| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-mavx512f -O2" } */ |
| 3 |
/* { dg-require-effective-target avx512f } */ |
| 4 |
|
| 5 |
#include "avx512f-check.h" |
| 6 |
|
| 7 |
static void |
| 8 |
init_vpadd_mask (int* dst, int *src1, int *src2, int seed) |
| 9 |
{ |
| 10 |
int i; |
| 11 |
|
| 12 |
for (i = 0; i < 16; i++) |
| 13 |
{ |
| 14 |
dst[i] = -1; |
| 15 |
src1[i] = seed * 2 * i + 1; |
| 16 |
src2[i] = seed * 2 * i; |
| 17 |
} |
| 18 |
} |
| 19 |
|
| 20 |
static inline void |
| 21 |
calc_vpadd_mask_zeroed (int *dst, __mmask16 m, int *src1, int *src2) |
| 22 |
{ |
| 23 |
int i; |
| 24 |
|
| 25 |
for (i = 0; i < 16; i++) |
| 26 |
{ |
| 27 |
if (m & (1 << i)) |
| 28 |
dst[i] = src1[i] + src2[i]; |
| 29 |
else |
| 30 |
dst[i] = 0; |
| 31 |
} |
| 32 |
} |
| 33 |
|
| 34 |
void static |
| 35 |
avx512f_test (void) |
| 36 |
{ |
| 37 |
/* Checking mask arithmetic instruction */ |
| 38 |
|
| 39 |
__mmask16 msk_dst, msk_src1, msk_src2, msk_dst_ref; |
| 40 |
|
| 41 |
msk_src1 = 0x0FFB; |
| 42 |
msk_src2 = 0x0F0F; |
| 43 |
|
| 44 |
asm ("kandw\t%2, %1, %0" |
| 45 |
: "=k" (msk_dst) |
| 46 |
: "k" (msk_src1), "k" (msk_src2)); |
| 47 |
|
| 48 |
msk_dst_ref = _mm512_kand (msk_src1, msk_src2); |
| 49 |
if (msk_dst != msk_dst_ref) |
| 50 |
abort (); |
| 51 |
|
| 52 |
|
| 53 |
/* Checking zero-masked vector instruction */ |
| 54 |
union512i_d dst, src1, src2; |
| 55 |
int dst_ref[16]; |
| 56 |
|
| 57 |
init_vpadd_mask (dst.a, src1.a, src2.a, 1); |
| 58 |
init_vpadd_mask (dst_ref, src1.a, src2.a, 1); |
| 59 |
|
| 60 |
asm ("vpaddd\t%2, %1, %0 %{%3%}%{z%}" |
| 61 |
: "=x" (dst.x) |
| 62 |
: "x" (src1.x), "x" (src2.x), "Yk" (msk_dst)); |
| 63 |
|
| 64 |
calc_vpadd_mask_zeroed (dst_ref, msk_dst, src1.a, src2.a); |
| 65 |
|
| 66 |
if (check_union512i_d (dst, dst_ref)) |
| 67 |
abort (); |
| 68 |
} |