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

   1 /* { dg-do run } */
   2 /* { dg-options "-O2 -msse3 -mfpmath=sse" } */
   3
   4 #ifndef CHECK_H
   5 #define CHECK_H "sse3-check.h"
   6 #endif
   7
   8 #ifndef TEST
   9 #define TEST sse3_test
  10 #endif
  11
  12 #include CHECK_H
  13
  14 #include <pmmintrin.h>
  15
  16 static void
  17 sse3_test_movshdup_reg (float *i1, float *r)
  18 {
  19   __m128 t1 = _mm_loadu_ps (i1);
  20   __m128 t2 = _mm_movehdup_ps (t1);
  21
  22   _mm_storeu_ps (r, t2);
  23 }
  24
  25 static void
  26 sse3_test_movshdup_reg_subsume (float *i1, float *r)
  27 {
  28   __m128 t1 = _mm_load_ps (i1);
  29   __m128 t2 = _mm_movehdup_ps (t1);
  30
  31   _mm_storeu_ps (r, t2);
  32 }
  33
  34 static int
  35 chk_ps (float *v1, float *v2)
  36 {
  37   int i;
  38   int n_fails = 0;
  39
  40   for (i = 0; i < 4; i++)
  41     if (v1[i] != v2[i])
  42       n_fails += 1;
  43
  44   return n_fails;
  45 }
  46
  47 static float p1[4] __attribute__ ((aligned(16)));
  48 static float p2[4];
  49 static float ck[4];
  50
  51 static float vals[80] =
  52   {
  53     100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
  54     1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
  55     32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
  56     12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
  57     541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
  58     23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
  59     23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
  60     1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
  61     -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
  62     9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
  63   };
  64
  65 static void
  66 TEST (void)
  67 {
  68   int i;
  69   int fail = 0;
  70
  71   for (i = 0; i < 80; i += 2)
  72     {
  73       p1[0] = 0.0;
  74       p1[1] = vals[i+0];
  75       p1[2] = 1.0;
  76       p1[3] = vals[i+1];
  77
  78       ck[0] = p1[1];
  79       ck[1] = p1[1];
  80       ck[2] = p1[3];
  81       ck[3] = p1[3];
  82
  83       sse3_test_movshdup_reg (p1, p2);
  84
  85       fail += chk_ps (ck, p2);
  86
  87       sse3_test_movshdup_reg_subsume (p1, p2);
  88
  89       fail += chk_ps (ck, p2);
  90     }
  91
  92   if (fail != 0)
  93     abort ();
  94 }