gcc/testsuite/gcc.dg/vect/pr97428.c

   1 /* { dg-do compile } */
   2 /* { dg-require-effective-target vect_double } */
   3 /* { dg-additional-options "--param vect-epilogues-nomask=0" { target riscv*-*-* } } */
   4
   5 typedef struct { double re, im; } dcmlx_t;
   6 typedef struct { double re[4], im[4]; } dcmlx4_t;
   7
   8 void foo_i2(dcmlx4_t dst[], const dcmlx_t src[], int n)
   9 {
  10   for (int i = 0; i < n; ++i) {
  11     dcmlx_t s00 = src[i*4+0];
  12     dcmlx_t s01 = src[i*4+1];
  13     dcmlx_t s02 = src[i*4+2];
  14     dcmlx_t s03 = src[i*4+3];
  15
  16     dcmlx_t s10 = src[i*4+0+n];
  17     dcmlx_t s11 = src[i*4+1+n];
  18     dcmlx_t s12 = src[i*4+2+n];
  19     dcmlx_t s13 = src[i*4+3+n];
  20
  21     dst[i*2+0].re[0] = s00.re;
  22     dst[i*2+0].re[1] = s01.re;
  23     dst[i*2+0].re[2] = s02.re;
  24     dst[i*2+0].re[3] = s03.re;
  25     dst[i*2+0].im[0] = s00.im;
  26     dst[i*2+0].im[1] = s01.im;
  27     dst[i*2+0].im[2] = s02.im;
  28     dst[i*2+0].im[3] = s03.im;
  29
  30     dst[i*2+1].re[0] = s10.re;
  31     dst[i*2+1].re[1] = s11.re;
  32     dst[i*2+1].re[2] = s12.re;
  33     dst[i*2+1].re[3] = s13.re;
  34     dst[i*2+1].im[0] = s10.im;
  35     dst[i*2+1].im[1] = s11.im;
  36     dst[i*2+1].im[2] = s12.im;
  37     dst[i*2+1].im[3] = s13.im;
  38   }
  39 }
  40
  41 /* The first step to produce optimal code is to appropriately detect the
  42    load and store groups.  */
  43 /* { dg-final { scan-tree-dump "Detected interleaving load of size 8" "vect" } } */
  44 /* { dg-final { scan-tree-dump "Detected interleaving store of size 16" "vect" } } */
  45 /* We're not able to peel & apply re-aligning to make accesses well-aligned for !vect_hw_misalign,
  46    but we could by peeling the stores for alignment and applying re-aligning loads.  */
  47 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { xfail { ! vect_hw_misalign } } } } */
  48 /* { dg-final { scan-tree-dump-not "gap of 6 elements" "vect" } } */