AVX512FP16: Add testcase for vcvtsh2sd/vcvtsh2ss/vcvtsd2sh/vcvtss2sh.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/avx512fp16-vcvtsd2sh-1a.c: New test.
	* gcc.target/i386/avx512fp16-vcvtsd2sh-1b.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtsh2sd-1a.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtsh2sd-1b.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtsh2ss-1a.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtsh2ss-1b.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtss2sh-1a.c: Ditto.
	* gcc.target/i386/avx512fp16-vcvtss2sh-1b.c: Ditto.
This commit is contained in:
liuhongt 2020-03-02 17:25:19 +08:00
parent 90429b962e
commit 1ef291e68f
8 changed files with 336 additions and 0 deletions

View File

@ -0,0 +1,25 @@
/* { dg-do compile } */
/* { dg-options "-mavx512fp16 -O2" } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsd2sh\[ \\t\]+\{rz-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
#include <immintrin.h>
volatile __m128h res, x1;
volatile __m128d x2;
volatile __mmask8 m8;
void extern
avx512f_test (void)
{
res = _mm_cvtsd_sh (x1, x2);
res = _mm_mask_cvtsd_sh (res, m8, x1, x2);
res = _mm_maskz_cvtsd_sh (m8, x1, x2);
res = _mm_cvt_roundsd_sh (x1, x2, 8);
res = _mm_mask_cvt_roundsd_sh (res, m8, x1, x2, 8);
res = _mm_maskz_cvt_roundsd_sh (m8, x1, x2, 11);
}

View File

@ -0,0 +1,60 @@
/* { dg-do run { target avx512fp16 } } */
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
#define AVX512FP16
#include "avx512fp16-helper.h"
#define N_ELEMS 8
void NOINLINE
emulate_vcvtsd2sh(V512 * dest, V512 op1, V512 op2,
__mmask8 k, int zero_mask)
{
V512 v1, v2, v5, v6, v7, v8;
int i;
unpack_ph_2twops(op1, &v1, &v2);
unpack_ph_2twops(*dest, &v7, &v8);
if ((k&1) || !k)
v5.f32[0] = (float)op2.f64[0];
else if (zero_mask)
v5.f32[0] = 0;
else
v5.f32[0] = v7.f32[0];
for (i = 1; i < 8; i++)
v5.f32[i] = v1.f32[i];
*dest = pack_twops_2ph(v5, v6);
}
void
test_512 (void)
{
V512 res;
V512 exp;
init_src();
emulate_vcvtsd2sh(&exp, src1, src2, 0x1, 0);
res.xmmh[0] = _mm_cvt_roundsd_sh(src1.xmmh[0], src2.xmmd[0],
_ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "_mm_cvt_roundsd_sh");
init_dest(&res, &exp);
emulate_vcvtsd2sh(&exp, src1, src2, 0x1, 0);
res.xmmh[0] = _mm_mask_cvt_roundsd_sh(res.xmmh[0], 0x1, src1.xmmh[0],
src2.xmmd[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_mask_cvt_roundsd_sh");
emulate_vcvtsd2sh(&exp, src1, src2, 0x2, 1);
res.xmmh[0] = _mm_maskz_cvt_roundsd_sh(0x2, src1.xmmh[0],
src2.xmmd[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_maskz_cvt_roundsd_sh");
if (n_errs != 0) {
abort ();
}
}

View File

@ -0,0 +1,25 @@
/* { dg-do compile } */
/* { dg-options "-mavx512fp16 -O2" } */
/* { dg-final { scan-assembler-times "vcvtsh2sd\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2sd\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2sd\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2sd\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2sd\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
#include <immintrin.h>
volatile __m128d res;
volatile __m128d x1;
volatile __m128h x2;
volatile __mmask8 m8;
void extern
avx512f_test (void)
{
res = _mm_cvtsh_sd (x1, x2);
res = _mm_mask_cvtsh_sd (res, m8, x1, x2);
res = _mm_maskz_cvtsh_sd (m8, x1, x2);
res = _mm_cvt_roundsh_sd (x1, x2, 8);
res = _mm_mask_cvt_roundsh_sd (res, m8, x1, x2, 8);
res = _mm_maskz_cvt_roundsh_sd (m8, x1, x2, 4);
}

