gcc/testsuite/gcc.target/i386/sse-23.c

   1 /* { 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}mmintrin.h,
   8    xopintrin.h, lwpintrin.h, popcntintrin.h and mm3dnow.h that
   9    reference the proper builtin functions.  Defining away "extern" and
  10    "__inline" results in all of them being compiled as proper
  11    functions.  */
  12
  13 #define extern
  14 #define __inline
  15
  16 /* Following intrinsics require immediate arguments. */
  17
  18 /* ammintrin.h */
  19 #define __builtin_ia32_extrqi(X, I, L)  __builtin_ia32_extrqi(X, 1, 1)
  20 #define __builtin_ia32_insertqi(X, Y, I, L) __builtin_ia32_insertqi(X, Y, 1, 1)
  21
  22 /* wmmintrin.h */
  23 #define __builtin_ia32_aeskeygenassist128(X, C) __builtin_ia32_aeskeygenassist128(X, 1)
  24 #define __builtin_ia32_pclmulqdq128(X, Y, I) __builtin_ia32_pclmulqdq128(X, Y, 1)
  25
  26 /* smmintrin.h */
  27 #define __builtin_ia32_roundpd(V, M) __builtin_ia32_roundpd(V, 1)
  28 #define __builtin_ia32_roundsd(D, V, M) __builtin_ia32_roundsd(D, V, 1)
  29 #define __builtin_ia32_roundps(V, M) __builtin_ia32_roundps(V, 1)
  30 #define __builtin_ia32_roundss(D, V, M) __builtin_ia32_roundss(D, V, 1)
  31
  32 #define __builtin_ia32_pblendw128(X, Y, M) __builtin_ia32_pblendw128 (X, Y, 1)
  33 #define __builtin_ia32_blendps(X, Y, M) __builtin_ia32_blendps(X, Y, 1)
  34 #define __builtin_ia32_blendpd(X, Y, M) __builtin_ia32_blendpd(X, Y, 1)
  35 #define __builtin_ia32_dpps(X, Y, M) __builtin_ia32_dpps(X, Y, 1)
  36 #define __builtin_ia32_dppd(X, Y, M) __builtin_ia32_dppd(X, Y, 1)
  37 #define __builtin_ia32_insertps128(D, S, N) __builtin_ia32_insertps128(D, S, 1)
  38 #define __builtin_ia32_vec_ext_v4sf(X, N) __builtin_ia32_vec_ext_v4sf(X, 1)
  39 #define __builtin_ia32_vec_set_v16qi(D, S, N) __builtin_ia32_vec_set_v16qi(D, S, 1)
  40 #define __builtin_ia32_vec_set_v4si(D, S, N) __builtin_ia32_vec_set_v4si(D, S, 1)
  41 #define __builtin_ia32_vec_set_v2di(D, S, N) __builtin_ia32_vec_set_v2di(D, S, 1)
  42 #define __builtin_ia32_vec_ext_v16qi(X, N) __builtin_ia32_vec_ext_v16qi(X, 1)
  43 #define __builtin_ia32_vec_ext_v4si(X, N) __builtin_ia32_vec_ext_v4si(X, 1)
  44 #define __builtin_ia32_vec_ext_v2di(X, N) __builtin_ia32_vec_ext_v2di(X, 1)
  45 #define __builtin_ia32_mpsadbw128(X, Y, M) __builtin_ia32_mpsadbw128(X, Y, 1)
  46 #define __builtin_ia32_pcmpistrm128(X, Y, M) \
  47   __builtin_ia32_pcmpistrm128(X, Y, 1)
  48 #define __builtin_ia32_pcmpistri128(X, Y, M) \
  49   __builtin_ia32_pcmpistri128(X, Y, 1)
  50 #define __builtin_ia32_pcmpestrm128(X, LX, Y, LY, M) \
  51   __builtin_ia32_pcmpestrm128(X, LX, Y, LY, 1)
  52 #define __builtin_ia32_pcmpestri128(X, LX, Y, LY, M) \
  53   __builtin_ia32_pcmpestri128(X, LX, Y, LY, 1)
  54 #define __builtin_ia32_pcmpistria128(X, Y, M) \
  55   __builtin_ia32_pcmpistria128(X, Y, 1)
