gcc/testsuite/gcc.target/i386/sse3-hsubps.c

   1 /* { dg-do run } */
   2 /* { dg-require-effective-target sse3 } */
   3 /* { dg-options "-O2 -msse3 -mfpmath=sse" } */
   4
   5 #ifndef CHECK_H
   6 #define CHECK_H "sse3-check.h"
   7 #endif
   8
   9 #ifndef TEST
  10 #define TEST sse3_test
  11 #endif
  12
  13 #include CHECK_H
  14 #include <pmmintrin.h>
  15
  16 static void
  17 sse3_test_hsubps (float *i1, float *i2, float *r)
  18 {
  19   __m128 t1 = _mm_loadu_ps (i1);
  20   __m128 t2 = _mm_loadu_ps (i2);
  21
  22   t1 = _mm_hsub_ps (t1, t2);
  23
  24   _mm_storeu_ps (r, t1);
  25 }
  26
  27 static void
  28 sse3_test_hsubps_subsume (float *i1, float *i2, float *r)
  29 {
  30   __m128 t1 = _mm_load_ps (i1);
  31   __m128 t2 = _mm_load_ps (i2);
  32
  33   t1 = _mm_hsub_ps (t1, t2);
  34
  35   _mm_storeu_ps (r, t1);
  36 }
  37
  38 static int
  39 chk_ps (float *v1, float *v2)
  40 {
  41   int i;
  42   int n_fails = 0;
  43
  44   for (i = 0; i < 4; i++) {
  45     if (v1[i] != v2[i]) {
  46       n_fails += 1;
  47     }
  48   }
  49
  50   return n_fails;
  51 }
  52
  53 static float p1[4] __attribute__ ((aligned(16)));
  54 static float p2[4] __attribute__ ((aligned(16)));
  55 static float p3[4];
  56 static float ck[4];
  57
  58 static float vals[80] =
  59   {
  60     100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
  61     1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
  62     32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
  63     12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
  64     541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
  65     23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
  66     23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
  67     1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
  68     -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
  69     9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
  70   };
  71
  72 static void
  73 TEST (void)
  74 {
  75   int i;
  76   int fail = 0;
  77
  78   for (i = 0; i < 80; i += 8)
  79     {
  80       p1[0] = vals[i+0];
  81       p1[1] = vals[i+1];
  82       p1[2] = vals[i+2];
  83       p1[3] = vals[i+3];
  84
  85       p2[0] = vals[i+4];
  86       p2[1] = vals[i+5];
  87       p2[2] = vals[i+6];
  88       p2[3] = vals[i+7];
  89
  90       ck[0] = p1[0] - p1[1];
  91       ck[1] = p1[2] - p1[3];
  92       ck[2] = p2[0] - p2[1];
  93       ck[3] = p2[2] - p2[3];
  94
  95       sse3_test_hsubps (p1, p2, p3);
  96
  97       fail += chk_ps (ck, p3);
  98
  99       sse3_test_hsubps_subsume (p1, p2, p3);
 100
 101       fail += chk_ps (ck, p3);
 102     }
 103
 104   if (fail != 0)
 105     abort ();
 106 }