gcc/testsuite/gcc.target/aarch64/mla_intrinsic_1.c

   1 /* { dg-do run } */
   2 /* { dg-options "-O3 --save-temps" } */
   3
   4 #include <arm_neon.h>
   5
   6 extern void abort (void);
   7
   8 #define MAPs(size, xx) int##size##xx##_t
   9 #define MAPu(size, xx) uint##size##xx##_t
  10
  11
  12 #define TEST_VMLA(q, su, size, in1_lanes, in2_lanes)            \
  13 static void                                                     \
  14 test_vmlaq_lane##q##_##su##size (MAP##su (size, ) * res,        \
  15                                  const MAP##su(size, ) *in1,    \
  16                                  const MAP##su(size, ) *in2)    \
  17 {                                                               \
  18   MAP##su (size, x##in1_lanes) a = vld1q_##su##size (res);      \
  19   MAP##su (size, x##in1_lanes) b = vld1q_##su##size (in1);      \
  20   MAP##su (size, x##in2_lanes) c = vld1##q##_##su##size (in2);  \
  21   a = vmlaq_lane##q##_##su##size (a, b, c, 1);                  \
  22   vst1q_##su##size (res, a);                                    \
  23 }
  24
  25 #define BUILD_VARS(width, n_lanes, n_half_lanes)                \
  26 TEST_VMLA (, s, width, n_lanes, n_half_lanes)                   \
  27 TEST_VMLA (q, s, width, n_lanes, n_lanes)                       \
  28 TEST_VMLA (, u, width, n_lanes, n_half_lanes)                   \
  29 TEST_VMLA (q, u, width, n_lanes, n_lanes)                       \
  30
  31 BUILD_VARS (32, 4, 2)
  32 BUILD_VARS (16, 8, 4)
  33
  34 #define POOL4 {0, 1, 2, 3}
  35 #define POOL8 {0, 1, 2, 3, 4, 5, 6, 7}
  36 #define EMPTY4 {0, 0, 0, 0}
  37 #define EMPTY8 {0, 0, 0, 0, 0, 0, 0, 0}
  38
  39 #define BUILD_TEST(su, size, lanes)                             \
  40 static void                                                     \
  41 test_##su##size (void)                                          \
  42 {                                                               \
  43   int i;                                                        \
  44   MAP##su (size,) pool[lanes] = POOL##lanes;                    \
  45   MAP##su (size,) res[lanes] = EMPTY##lanes;                    \
  46   MAP##su (size,) res2[lanes] = EMPTY##lanes;                   \
  47                                                                 \
  48   /* Forecfully avoid optimization.  */                         \
  49   asm volatile ("" : : : "memory");                             \
  50   test_vmlaq_lane_##su##size (res, pool, pool);                 \
  51   for (i = 0; i < lanes; i++)                                   \
  52     if (res[i] != pool[i])                                      \
  53       abort ();                                                 \
  54                                                                 \
  55   /* Forecfully avoid optimization.  */                         \
  56   asm volatile ("" : : : "memory");                             \
  57   test_vmlaq_laneq_##su##size (res2, pool, pool);               \
  58   for (i = 0; i < lanes; i++)                                   \
  59     if (res2[i] != pool[i])                                     \
  60       abort ();                                                 \
  61 }
  62
  63 #undef BUILD_VARS
  64 #define BUILD_VARS(size, lanes)                                 \
  65 BUILD_TEST (s, size, lanes)                                     \
  66 BUILD_TEST (u, size, lanes)
  67
  68 BUILD_VARS (32, 4)
  69 BUILD_VARS (16, 8)
  70
  71 int
  72 main (int argc, char **argv)
  73 {
  74   test_s32 ();
  75   test_u32 ();
  76   test_s16 ();
  77   test_u16 ();
  78   return 0;
  79 }
  80
  81 /* { dg-final { scan-assembler-times "mla\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s, v\[0-9\]+\.4s\\\[\[0-9\]+\\\]" 4 } } */
  82 /* { dg-final { scan-assembler-times "mla\\tv\[0-9\]+\.8h, v\[0-9\]+\.8h, v\[0-9\]+\.8h\\\[\[0-9\]+\\\]" 4 } } */
  83 /* { dg-final { cleanup-saved-temps } } */
  84