| 1 |
#include <stdint-gcc.h> |
| 2 |
|
| 3 |
typedef int8_t vnx2qi __attribute__ ((vector_size (2))); |
| 4 |
typedef int8_t vnx4qi __attribute__ ((vector_size (4))); |
| 5 |
typedef int8_t vnx8qi __attribute__ ((vector_size (8))); |
| 6 |
typedef int8_t vnx16qi __attribute__ ((vector_size (16))); |
| 7 |
typedef int8_t vnx32qi __attribute__ ((vector_size (32))); |
| 8 |
typedef int8_t vnx64qi __attribute__ ((vector_size (64))); |
| 9 |
typedef int8_t vnx128qi __attribute__ ((vector_size (128))); |
| 10 |
|
| 11 |
typedef int16_t vnx2hi __attribute__ ((vector_size (4))); |
| 12 |
typedef int16_t vnx4hi __attribute__ ((vector_size (8))); |
| 13 |
typedef int16_t vnx8hi __attribute__ ((vector_size (16))); |
| 14 |
typedef int16_t vnx16hi __attribute__ ((vector_size (32))); |
| 15 |
typedef int16_t vnx32hi __attribute__ ((vector_size (64))); |
| 16 |
typedef int16_t vnx64hi __attribute__ ((vector_size (128))); |
| 17 |
|
| 18 |
typedef int32_t vnx2si __attribute__ ((vector_size (8))); |
| 19 |
typedef int32_t vnx4si __attribute__ ((vector_size (16))); |
| 20 |
typedef int32_t vnx8si __attribute__ ((vector_size (32))); |
| 21 |
typedef int32_t vnx16si __attribute__ ((vector_size (64))); |
| 22 |
typedef int32_t vnx32si __attribute__ ((vector_size (128))); |
| 23 |
|
| 24 |
typedef int64_t vnx2di __attribute__ ((vector_size (16))); |
| 25 |
typedef int64_t vnx4di __attribute__ ((vector_size (32))); |
| 26 |
typedef int64_t vnx8di __attribute__ ((vector_size (64))); |
| 27 |
typedef int64_t vnx16di __attribute__ ((vector_size (128))); |
| 28 |
|
| 29 |
typedef float vnx2sf __attribute__ ((vector_size (8))); |
| 30 |
typedef float vnx4sf __attribute__ ((vector_size (16))); |
| 31 |
typedef float vnx8sf __attribute__ ((vector_size (32))); |
| 32 |
typedef float vnx16sf __attribute__ ((vector_size (64))); |
| 33 |
typedef float vnx32sf __attribute__ ((vector_size (128))); |
| 34 |
|
| 35 |
typedef double vnx2df __attribute__ ((vector_size (16))); |
| 36 |
typedef double vnx4df __attribute__ ((vector_size (32))); |
| 37 |
typedef double vnx8df __attribute__ ((vector_size (64))); |
| 38 |
typedef double vnx16df __attribute__ ((vector_size (128))); |
| 39 |
|
| 40 |
#define INIT_PERMUTE(NUNITS, NUM1, NUM2, TYPE) \ |
| 41 |
TYPE v_##TYPE##_in1; \ |
| 42 |
TYPE v_##TYPE##_in2; \ |
| 43 |
TYPE v_##TYPE##_out = {0}; \ |
| 44 |
for (int i = 0; i < NUNITS; i++) \ |
| 45 |
{ \ |
| 46 |
v_##TYPE##_in1[i] = i * NUM1 + NUM2; \ |
| 47 |
v_##TYPE##_in2[i] = i * NUM1 - NUM2; \ |
| 48 |
} |
| 49 |
|
| 50 |
#define CHECK_PERMUTE_SINGLE(NUNITS, VALUE, TYPE) \ |
| 51 |
for (int i = 0; i < NUNITS; i++) \ |
| 52 |
if (v_##TYPE##_out[i] != VALUE) \ |
| 53 |
__builtin_abort (); |
| 54 |
|
| 55 |
#define CHECK_PERMUTE_REVERSE(NUNITS, TYPE) \ |
| 56 |
for (int i = 0; i < NUNITS; i++) \ |
| 57 |
if (v_##TYPE##_out[i] != v_##TYPE##_in1[NUNITS - 1 - i]) \ |
| 58 |
__builtin_abort (); |
| 59 |
|
| 60 |
#define CHECK_PERMUTE_DOUBLE(NUNITS, TYPE) \ |
| 61 |
for (int i = 0; i < NUNITS; i++) \ |
| 62 |
{ \ |
| 63 |
int new_index = i * 2; \ |
| 64 |
if (new_index < NUNITS \ |
| 65 |
&& v_##TYPE##_out[i] != v_##TYPE##_in1[new_index]) \ |
| 66 |
__builtin_abort (); \ |
| 67 |
if (new_index >= NUNITS \ |
| 68 |
&& v_##TYPE##_out[i] != v_##TYPE##_in2[new_index % NUNITS]) \ |
| 69 |
__builtin_abort (); \ |
| 70 |
} |