gcc/testsuite/gcc.target/aarch64/sve/reduc_2.c

   1 /* { dg-do compile } */
   2 /* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
   3
   4 #include <stdint.h>
   5
   6 #define NUM_ELEMS(TYPE) (1024 / sizeof (TYPE))
   7
   8 #define DEF_REDUC_PLUS(TYPE)                                    \
   9 void __attribute__ ((noinline, noclone))                        \
  10 reduc_plus_##TYPE (TYPE (*restrict a)[NUM_ELEMS (TYPE)],        \
  11                    TYPE *restrict r, int n)                     \
  12 {                                                               \
  13   for (int i = 0; i < n; i++)                                   \
  14     {                                                           \
  15       r[i] = 0;                                                 \
  16       for (int j = 0; j < NUM_ELEMS (TYPE); j++)                \
  17         r[i] += a[i][j];                                        \
  18     }                                                           \
  19 }
  20
  21 #define TEST_PLUS(T)                            \
  22   T (int8_t)                                    \
  23   T (int16_t)                                   \
  24   T (int32_t)                                   \
  25   T (int64_t)                                   \
  26   T (uint8_t)                                   \
  27   T (uint16_t)                                  \
  28   T (uint32_t)                                  \
  29   T (uint64_t)                                  \
  30   T (_Float16)                                  \
  31   T (float)                                     \
  32   T (double)
  33
  34 TEST_PLUS (DEF_REDUC_PLUS)
  35
  36 #define DEF_REDUC_MAXMIN(TYPE, NAME, CMP_OP)                    \
  37 void __attribute__ ((noinline, noclone))                        \
  38 reduc_##NAME##_##TYPE (TYPE (*restrict a)[NUM_ELEMS (TYPE)],    \
  39                        TYPE *restrict r, int n)                 \
  40 {                                                               \
  41   for (int i = 0; i < n; i++)                                   \
  42     {                                                           \
  43       r[i] = a[i][0];                                           \
  44       for (int j = 0; j < NUM_ELEMS (TYPE); j++)                \
  45         r[i] = a[i][j] CMP_OP r[i] ? a[i][j] : r[i];            \
  46     }                                                           \
  47 }
  48
  49 #define TEST_MAXMIN(T)                          \
  50   T (int8_t, max, >)                            \
  51   T (int16_t, max, >)                           \
  52   T (int32_t, max, >)                           \
  53   T (int64_t, max, >)                           \
  54   T (uint8_t, max, >)                           \
  55   T (uint16_t, max, >)                          \
  56   T (uint32_t, max, >)                          \
  57   T (uint64_t, max, >)                          \
  58   T (_Float16, max, >)                          \
  59   T (float, max, >)                             \
  60   T (double, max, >)                            \
  61                                                 \
  62   T (int8_t, min, <)                            \
  63   T (int16_t, min, <)                           \
  64   T (int32_t, min, <)                           \
  65   T (int64_t, min, <)                           \
  66   T (uint8_t, min, <)                           \
  67   T (uint16_t, min, <)                          \
  68   T (uint32_t, min, <)                          \
  69   T (uint64_t, min, <)                          \
  70   T (_Float16, min, <)                          \
  71   T (float, min, <)                             \
  72   T (double, min, <)
  73
  74 TEST_MAXMIN (DEF_REDUC_MAXMIN)
  75
  76 /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */
  77 /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  78 /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.s\n} 2 } } */
  79 /* { dg-final { scan-assembler-times {\tuaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 2 } } */
  80 /* { dg-final { scan-assembler-times {\tfaddv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  81 /* { dg-final { scan-assembler-times {\tfaddv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
  82 /* { dg-final { scan-assembler-times {\tfaddv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
  83
  84 /* { dg-final { scan-assembler-times {\tsmaxv\tb[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */
  85 /* { dg-final { scan-assembler-times {\tsmaxv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  86 /* { dg-final { scan-assembler-times {\tsmaxv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
  87 /* { dg-final { scan-assembler-times {\tsmaxv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
  88 /* { dg-final { scan-assembler-times {\tumaxv\tb[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */
  89 /* { dg-final { scan-assembler-times {\tumaxv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  90 /* { dg-final { scan-assembler-times {\tumaxv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
  91 /* { dg-final { scan-assembler-times {\tumaxv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
  92 /* { dg-final { scan-assembler-times {\tfmaxnmv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  93 /* { dg-final { scan-assembler-times {\tfmaxnmv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
  94 /* { dg-final { scan-assembler-times {\tfmaxnmv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
  95
  96 /* { dg-final { scan-assembler-times {\tsminv\tb[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */
  97 /* { dg-final { scan-assembler-times {\tsminv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
  98 /* { dg-final { scan-assembler-times {\tsminv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
  99 /* { dg-final { scan-assembler-times {\tsminv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
 100 /* { dg-final { scan-assembler-times {\tuminv\tb[0-9]+, p[0-7], z[0-9]+\.b\n} 1 } } */
 101 /* { dg-final { scan-assembler-times {\tuminv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
 102 /* { dg-final { scan-assembler-times {\tuminv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
 103 /* { dg-final { scan-assembler-times {\tuminv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */
 104 /* { dg-final { scan-assembler-times {\tfminnmv\th[0-9]+, p[0-7], z[0-9]+\.h\n} 1 } } */
 105 /* { dg-final { scan-assembler-times {\tfminnmv\ts[0-9]+, p[0-7], z[0-9]+\.s\n} 1 } } */
 106 /* { dg-final { scan-assembler-times {\tfminnmv\td[0-9]+, p[0-7], z[0-9]+\.d\n} 1 } } */