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

   1 /* { dg-do run } */
   2 /* { dg-options "-O2 -msse3 -mfpmath=sse" } */
   3
   4 #include "sse3-check.h"
   5
   6 #include <pmmintrin.h>
   7
   8 static void
   9 sse3_test_addsubpd (double *i1, double *i2, double *r)
  10 {
  11   __m128d t1 = _mm_loadu_pd (i1);
  12   __m128d t2 = _mm_loadu_pd (i2);
  13
  14   t1 = _mm_addsub_pd (t1, t2);
  15
  16   _mm_storeu_pd (r, t1);
  17 }
  18
  19 static void
  20 sse3_test_addsubpd_subsume (double *i1, double *i2, double *r)
  21 {
  22   __m128d t1 = _mm_load_pd (i1);
  23   __m128d t2 = _mm_load_pd (i2);
  24
  25   t1 = _mm_addsub_pd (t1, t2);
  26
  27   _mm_storeu_pd (r, t1);
  28 }
  29
  30 static int
  31 chk_pd (double *v1, double *v2)
  32 {
  33   int i;
  34   int n_fails = 0;
  35
  36   for (i = 0; i < 2; i++)
  37     if (v1[i] != v2[i])
  38       n_fails += 1;
  39
  40   return n_fails;
  41 }
  42
  43 static double p1[2] __attribute__ ((aligned(16)));
  44 static double p2[2] __attribute__ ((aligned(16)));
  45 static double p3[2];
  46 static double ck[2];
  47
  48 double vals[80] =
  49   {
  50     100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
  51     1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
  52     32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
  53     12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
  54     541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
  55     23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
  56     23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
  57     1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
  58     -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
  59     9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
  60   };
  61
  62 static void
  63 sse3_test (void)
  64 {
  65   int i;
  66   int fail = 0;
  67
  68   for (i = 0; i < 80; i += 4)
  69     {
  70       p1[0] = vals[i+0];
  71       p1[1] = vals[i+1];
  72
  73       p2[0] = vals[i+2];
  74       p2[1] = vals[i+3];
  75
  76       ck[0] = p1[0] - p2[0];
  77       ck[1] = p1[1] + p2[1];
  78
  79       sse3_test_addsubpd (p1, p2, p3);
  80
  81       fail += chk_pd (ck, p3);
  82
  83       sse3_test_addsubpd_subsume (p1, p2, p3);
  84
  85       fail += chk_pd (ck, p3);
  86     }
  87
  88   if (fail != 0)
  89     abort ();
  90
  91   exit (0);
  92 }