gcc/testsuite/gcc.target/i386/sse3-addsubpd.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
  15 #include <pmmintrin.h>
  16
  17 static void
  18 sse3_test_addsubpd (double *i1, double *i2, double *r)
  19 {
  20   __m128d t1 = _mm_loadu_pd (i1);
  21   __m128d t2 = _mm_loadu_pd (i2);
  22
  23   t1 = _mm_addsub_pd (t1, t2);
  24
  25   _mm_storeu_pd (r, t1);
  26 }
  27
  28 static void
  29 sse3_test_addsubpd_subsume (double *i1, double *i2, double *r)
  30 {
  31   __m128d t1 = _mm_load_pd (i1);
  32   __m128d t2 = _mm_load_pd (i2);
  33
  34   t1 = _mm_addsub_pd (t1, t2);
  35
  36   _mm_storeu_pd (r, t1);
  37 }
  38
  39 static int
  40 chk_pd (double *v1, double *v2)
  41 {
  42   int i;
  43   int n_fails = 0;
  44
  45   for (i = 0; i < 2; i++)
  46     if (v1[i] != v2[i])
  47       n_fails += 1;
  48
  49   return n_fails;
  50 }
  51
  52 static double p1[2] __attribute__ ((aligned(16)));
  53 static double p2[2] __attribute__ ((aligned(16)));
  54 static double p3[2];
  55 static double ck[2];
  56
  57 double vals[80] =
  58   {
  59     100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
  60     1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
  61     32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
  62     12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
  63     541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
  64     23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
  65     23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
  66     1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
  67     -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
  68     9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
  69   };
  70
  71 static void
  72 TEST (void)
  73 {
  74   int i;
  75   int fail = 0;
  76
  77   for (i = 0; i < 80; i += 4)
  78     {
  79       p1[0] = vals[i+0];
  80       p1[1] = vals[i+1];
  81
  82       p2[0] = vals[i+2];
  83       p2[1] = vals[i+3];
  84
  85       ck[0] = p1[0] - p2[0];
  86       ck[1] = p1[1] + p2[1];
  87
  88       sse3_test_addsubpd (p1, p2, p3);
  89
  90       fail += chk_pd (ck, p3);
  91
  92       sse3_test_addsubpd_subsume (p1, p2, p3);
  93
  94       fail += chk_pd (ck, p3);
  95     }
  96
  97   if (fail != 0)
  98     abort ();
  99 }