gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c

   1 /* Disabling epilogues until we find a better way to deal with scans.  */
   2 /* { dg-additional-options "--param vect-epilogues-nomask=0" } */
   3 #include "tree-vect.h"
   4
   5 extern float copysignf (float, float);
   6 extern float sqrtf (float);
   7 extern float fabsf (float);
   8 extern void abort (void);
   9 float a[64], b[64], c[64], d[64];
  10
  11 __attribute__((noinline, noclone)) void
  12 f1 (int n)
  13 {
  14   int i;
  15   for (i = 0; i < n; i++)
  16     {
  17       a[4 * i + 0] = copysignf (b[4 * i + 0], c[4 * i + 0]) + 1.0f + sqrtf (d[4 * i + 0]);
  18       a[4 * i + 1] = copysignf (b[4 * i + 1], c[4 * i + 1]) + 2.0f + sqrtf (d[4 * i + 1]);
  19       a[4 * i + 2] = copysignf (b[4 * i + 2], c[4 * i + 2]) + 3.0f + sqrtf (d[4 * i + 2]);
  20       a[4 * i + 3] = copysignf (b[4 * i + 3], c[4 * i + 3]) + 4.0f + sqrtf (d[4 * i + 3]);
  21     }
  22 }
  23
  24 __attribute__((noinline, noclone)) void
  25 f2 (int n)
  26 {
  27   int i;
  28   for (i = 0; i < 2 * n; i++)
  29     {
  30       a[2 * i + 0] = copysignf (b[2 * i + 0], c[2 * i + 0]) + 1.0f + sqrtf (d[2 * i + 0]);
  31       a[2 * i + 1] = copysignf (b[2 * i + 1], c[2 * i + 1]) + 2.0f + sqrtf (d[2 * i + 1]);
  32     }
  33 }
  34
  35 __attribute__((noinline, noclone)) void
  36 f3 (void)
  37 {
  38   int i;
  39   for (i = 0; i < 64; i++)
  40     a[i] = copysignf (b[i], c[i]) + 1.0f + sqrtf (d[i]);
  41 }
  42
  43 __attribute__((noinline, noclone)) void
  44 f4 (int n)
  45 {
  46   int i;
  47   for (i = 0; i < 2 * n; i++)
  48     {
  49       a[3 * i + 0] = copysignf (b[3 * i + 0], c[3 * i + 0]) + 1.0f + sqrtf (d[3 * i + 0]);
  50       a[3 * i + 1] = copysignf (b[3 * i + 1], c[3 * i + 1]) + 2.0f + sqrtf (d[3 * i + 1]);
  51       a[3 * i + 2] = copysignf (b[3 * i + 2], c[3 * i + 2]) + 3.0f + sqrtf (d[3 * i + 2]);
  52     }
  53 }
  54
  55 __attribute__((noinline, noclone)) int
  56 main1 ()
  57 {
  58   int i;
  59
  60   for (i = 0; i < 64; i++)
  61     {
  62       asm ("");
  63       b[i] = (i & 1) ? -4 * i : 4 * i;
  64       c[i] = (i & 2) ? -8 * i : 8 * i;
  65       d[i] = i * i;
  66     }
  67   f1 (16);
  68 #pragma GCC novector
  69   for (i = 0; i < 64; i++)
  70     if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + (i & 3) + i - a[i]) >= 0.0001f)
  71       abort ();
  72     else
  73       a[i] = 131.25;
  74   f2 (16);
  75 #pragma GCC novector
  76   for (i = 0; i < 64; i++)
  77     if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + (i & 1) + i - a[i]) >= 0.0001f)
  78       abort ();
  79     else
  80       a[i] = 131.25;
  81   f3 ();
  82 #pragma GCC novector
  83   for (i = 0; i < 64; i++)
  84     if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + i - a[i]) >= 0.0001f)
  85       abort ();
  86     else
  87       a[i] = 131.25;
  88   f4 (10);
  89 #pragma GCC novector
  90   for (i = 0; i < 60; i++)
  91     if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + (i % 3) + i - a[i]) >= 0.0001f)
  92       abort ();
  93   return 0;
  94 }
  95
  96 int
  97 main ()
  98 {
  99   check_vect ();
 100   return main1 ();
 101 }
 102
 103 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 4 "vect" { target { vect_call_copysignf && vect_call_sqrtf } } } } */
 104 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 4 "vect" { target { { vect_call_copysignf && vect_call_sqrtf } && vect_perm3_int } } } } */