
|
| 163 |
|
|
#define __builtin_ia32_permti256(X, Y, Z) __builtin_ia32_permti256(X, Y, 1)
|
| 164 |
|
|
#define __builtin_ia32_extract128i256(X, Y) __builtin_ia32_extract128i256(X, 1)
|
| 165 |
|
|
#define __builtin_ia32_insert128i256(X, Y, Z) __builtin_ia32_insert128i256(X, Y, 1)
|
| 166 |
|
|
#define __builtin_ia32_gathersiv2df(X, Y, Z, K, M) __builtin_ia32_gathersiv2df(X, Y, Z, K, 1)
|
| 167 |
|
|
#define __builtin_ia32_gathersiv4df(X, Y, Z, K, M) __builtin_ia32_gathersiv4df(X, Y, Z, K, 1)
|
| 168 |
|
|
#define __builtin_ia32_gatherdiv2df(X, Y, Z, K, M) __builtin_ia32_gatherdiv2df(X, Y, Z, K, 1)
|
| 169 |
|
|
#define __builtin_ia32_gatherdiv4df(X, Y, Z, K, M) __builtin_ia32_gatherdiv4df(X, Y, Z, K, 1)
|
| 170 |
|
|
#define __builtin_ia32_gathersiv4sf(X, Y, Z, K, M) __builtin_ia32_gathersiv4sf(X, Y, Z, K, 1)
|
| 171 |
|
|
#define __builtin_ia32_gathersiv8sf(X, Y, Z, K, M) __builtin_ia32_gathersiv8sf(X, Y, Z, K, 1)
|
| 172 |
|
|
#define __builtin_ia32_gatherdiv4sf(X, Y, Z, K, M) __builtin_ia32_gatherdiv4sf(X, Y, Z, K, 1)
|
| 173 |
|
|
#define __builtin_ia32_gatherdiv4sf256(X, Y, Z, K, M) __builtin_ia32_gatherdiv4sf256(X, Y, Z, K, 1)
|
| 174 |
|
|
#define __builtin_ia32_gathersiv2di(X, Y, Z, K, M) __builtin_ia32_gathersiv2di(X, Y, Z, K, 1)
|
| 175 |
|
|
#define __builtin_ia32_gathersiv4di(X, Y, Z, K, M) __builtin_ia32_gathersiv4di(X, Y, Z, K, 1)
|
| 176 |
|
|
#define __builtin_ia32_gatherdiv2di(X, Y, Z, K, M) __builtin_ia32_gatherdiv2di(X, Y, Z, K, 1)
|
| 177 |
|
|
#define __builtin_ia32_gatherdiv4di(X, Y, Z, K, M) __builtin_ia32_gatherdiv4di(X, Y, Z, K, 1)
|
| 178 |
|
|
#define __builtin_ia32_gathersiv4si(X, Y, Z, K, M) __builtin_ia32_gathersiv4si(X, Y, Z, K, 1)
|
| 179 |
|
|
#define __builtin_ia32_gathersiv8si(X, Y, Z, K, M) __builtin_ia32_gathersiv8si(X, Y, Z, K, 1)
|
| 180 |
|
|
#define __builtin_ia32_gatherdiv4si(X, Y, Z, K, M) __builtin_ia32_gatherdiv4si(X, Y, Z, K, 1)
|
| 181 |
|
|
#define __builtin_ia32_gatherdiv4si256(X, Y, Z, K, M) __builtin_ia32_gatherdiv4si256(X, Y, Z, K, 1)
|
| 182 |
|
|
|
| 183 |
|
|
#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma")
|
| 184 |
|
|
#include <wmmintrin.h>
|
| 185 |
|
|
#include <smmintrin.h>
|
| 186 |
|
|
#include <mm3dnow.h>
|
| 187 |
|
|
#include <x86intrin.h>
|
