| 1 |
/* Test the `v[min|max]{nm}{q}_f*' AArch64 SIMD intrinsic. */ |
| 2 |
|
| 3 |
/* { dg-do run } */ |
| 4 |
/* { dg-options "-O2" } */ |
| 5 |
|
| 6 |
#include "arm_neon.h" |
| 7 |
|
| 8 |
extern void abort (); |
| 9 |
|
| 10 |
#define CHECK(R, E) \ |
| 11 |
{\ |
| 12 |
if (__builtin_memcmp (&R, &E, sizeof (R)) != 0)\ |
| 13 |
abort ();\ |
| 14 |
} |
| 15 |
|
| 16 |
int |
| 17 |
main (int argc, char **argv) |
| 18 |
{ |
| 19 |
/* v{min|max}nm_f32 normal. */ |
| 20 |
float32x2_t f32x2_input1 = vdup_n_f32 (-1.0); |
| 21 |
float32x2_t f32x2_input2 = vdup_n_f32 (0.0); |
| 22 |
float32x2_t f32x2_exp_minnm = vdup_n_f32 (-1.0); |
| 23 |
float32x2_t f32x2_exp_maxnm = vdup_n_f32 (0.0); |
| 24 |
float32x2_t f32x2_ret_minnm = vminnm_f32 (f32x2_input1, f32x2_input2); |
| 25 |
float32x2_t f32x2_ret_maxnm = vmaxnm_f32 (f32x2_input1, f32x2_input2); |
| 26 |
|
| 27 |
CHECK (f32x2_ret_minnm, f32x2_exp_minnm); |
| 28 |
CHECK (f32x2_ret_maxnm, f32x2_exp_maxnm); |
| 29 |
|
| 30 |
/* v{min|max}nm_f32 NaN. */ |
| 31 |
f32x2_input1 = vdup_n_f32 (__builtin_nanf ("")); |
| 32 |
f32x2_input2 = vdup_n_f32 (1.0); |
| 33 |
f32x2_exp_minnm = vdup_n_f32 (1.0); |
| 34 |
f32x2_exp_maxnm = vdup_n_f32 (1.0); |
| 35 |
f32x2_ret_minnm = vminnm_f32 (f32x2_input1, f32x2_input2); |
| 36 |
f32x2_ret_maxnm = vmaxnm_f32 (f32x2_input1, f32x2_input2); |
| 37 |
|
| 38 |
CHECK (f32x2_ret_minnm, f32x2_exp_minnm); |
| 39 |
CHECK (f32x2_ret_maxnm, f32x2_exp_maxnm); |
| 40 |
|
| 41 |
/* v{min|max}nmq_f32 normal. */ |
| 42 |
float32x4_t f32x4_input1 = vdupq_n_f32 (-1024.0); |
| 43 |
float32x4_t f32x4_input2 = vdupq_n_f32 (77.0); |
| 44 |
float32x4_t f32x4_exp_minnm = vdupq_n_f32 (-1024.0); |
| 45 |
float32x4_t f32x4_exp_maxnm = vdupq_n_f32 (77.0); |
| 46 |
float32x4_t f32x4_ret_minnm = vminnmq_f32 (f32x4_input1, f32x4_input2); |
| 47 |
float32x4_t f32x4_ret_maxnm = vmaxnmq_f32 (f32x4_input1, f32x4_input2); |
| 48 |
|
| 49 |
CHECK (f32x4_ret_minnm, f32x4_exp_minnm); |
| 50 |
CHECK (f32x4_ret_maxnm, f32x4_exp_maxnm); |
| 51 |
|
| 52 |
/* v{min|max}nmq_f32 NaN. */ |
| 53 |
f32x4_input1 = vdupq_n_f32 (-__builtin_nanf ("")); |
| 54 |
f32x4_input2 = vdupq_n_f32 (-1.0); |
| 55 |
f32x4_exp_minnm = vdupq_n_f32 (-1.0); |
| 56 |
f32x4_exp_maxnm = vdupq_n_f32 (-1.0); |
| 57 |
f32x4_ret_minnm = vminnmq_f32 (f32x4_input1, f32x4_input2); |
| 58 |
f32x4_ret_maxnm = vmaxnmq_f32 (f32x4_input1, f32x4_input2); |
| 59 |
|
| 60 |
CHECK (f32x4_ret_minnm, f32x4_exp_minnm); |
| 61 |
CHECK (f32x4_ret_maxnm, f32x4_exp_maxnm); |
| 62 |
|
| 63 |
/* v{min|max}nm_f64 normal. */ |
| 64 |
float64x1_t f64x1_input1 = vdup_n_f64 (1.23); |
| 65 |
float64x1_t f64x1_input2 = vdup_n_f64 (4.56); |
| 66 |
float64x1_t f64x1_exp_minnm = vdup_n_f64 (1.23); |
| 67 |
float64x1_t f64x1_exp_maxnm = vdup_n_f64 (4.56); |
