gcc/testsuite/gcc.dg/vect/vect-outer-fir.c

   1 /* { dg-require-effective-target vect_float } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #define N 40
   7 #define M 128
   8 float in[N+M];
   9 float coeff[M];
  10 float out[N];
  11 float fir_out[N];
  12
  13 /* Should be vectorized. Fixed misaligment in the inner-loop.  */
  14 /* Currently not vectorized because we get too many BBs in the inner-loop,
  15    because the compiler doesn't realize that the inner-loop executes at
  16    least once (cause k<4), and so there's no need to create a guard code
  17    to skip the inner-loop in case it doesn't execute.  */
  18 __attribute__ ((noinline))
  19 void foo (){
  20  int i,j,k;
  21  float diff;
  22
  23  for (i = 0; i < N; i++) {
  24   out[i] = 0;
  25  }
  26
  27  for (k = 0; k < 4; k++) {
  28   for (i = 0; i < N; i++) {
  29     diff = 0;
  30     for (j = k; j < M; j+=4) {
  31       diff += in[j+i]*coeff[j];
  32     }
  33     out[i] += diff;
  34   }
  35  }
  36
  37 }
  38
  39 /* Vectorized. Changing misalignment in the inner-loop.  */
  40 __attribute__ ((noinline))
  41 void fir (){
  42   int i,j,k;
  43   float diff;
  44
  45   for (i = 0; i < N; i++) {
  46     diff = 0;
  47     for (j = 0; j < M; j++) {
  48       diff += in[j+i]*coeff[j];
  49     }
  50     fir_out[i] = diff;
  51   }
  52 }
  53
  54
  55 int main (void)
  56 {
  57   check_vect ();
  58   int i, j;
  59   float diff;
  60
  61   for (i = 0; i < M; i++)
  62     coeff[i] = i;
  63   for (i = 0; i < N+M; i++)
  64     in[i] = i;
  65
  66   foo ();
  67   fir ();
  68
  69   for (i = 0; i < N; i++) {
  70     if (out[i] != fir_out[i])
  71       abort ();
  72   }
  73
  74   return 0;
  75 }
  76
  77 /* { dg-final { scan-tree-dump-times "OUTER LOOP VECTORIZED" 2 "vect" { xfail *-*-* } } } */
  78 /* { dg-final { scan-tree-dump-times "OUTER LOOP VECTORIZED" 1 "vect" { xfail vect_no_align } } } */
  79 /* { dg-final { cleanup-tree-dump "vect" } } */