gcc/testsuite/gcc.target/gcn/simd-math-4.c

   1 /* Test that signed division and modulus give the correct result with
   2    different variations of signedness.  */
   3
   4 /* Setting it this way ensures the run tests use the same flag as the
   5    compile tests.  */
   6 #pragma GCC optimize("O2")
   7
   8 typedef char v64qi __attribute__ ((vector_size (64)));
   9 typedef short v64hi __attribute__ ((vector_size (128)));
  10 typedef int v64si __attribute__ ((vector_size (256)));
  11 typedef long v64di __attribute__ ((vector_size (512)));
  12
  13 #ifndef TYPE
  14 #define TYPE v64si
  15 #endif
  16 #define N 64
  17
  18 TYPE a;
  19 TYPE b;
  20
  21 int main()
  22 {
  23   int i;
  24   TYPE squot, srem;
  25   TYPE usquot, usrem;
  26   TYPE vquot, vrem;
  27   TYPE vquot2, vrem2;
  28   TYPE refquot, refrem;
  29
  30   for (i = 0; i < 64; i++)
  31     {
  32       a[i] = i * (i >> 2) * (i&1 ? -1 : 1);
  33       b[i] = i * (i&2 ? -1 : 1);
  34     }
  35
  36   for (i = 0; i < N; i++)
  37     {
  38       /* Calculate reference values using regular scalar div and mod.  */
  39       refquot[i] = a[i] / b[i];
  40       __asm__ ("" ::: "memory");
  41       refrem[i] = a[i] % b[i];
  42     }
  43
  44   __asm__ ("" ::: "memory");
  45   /* Scalar with divmod.  */
  46   for (i = 0; i < N; i++)
  47     {
  48       squot[i] = a[i] / b[i];
  49       srem[i] = a[i] % b[i];
  50     }
  51
  52   __asm__ ("" ::: "memory");
  53   /* Vectorized with divmod.  */
  54   vquot = a / b;
  55   vrem = a % b;
  56
  57   __asm__ ("" ::: "memory");
  58   /* Vectorized with separte div and mod.  */
  59   vquot2 = a / b;
  60   __asm__ ("" ::: "memory");
  61   vrem2 = a % b;
  62
  63 #ifdef DEBUG
  64 #define DUMP(VAR) \
  65   __builtin_printf ("%8s: ", #VAR); \
  66   for (i = 0; i < N; i++) \
  67     __builtin_printf ("%d ", (int)VAR[i]); \
  68   __builtin_printf ("\n");
  69   DUMP (refquot)
  70   DUMP (squot)
  71   DUMP (vquot)
  72   DUMP (vquot2)
  73   __builtin_printf ("\n");
  74   DUMP (refrem)
  75   DUMP (srem)
  76   DUMP (vrem)
  77   DUMP (vrem2)
  78   __builtin_printf ("\n");
  79 #endif
  80
  81   for (i = 0; i < N; i++)
  82     if (squot[i] != refquot[i]
  83         || vquot[i] != refquot[i]
  84         || vquot2[i] != refquot[i]
  85         || srem[i] != refrem[i]
  86         || vrem[i] != refrem[i]
  87         || vrem2[i] != refrem[i])
  88       __builtin_abort ();
  89   return 0;
  90 }
  91
  92 /* { dg-final { scan-assembler-times {__divmodv64si4@rel32@lo} 1 { xfail *-*-* } } } */
  93 /* { dg-final { scan-assembler-times {__udivmodv64si4@rel32@lo} 0 } } */
  94 /* { dg-final { scan-assembler-times {__divv64si3@rel32@lo} 1 } } */
  95 /* { dg-final { scan-assembler-times {__udivv64si3@rel32@lo} 0 } } */
  96 /* { dg-final { scan-assembler-times {__modv64si3@rel32@lo} 1 } } */
  97 /* { dg-final { scan-assembler-times {__umodv64si3@rel32@lo} 0 } } */
  98 /* { dg-final { scan-assembler-times {__divsi3@rel32@lo} 1 } } */
  99 /* { dg-final { scan-assembler-times {__udivsi3@rel32@lo} 0 } } */