  56 #define __builtin_ia32_pcmpistric128(X, Y, M) \
  57   __builtin_ia32_pcmpistric128(X, Y, 1)
  58 #define __builtin_ia32_pcmpistrio128(X, Y, M) \
  59   __builtin_ia32_pcmpistrio128(X, Y, 1)
  60 #define __builtin_ia32_pcmpistris128(X, Y, M) \
  61   __builtin_ia32_pcmpistris128(X, Y, 1)
  62 #define __builtin_ia32_pcmpistriz128(X, Y, M) \
  63   __builtin_ia32_pcmpistriz128(X, Y, 1)
  64 #define __builtin_ia32_pcmpestria128(X, LX, Y, LY, M) \
  65   __builtin_ia32_pcmpestria128(X, LX, Y, LY, 1)
  66 #define __builtin_ia32_pcmpestric128(X, LX, Y, LY, M) \
  67   __builtin_ia32_pcmpestric128(X, LX, Y, LY, 1)
  68 #define __builtin_ia32_pcmpestrio128(X, LX, Y, LY, M) \
  69   __builtin_ia32_pcmpestrio128(X, LX, Y, LY, 1)
  70 #define __builtin_ia32_pcmpestris128(X, LX, Y, LY, M) \
  71   __builtin_ia32_pcmpestris128(X, LX, Y, LY, 1)
  72 #define __builtin_ia32_pcmpestriz128(X, LX, Y, LY, M) \
  73   __builtin_ia32_pcmpestriz128(X, LX, Y, LY, 1)
  74
  75 /* tmmintrin.h */
  76 #define __builtin_ia32_palignr128(X, Y, N) __builtin_ia32_palignr128(X, Y, 8)
  77 #define __builtin_ia32_palignr(X, Y, N) __builtin_ia32_palignr(X, Y, 8)
  78
  79 /* emmintrin.h */
  80 #define __builtin_ia32_psrldqi128(A, B) __builtin_ia32_psrldqi128(A, 8)
  81 #define __builtin_ia32_pslldqi128(A, B) __builtin_ia32_pslldqi128(A, 8)
  82 #define __builtin_ia32_pshufhw(A, N) __builtin_ia32_pshufhw(A, 0)
  83 #define __builtin_ia32_pshuflw(A, N) __builtin_ia32_pshuflw(A, 0)
  84 #define __builtin_ia32_pshufd(A, N) __builtin_ia32_pshufd(A, 0)
  85 #define __builtin_ia32_vec_set_v8hi(A, D, N) \
  86   __builtin_ia32_vec_set_v8hi(A, D, 0)
  87 #define __builtin_ia32_vec_ext_v8hi(A, N) __builtin_ia32_vec_ext_v8hi(A, 0)
  88 #define __builtin_ia32_shufpd(A, B, N) __builtin_ia32_shufpd(A, B, 0)
  89
  90 /* xmmintrin.h */
  91 #define __builtin_prefetch(P, A, I) __builtin_prefetch(P, A, _MM_HINT_NTA)
  92 #define __builtin_ia32_pshufw(A, N) __builtin_ia32_pshufw(A, 0)
  93 #define __builtin_ia32_vec_set_v4hi(A, D, N) \
  94   __builtin_ia32_vec_set_v4hi(A, D, 0)
  95 #define __builtin_ia32_vec_ext_v4hi(A, N) __builtin_ia32_vec_ext_v4hi(A, 0)
  96 #define __builtin_ia32_shufps(A, B, N) __builtin_ia32_shufps(A, B, 0)
  97
  98 /* immintrin.h */
  99 #define __builtin_ia32_blendpd256(X, Y, M) __builtin_ia32_blendpd256(X, Y, 1)
 100 #define __builtin_ia32_blendps256(X, Y, M) __builtin_ia32_blendps256(X, Y, 1)
 101 #define __builtin_ia32_dpps256(X, Y, M) __builtin_ia32_dpps256(X, Y, 1)
 102 #define __builtin_ia32_shufpd256(X, Y, M) __builtin_ia32_shufpd256(X, Y, 1)
 103 #define __builtin_ia32_shufps256(X, Y, M) __builtin_ia32_shufps256(X, Y, 1)
 104 #define __builtin_ia32_cmpsd(X, Y, O) __builtin_ia32_cmpsd(X, Y, 1)
 105 #define __builtin_ia32_cmpss(X, Y, O) __builtin_ia32_cmpss(X, Y, 1)
