gcc/testsuite/gcc.target/powerpc/sse3-hsubps.c

   1 /* { dg-do run } */
   2 /* { dg-options "-O3 -mvsx -Wno-psabi" } */
   3 /* { dg-additional-options "-mdejagnu-cpu=power8" { target { ! has_arch_pwr8 } } } */
   4 /* { dg-require-effective-target p8vector_hw } */
   5
   6 #ifndef CHECK_H
   7 #define CHECK_H "sse3-check.h"
   8 #endif
   9
  10 #include CHECK_H
  11
  12 #ifndef TEST
  13 #define TEST sse3_test_hsubps_1
  14 #endif
  15
  16 #define NO_WARN_X86_INTRINSICS 1
  17 #include <pmmintrin.h>
  18
  19 static void
  20 sse3_test_hsubps (float *i1, float *i2, float *r)
  21 {
  22   __m128 t1 = _mm_loadu_ps (i1);
  23   __m128 t2 = _mm_loadu_ps (i2);
  24
  25   t1 = _mm_hsub_ps (t1, t2);
  26
  27   _mm_storeu_ps (r, t1);
  28 }
  29
  30 static void
  31 sse3_test_hsubps_subsume (float *i1, float *i2, float *r)
  32 {
  33   __m128 t1 = _mm_load_ps (i1);
  34   __m128 t2 = _mm_load_ps (i2);
  35
  36   t1 = _mm_hsub_ps (t1, t2);
  37
  38   _mm_storeu_ps (r, t1);
  39 }
  40
  41 static int
  42 chk_ps(float *v1, float *v2)
  43 {
  44   int i;
  45   int n_fails = 0;
  46
  47   for (i = 0; i < 4; i++)
  48     if (v1[i] != v2[i])
  49       n_fails += 1;
  50
  51   return n_fails;
  52 }
  53
  54 static float p1[4] __attribute__ ((aligned(16)));
  55 static float p2[4] __attribute__ ((aligned(16)));
  56 static float p3[4];
  57 static float ck[4];
  58
  59 static float vals[] =
  60   {
  61     100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
  62     1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
  63     32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
  64     12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
  65     541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
  66     23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
  67     23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
  68     1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
  69     -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
  70     9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
  71   };
  72
  73 //static
  74 void
  75 TEST ()
  76 {
  77   int i;
  78   int fail = 0;
  79
  80   for (i = 0; i < sizeof (vals) / sizeof (vals[0]); i += 8)
  81     {
  82       p1[0] = vals[i+0];
  83       p1[1] = vals[i+1];
  84       p1[2] = vals[i+2];
  85       p1[3] = vals[i+3];
  86
  87       p2[0] = vals[i+4];
  88       p2[1] = vals[i+5];
  89       p2[2] = vals[i+6];
  90       p2[3] = vals[i+7];
  91
  92       ck[0] = p1[0] - p1[1];
  93       ck[1] = p1[2] - p1[3];
  94       ck[2] = p2[0] - p2[1];
  95       ck[3] = p2[2] - p2[3];
  96
  97       sse3_test_hsubps (p1, p2, p3);
  98
  99       fail += chk_ps (ck, p3);
 100
 101       sse3_test_hsubps_subsume (p1, p2, p3);
 102
 103       fail += chk_ps (ck, p3);
 104     }
 105
 106   if (fail != 0)
 107     abort ();
 108 }