gcc/testsuite/gcc.target/arm/mve/dlstp-compile-asm-3.c

   1
   2 /* { dg-do compile } */
   3 /* { dg-require-effective-target arm_v8_1m_mve_ok } */
   4 /* { dg-options "-O3 -save-temps" } */
   5 /* { dg-add-options arm_v8_1m_mve } */
   6
   7 #include <arm_mve.h>
   8
   9 /* We don't support pattern recognition of signed N values when computing num_iter.  */
  10 void test3 (uint8_t *a, uint8_t *b, uint8_t *c, int n)
  11 {
  12     int num_iter = (n + 15)/16;
  13     for (int i = 0; i < num_iter; i++)
  14     {
  15         mve_pred16_t p = vctp8q (n);
  16         uint8x16_t va = vldrbq_z_u8 (a, p);
  17         uint8x16_t vb = vldrbq_z_u8 (b, p);
  18         uint8x16_t vc = vaddq_x_u8 (va, vb, p);
  19         vstrbq_p_u8 (c, vc, p);
  20         n-=16;
  21         a += 16;
  22         b += 16;
  23         c += 16;
  24     }
  25 }
  26
  27 /* Using a predicated vcmp to generate a new predicate value in the
  28    loop and then using it in a predicated store insn.  */
  29 void test17 (int32_t *a, int32_t *b, int32_t *c, int n)
  30 {
  31   while (n > 0)
  32     {
  33       mve_pred16_t p = vctp32q (n);
  34       int32x4_t va = vldrwq_z_s32 (a, p);
  35       int32x4_t vb = vldrwq_z_s32 (b, p);
  36       int32x4_t vc = vaddq_s32 (va, vb);
  37       mve_pred16_t p1 = vcmpeqq_m_s32 (va, vc, p);
  38       vstrwq_p_s32 (c, vc, p1);
  39       c += 4;
  40       a += 4;
  41       b += 4;
  42       n -= 4;
  43     }
  44 }
  45 /* This is an example of a loop that we could tail predicate but currently don't.  */
  46 /* { dg-final { scan-assembler "letp" { xfail *-*-* } } } */