| 68 |
float64x1_t f64x1_ret_minnm = vminnm_f64 (f64x1_input1, f64x1_input2); |
| 69 |
float64x1_t f64x1_ret_maxnm = vmaxnm_f64 (f64x1_input1, f64x1_input2); |
| 70 |
CHECK (f64x1_ret_minnm, f64x1_exp_minnm); |
| 71 |
CHECK (f64x1_ret_maxnm, f64x1_exp_maxnm); |
| 72 |
|
| 73 |
/* v{min|max}_f64 normal. */ |
| 74 |
float64x1_t f64x1_exp_min = vdup_n_f64 (1.23); |
| 75 |
float64x1_t f64x1_exp_max = vdup_n_f64 (4.56); |
| 76 |
float64x1_t f64x1_ret_min = vmin_f64 (f64x1_input1, f64x1_input2); |
| 77 |
float64x1_t f64x1_ret_max = vmax_f64 (f64x1_input1, f64x1_input2); |
| 78 |
CHECK (f64x1_ret_min, f64x1_exp_min); |
| 79 |
CHECK (f64x1_ret_max, f64x1_exp_max); |
| 80 |
|
| 81 |
/* v{min|max}nmq_f64 normal. */ |
| 82 |
float64x2_t f64x2_input1 = vdupq_n_f64 (1.23); |
| 83 |
float64x2_t f64x2_input2 = vdupq_n_f64 (4.56); |
| 84 |
float64x2_t f64x2_exp_minnm = vdupq_n_f64 (1.23); |
| 85 |
float64x2_t f64x2_exp_maxnm = vdupq_n_f64 (4.56); |
| 86 |
float64x2_t f64x2_ret_minnm = vminnmq_f64 (f64x2_input1, f64x2_input2); |
| 87 |
float64x2_t f64x2_ret_maxnm = vmaxnmq_f64 (f64x2_input1, f64x2_input2); |
| 88 |
CHECK (f64x2_ret_minnm, f64x2_exp_minnm); |
| 89 |
CHECK (f64x2_ret_maxnm, f64x2_exp_maxnm); |
| 90 |
|
| 91 |
/* v{min|max}nm_f64 NaN. */ |
| 92 |
f64x1_input1 = vdup_n_f64 (-__builtin_nanf ("")); |
| 93 |
f64x1_input2 = vdup_n_f64 (1.0); |
| 94 |
f64x1_exp_minnm = vdup_n_f64 (1.0); |
| 95 |
f64x1_exp_maxnm = vdup_n_f64 (1.0); |
| 96 |
f64x1_ret_minnm = vminnm_f64 (f64x1_input1, f64x1_input2); |
| 97 |
f64x1_ret_maxnm = vmaxnm_f64 (f64x1_input1, f64x1_input2); |
| 98 |
|
| 99 |
CHECK (f64x1_ret_minnm, f64x1_exp_minnm); |
| 100 |
CHECK (f64x1_ret_maxnm, f64x1_exp_maxnm); |
| 101 |
|
| 102 |
/* v{min|max}_f64 NaN. */ |
| 103 |
f64x1_input1 = vdup_n_f64 (-__builtin_nanf ("")); |
| 104 |
f64x1_input2 = vdup_n_f64 (1.0); |
| 105 |
f64x1_exp_minnm = vdup_n_f64 (-__builtin_nanf ("")); |
| 106 |
f64x1_exp_maxnm = vdup_n_f64 (-__builtin_nanf ("")); |
| 107 |
f64x1_ret_minnm = vmin_f64 (f64x1_input1, f64x1_input2); |
| 108 |
f64x1_ret_maxnm = vmax_f64 (f64x1_input1, f64x1_input2); |
| 109 |
|
| 110 |
CHECK (f64x1_ret_minnm, f64x1_exp_minnm); |
| 111 |
CHECK (f64x1_ret_maxnm, f64x1_exp_maxnm); |
| 112 |
|
| 113 |
/* v{min|max}nmq_f64 NaN. */ |
| 114 |
f64x2_input1 = vdupq_n_f64 (-__builtin_nan ("")); |
| 115 |
f64x2_input2 = vdupq_n_f64 (1.0); |
| 116 |
f64x2_exp_minnm = vdupq_n_f64 (1.0); |
| 117 |
f64x2_exp_maxnm = vdupq_n_f64 (1.0); |
| 118 |
f64x2_ret_minnm = vminnmq_f64 (f64x2_input1, f64x2_input2); |
| 119 |
f64x2_ret_maxnm = vmaxnmq_f64 (f64x2_input1, f64x2_input2); |
| 120 |
|
| 121 |
CHECK (f64x2_ret_minnm, f64x2_exp_minnm); |
| 122 |
CHECK (f64x2_ret_maxnm, f64x2_exp_maxnm); |
| 123 |
|
| 124 |
return 0; |
| 125 |
} |