gcc/testsuite/gcc.target/powerpc/pr71805.c

   1 /* { dg-require-effective-target p9vector_hw } */
   2 /* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
   3 /* { dg-options "-mcpu=power9 -O3 --param tree-reassoc-width=1" } */
   4
   5 /* Originally from gcc.dg/vect/pr45752.c.  */
   6 #include <stdarg.h>
   7
   8 #ifdef __cplusplus
   9 extern "C" {
  10 #endif
  11 extern void abort (void);
  12 extern void exit (int);
  13 #ifdef __cplusplus
  14 }
  15 #endif
  16
  17 #define M00 100
  18 #define M10 216
  19 #define M20 23
  20 #define M30 237
  21 #define M40 437
  22
  23 #define M01 1322
  24 #define M11 13
  25 #define M21 27271
  26 #define M31 2280
  27 #define M41 284
  28
  29 #define M02 74
  30 #define M12 191
  31 #define M22 500
  32 #define M32 111
  33 #define M42 1114
  34
  35 #define M03 134
  36 #define M13 117
  37 #define M23 11
  38 #define M33 771
  39 #define M43 71
  40
  41 #define M04 334
  42 #define M14 147
  43 #define M24 115
  44 #define M34 7716
  45 #define M44 16
  46
  47 #define N 20
  48
  49 void foo (unsigned int *__restrict__ pInput,
  50           unsigned int *__restrict__ pOutput,
  51           unsigned int *__restrict__ pInput2,
  52           unsigned int *__restrict__ pOutput2)
  53 {
  54   unsigned int i, a, b, c, d, e;
  55
  56   for (i = 0; i < N / 5; i++)
  57     {
  58        a = *pInput++;
  59        b = *pInput++;
  60        c = *pInput++;
  61        d = *pInput++;
  62        e = *pInput++;
  63
  64        *pOutput++ = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  65        *pOutput++ = M10 * a + M11 * b + M12 * c + M13 * d + M14 * e;
  66        *pOutput++ = M20 * a + M21 * b + M22 * c + M23 * d + M24 * e;
  67        *pOutput++ = M30 * a + M31 * b + M32 * c + M33 * d + M34 * e;
  68        *pOutput++ = M40 * a + M41 * b + M42 * c + M43 * d + M44 * e;
  69
  70
  71        a = *pInput2++;
  72        b = *pInput2++;
  73        c = *pInput2++;
  74        d = *pInput2++;
  75        e = *pInput2++;
  76
  77        *pOutput2++ = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  78        *pOutput2++ = M10 * a + M11 * b + M12 * c + M13 * d + M14 * e;
  79        *pOutput2++ = M20 * a + M21 * b + M22 * c + M23 * d + M24 * e;
  80        *pOutput2++ = M30 * a + M31 * b + M32 * c + M33 * d + M34 * e;
  81        *pOutput2++ = M40 * a + M41 * b + M42 * c + M43 * d + M44 * e;
  82
  83     }
  84 }
  85
  86 int main (int argc, const char* argv[])
  87 {
  88   unsigned int input[N], output[N], i, input2[N], output2[N];
  89   unsigned int check_results[N]
  90     = {3208, 1334, 28764, 35679, 2789, 13028, 4754, 168364, 91254, 12399,
  91     22848, 8174, 307964, 146829, 22009, 32668, 11594, 447564, 202404, 31619 };
  92   unsigned int check_results2[N]
  93     = {7136, 2702, 84604, 57909, 6633, 16956, 6122, 224204, 113484, 16243,
  94     26776, 9542, 363804, 169059, 25853, 36596, 12962, 503404, 224634, 35463 };
  95
  96   for (i = 0; i < N; i++)
  97     {
  98       input[i] = i%256;
  99       input2[i] = i + 2;
 100       output[i] = 0;
 101       output2[i] = 0;
 102       __asm__ volatile ("");
 103     }
 104
 105   foo (input, output, input2, output2);
 106
 107   for (i = 0; i < N; i++)
 108     if (output[i] != check_results[i]
 109         || output2[i] != check_results2[i])
 110       abort ();
 111
 112   return 0;
 113 }