 106 #define __builtin_ia32_cmppd(X, Y, O) __builtin_ia32_cmppd(X, Y, 1)
 107 #define __builtin_ia32_cmpps(X, Y, O) __builtin_ia32_cmpps(X, Y, 1)
 108 #define __builtin_ia32_cmppd256(X, Y, O) __builtin_ia32_cmppd256(X, Y, 1)
 109 #define __builtin_ia32_cmpps256(X, Y, O) __builtin_ia32_cmpps256(X, Y, 1)
 110 #define __builtin_ia32_vextractf128_pd256(X, N) __builtin_ia32_vextractf128_pd256(X, 1)
 111 #define __builtin_ia32_vextractf128_ps256(X, N) __builtin_ia32_vextractf128_ps256(X, 1)
 112 #define __builtin_ia32_vextractf128_si256(X, N) __builtin_ia32_vextractf128_si256(X, 1)
 113 #define __builtin_ia32_vpermilpd(X, N) __builtin_ia32_vpermilpd(X, 1)
 114 #define __builtin_ia32_vpermilpd256(X, N) __builtin_ia32_vpermilpd256(X, 1)
 115 #define __builtin_ia32_vpermilps(X, N) __builtin_ia32_vpermilps(X, 1)
 116 #define __builtin_ia32_vpermilps256(X, N) __builtin_ia32_vpermilps256(X, 1)
 117 #define __builtin_ia32_vpermil2pd(X, Y, C, I) __builtin_ia32_vpermil2pd(X, Y, C, 1)
 118 #define __builtin_ia32_vpermil2pd256(X, Y, C, I) __builtin_ia32_vpermil2pd256(X, Y, C, 1)
 119 #define __builtin_ia32_vpermil2ps(X, Y, C, I) __builtin_ia32_vpermil2ps(X, Y, C, 1)
 120 #define __builtin_ia32_vpermil2ps256(X, Y, C, I) __builtin_ia32_vpermil2ps256(X, Y, C, 1)
 121 #define __builtin_ia32_vperm2f128_pd256(X, Y, C) __builtin_ia32_vperm2f128_pd256(X, Y, 1)
 122 #define __builtin_ia32_vperm2f128_ps256(X, Y, C) __builtin_ia32_vperm2f128_ps256(X, Y, 1)
 123 #define __builtin_ia32_vperm2f128_si256(X, Y, C) __builtin_ia32_vperm2f128_si256(X, Y, 1)
 124 #define __builtin_ia32_vinsertf128_pd256(X, Y, C) __builtin_ia32_vinsertf128_pd256(X, Y, 1)
 125 #define __builtin_ia32_vinsertf128_ps256(X, Y, C) __builtin_ia32_vinsertf128_ps256(X, Y, 1)
 126 #define __builtin_ia32_vinsertf128_si256(X, Y, C) __builtin_ia32_vinsertf128_si256(X, Y, 1)
 127 #define __builtin_ia32_roundpd256(V, M) __builtin_ia32_roundpd256(V, 1)
 128 #define __builtin_ia32_roundps256(V, M) __builtin_ia32_roundps256(V, 1)
 129
 130 /* xopintrin.h */
 131 #define __builtin_ia32_vprotbi(A, B) __builtin_ia32_vprotbi(A,1)
 132 #define __builtin_ia32_vprotwi(A, B) __builtin_ia32_vprotwi(A,1)
 133 #define __builtin_ia32_vprotdi(A, B) __builtin_ia32_vprotdi(A,1)
 134 #define __builtin_ia32_vprotqi(A, B) __builtin_ia32_vprotqi(A,1)
 135
 136 /* lwpintrin.h */
 137 #define __builtin_ia32_lwpval32(D2, D1, F) __builtin_ia32_lwpval32 (D2, D1, 1)
 138 #define __builtin_ia32_lwpval64(D2, D1, F) __builtin_ia32_lwpval64 (D2, D1, 1)
 139 #define __builtin_ia32_lwpins32(D2, D1, F) __builtin_ia32_lwpins32 (D2, D1, 1)
 140 #define __builtin_ia32_lwpins64(D2, D1, F) __builtin_ia32_lwpins64 (D2, D1, 1)
 141
 142 #pragma GCC target ("3dnow,sse4,sse4a,aes,pclmul,xop,abm,popcnt,lwp")
 143 #include <wmmintrin.h>
 144 #include <smmintrin.h>
 145 #include <mm3dnow.h>
 146 #include <x86intrin.h>