| 1 |
/* { dg-require-effective-target arm_neon_fp16_hw { target { arm*-*-* } } } */ |
| 2 |
#include <arm_neon.h> |
| 3 |
#include "arm-neon-ref.h" |
| 4 |
#include "compute-ref-data.h" |
| 5 |
#include <math.h> |
| 6 |
|
| 7 |
/* Expected results for vcvt. */ |
| 8 |
VECT_VAR_DECL (expected,hfloat,32,4) [] = { 0x41800000, 0x41700000, |
| 9 |
0x41600000, 0x41500000 }; |
| 10 |
VECT_VAR_DECL (expected,hfloat,16,4) [] = { 0x3e00, 0x4100, 0x4300, 0x4480 }; |
| 11 |
|
| 12 |
/* Expected results for vcvt_high_f32_f16. */ |
| 13 |
VECT_VAR_DECL (expected_high,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, |
| 14 |
0xc1200000, 0xc1100000 }; |
| 15 |
/* Expected results for vcvt_high_f16_f32. */ |
| 16 |
VECT_VAR_DECL (expected_high,hfloat,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, |
| 17 |
0xcc00, 0xcb80, 0xcb00, 0xca80 }; |
| 18 |
|
| 19 |
void |
| 20 |
exec_vcvt (void) |
| 21 |
{ |
| 22 |
clean_results (); |
| 23 |
|
| 24 |
#define TEST_MSG "vcvt_f32_f16" |
| 25 |
{ |
| 26 |
VECT_VAR_DECL (buffer_src, float, 16, 4) [] = { 16.0, 15.0, 14.0, 13.0 }; |
| 27 |
|
| 28 |
DECL_VARIABLE (vector_src, float, 16, 4); |
| 29 |
|
| 30 |
VLOAD (vector_src, buffer_src, , float, f, 16, 4); |
| 31 |
DECL_VARIABLE (vector_res, float, 32, 4) = |
| 32 |
vcvt_f32_f16 (VECT_VAR (vector_src, float, 16, 4)); |
| 33 |
vst1q_f32 (VECT_VAR (result, float, 32, 4), |
| 34 |
VECT_VAR (vector_res, float, 32, 4)); |
| 35 |
|
| 36 |
CHECK_FP (TEST_MSG, float, 32, 4, PRIx32, expected, ""); |
| 37 |
} |
| 38 |
#undef TEST_MSG |
| 39 |
|
| 40 |
clean_results (); |
| 41 |
|
| 42 |
#define TEST_MSG "vcvt_f16_f32" |
| 43 |
{ |
| 44 |
VECT_VAR_DECL (buffer_src, float, 32, 4) [] = { 1.5, 2.5, 3.5, 4.5 }; |
| 45 |
DECL_VARIABLE (vector_src, float, 32, 4); |
| 46 |
|
| 47 |
VLOAD (vector_src, buffer_src, q, float, f, 32, 4); |
| 48 |
DECL_VARIABLE (vector_res, float, 16, 4) = |
| 49 |
vcvt_f16_f32 (VECT_VAR (vector_src, float, 32, 4)); |
| 50 |
vst1_f16 (VECT_VAR (result, float, 16, 4), |
| 51 |
VECT_VAR (vector_res, float, 16 ,4)); |
| 52 |
|
| 53 |
CHECK_FP (TEST_MSG, float, 16, 4, PRIx16, expected, ""); |
| 54 |
} |
| 55 |
#undef TEST_MSG |
| 56 |
|
| 57 |
/* We run more tests for AArch64 as the relevant intrinsics |
| 58 |
do not exist on AArch32. */ |
| 59 |
#if defined (__aarch64__) |
| 60 |
clean_results (); |
| 61 |
|
| 62 |
#define TEST_MSG "vcvt_high_f32_f16" |
| 63 |
{ |
| 64 |
DECL_VARIABLE (vector_src, float, 16, 8); |
| 65 |
VLOAD (vector_src, buffer, q, float, f, 16, 8); |
| 66 |
DECL_VARIABLE (vector_res, float, 32, 4); |
| 67 |
VECT_VAR (vector_res, float, 32, 4) = |
| 68 |
vcvt_high_f32_f16 (VECT_VAR (vector_src, float, 16, 8)); |
| 69 |
vst1q_f32 (VECT_VAR (result, float, 32, 4), |
| 70 |
VECT_VAR (vector_res, float, 32, 4)); |
| 71 |
CHECK_FP (TEST_MSG, float, 32, 4, PRIx32, expected_high, ""); |
| 72 |
} |
| 73 |
#undef TEST_MSG |
| 74 |
clean_results (); |
| 75 |
|
| 76 |
#define TEST_MSG "vcvt_high_f16_f32" |
| 77 |
{ |
| 78 |
DECL_VARIABLE (vector_low, float, 16, 4); |
| 79 |
VDUP (vector_low, , float, f, 16, 4, 2.0); |
| 80 |
|
| 81 |
DECL_VARIABLE (vector_src, float, 32, 4); |
| 82 |
VLOAD (vector_src, buffer, q, float, f, 32, 4); |
| 83 |
|
| 84 |
DECL_VARIABLE (vector_res, float, 16, 8) = |
| 85 |
vcvt_high_f16_f32 (VECT_VAR (vector_low, float, 16, 4), |
| 86 |
VECT_VAR (vector_src, float, 32, 4)); |
| 87 |
vst1q_f16 (VECT_VAR (result, float, 16, 8), |
| 88 |
VECT_VAR (vector_res, float, 16, 8)); |
| 89 |
|
| 90 |
CHECK_FP (TEST_MSG, float, 16, 8, PRIx16, expected_high, ""); |
| 91 |
} |
| 92 |
#endif |
| 93 |
} |
| 94 |
|
| 95 |
int |
| 96 |
main (void) |
| 97 |
{ |
| 98 |
exec_vcvt (); |
| 99 |
return 0; |
| 100 |
} |