1  /* { dg-do run } */
       2  /* { dg-options "-O3 --save-temps" } */
       3  
       4  #include <arm_neon.h>
       5  
       6  extern void abort (void);
       7  extern float fabsf (float);
       8  extern double fabs (double);
       9  
      10  #define NUM_TESTS 8
      11  
      12  float input_s1[] = {0.1f, -0.1f, 0.4f, 10.3f, 200.0f, -800.0f, -13.0f, -0.5f};
      13  float input_s2[] = {-0.2f, 0.4f, 0.04f, -100.3f, 2.0f, -80.0f, 13.0f, -0.5f};
      14  double input_d1[] = {0.1, -0.1, 0.4, 10.3, 200.0, -800.0, -13.0, -0.5};
      15  double input_d2[] = {-0.2, 0.4, 0.04, -100.3, 2.0, -80.0, 13.0, -0.5};
      16  
      17  #define TEST(T, CMP, SUFFIX, WIDTH, LANES, Q, F)			\
      18  int									\
      19  test_vca##T##_float##WIDTH##x##LANES##_t (void)				\
      20  {									\
      21    int ret = 0;								\
      22    int i = 0;								\
      23    uint##WIDTH##_t output[NUM_TESTS];					\
      24  									\
      25    for (i = 0; i < NUM_TESTS; i++)					\
      26      {									\
      27        float##WIDTH##_t f1 = fabs##F (input_##SUFFIX##1[i]);		\
      28        float##WIDTH##_t f2 = fabs##F (input_##SUFFIX##2[i]);		\
      29        /* Inhibit optimization of our linear test loop.  */		\
      30        asm volatile ("" : : : "memory");					\
      31        output[i] = f1 CMP f2 ? -1 : 0;					\
      32      }									\
      33  									\
      34    for (i = 0; i < NUM_TESTS; i += LANES)				\
      35      {									\
      36        float##WIDTH##x##LANES##_t in1 =					\
      37  	vld1##Q##_f##WIDTH (input_##SUFFIX##1 + i);			\
      38        float##WIDTH##x##LANES##_t in2 =					\
      39  	vld1##Q##_f##WIDTH (input_##SUFFIX##2 + i);			\
      40        uint##WIDTH##x##LANES##_t expected_out =				\
      41  	vld1##Q##_u##WIDTH (output + i);				\
      42        uint##WIDTH##x##LANES##_t out =					\
      43  	veor##Q##_u##WIDTH (vca##T##Q##_f##WIDTH (in1, in2),		\
      44  			    expected_out);				\
      45        vst1##Q##_u##WIDTH (output + i, out);				\
      46      }									\
      47  									\
      48    for (i = 0; i < NUM_TESTS; i++)					\
      49      ret |= output[i];							\
      50  									\
      51    return ret;								\
      52  }
      53  
      54  #define BUILD_VARIANTS(T, CMP)	\
      55  TEST (T, CMP, s, 32, 2,  , f)	\
      56  TEST (T, CMP, s, 32, 4, q, f)	\
      57  TEST (T, CMP, d, 64, 2, q,  )
      58  
      59  BUILD_VARIANTS (ge, >=)
      60  /* { dg-final { scan-assembler "facge\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
      61  /* { dg-final { scan-assembler "facge\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
      62  /* { dg-final { scan-assembler "facge\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
      63  
      64  BUILD_VARIANTS (gt, >)
      65  /* { dg-final { scan-assembler "facgt\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
      66  /* { dg-final { scan-assembler "facgt\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
      67  /* { dg-final { scan-assembler "facgt\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
      68  
      69  /* No need for another scan-assembler as these tests
      70     also generate facge, facgt instructions.  */
      71  BUILD_VARIANTS (le, <=)
      72  BUILD_VARIANTS (lt, <)
      73  
      74  #undef TEST
      75  #define TEST(T, CMP, SUFFIX, WIDTH, LANES, Q, F)	\
      76  if (test_vca##T##_float##WIDTH##x##LANES##_t ())	\
      77    abort ();
      78  
      79  int
      80  main (int argc, char **argv)
      81  {
      82  BUILD_VARIANTS (ge, >=)
      83  BUILD_VARIANTS (gt, >)
      84  BUILD_VARIANTS (le, <=)
      85  BUILD_VARIANTS (lt, <)
      86    return 0;
      87  }
      88