View File

@ -0,0 +1,57 @@
/* { dg-do run { target avx512fp16 } } */
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
#define AVX512FP16
#include "avx512fp16-helper.h"
#define N_ELEMS 8
void NOINLINE
emulate_vcvtsh2sd(V512 * dest, V512 op1, V512 op2,
__mmask8 k, int zero_mask)
{
V512 v1, v2, v3, v4, v5, v6, v7, v8;
unpack_ph_2twops(op2, &v3, &v4);
if ((k&1) || !k)
v5.f64[0] = v3.f32[0];
else if (zero_mask)
v5.f64[0] = 0;
else
v5.f64[0] = dest->f64[0];
v5.f64[1] = op1.f64[1];
*dest = v5;
}
void
test_512 (void)
{
V512 res;
V512 exp;
init_src();
emulate_vcvtsh2sd(&exp, src1, src2, 0x1, 0);
res.xmmd[0] = _mm_cvt_roundsh_sd(src1.xmmd[0], src2.xmmh[0],
_ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "_mm_cvt_roundsh_sd");
init_dest(&res, &exp);
emulate_vcvtsh2sd(&exp, src1, src2, 0x1, 0);
res.xmmd[0] = _mm_mask_cvt_roundsh_sd(res.xmmd[0], 0x1, src1.xmmd[0],
src2.xmmh[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_mask_cvt_roundsh_sd");
emulate_vcvtsh2sd(&exp, src1, src2, 0x2, 1);
res.xmmd[0] = _mm_maskz_cvt_roundsh_sd(0x2, src1.xmmd[0],
src2.xmmh[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_maskz_cvt_roundsh_sd");
if (n_errs != 0) {
abort ();
}
}

View File

@ -0,0 +1,25 @@
/* { dg-do compile } */
/* { dg-options "-mavx512fp16 -O2" } */
/* { dg-final { scan-assembler-times "vcvtsh2ss\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2ss\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2ss\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2ss\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtsh2ss\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
#include <immintrin.h>
volatile __m128 res;
volatile __m128 x1;
volatile __m128h x2;
volatile __mmask8 m8;
void extern
avx512f_test (void)
{
res = _mm_cvtsh_ss (x1, x2);
res = _mm_mask_cvtsh_ss (res, m8, x1, x2);
res = _mm_maskz_cvtsh_ss (m8, x1, x2);
res = _mm_cvt_roundsh_ss (x1, x2, 8);
res = _mm_mask_cvt_roundsh_ss (res, m8, x1, x2, 8);
res = _mm_maskz_cvt_roundsh_ss (m8, x1, x2, 4);
}

View File

@ -0,0 +1,59 @@
/* { dg-do run { target avx512fp16 } } */
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
#define AVX512FP16
#include "avx512fp16-helper.h"
#define N_ELEMS 8
void NOINLINE
emulate_vcvtsh2ss(V512 * dest, V512 op1, V512 op2,
__mmask8 k, int zero_mask)
{
V512 v1, v2, v3, v4, v5, v6, v7, v8;
int i;
unpack_ph_2twops(op2, &v3, &v4);
if ((k&1) || !k)
v5.f32[0] = v3.f32[0];
else if (zero_mask)
v5.f32[0] = 0;
else
v5.f32[0] = dest->f32[0];
for (i = 1; i < 4; i++)
v5.f32[i] = op1.f32[i];
*dest = v5;
}
void
test_512 (void)
{
V512 res;
V512 exp;
init_src();
emulate_vcvtsh2ss(&exp, src1, src2, 0x1, 0);
res.xmm[0] = _mm_cvt_roundsh_ss(src1.xmm[0], src2.xmmh[0],
_ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "_mm_cvt_roundsh_ss");
init_dest(&res, &exp);
emulate_vcvtsh2ss(&exp, src1, src2, 0x1, 0);
res.xmm[0] = _mm_mask_cvt_roundsh_ss(res.xmm[0], 0x1, src1.xmm[0],
src2.xmmh[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_mask_cvt_roundsh_ss");
emulate_vcvtsh2ss(&exp, src1, src2, 0x2, 1);
res.xmm[0] = _mm_maskz_cvt_roundsh_ss(0x2, src1.xmm[0],
src2.xmmh[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_maskz_cvt_roundsh_ss");
if (n_errs != 0) {
abort ();
}
}

View File

@ -0,0 +1,25 @@
/* { dg-do compile } */
/* { dg-options "-mavx512fp16 -O2" } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vcvtss2sh\[ \\t\]+\{rz-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
#include <immintrin.h>
volatile __m128h res, x1;
volatile __m128 x2;
volatile __mmask8 m8;
void extern
avx512f_test (void)
{
res = _mm_cvtss_sh (x1, x2);
res = _mm_mask_cvtss_sh (res, m8, x1, x2);
res = _mm_maskz_cvtss_sh (m8, x1, x2);
res = _mm_cvt_roundss_sh (x1, x2, 8);
res = _mm_mask_cvt_roundss_sh (res, m8, x1, x2, 8);
res = _mm_maskz_cvt_roundss_sh (m8, x1, x2, 11);
}

View File

@ -0,0 +1,60 @@
/* { dg-do run { target avx512fp16 } } */
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
#define AVX512FP16
#include "avx512fp16-helper.h"
#define N_ELEMS 8
void NOINLINE
emulate_vcvtss2sh(V512 * dest, V512 op1, V512 op2,
__mmask8 k, int zero_mask)
{
V512 v1, v2, v5, v6, v7, v8;
int i;
unpack_ph_2twops(op1, &v1, &v2);
unpack_ph_2twops(*dest, &v7, &v8);
if ((k&1) || !k)
v5.f32[0] = op2.f32[0];
else if (zero_mask)
v5.f32[0] = 0;
else
v5.f32[0] = v7.f32[0];
for (i = 1; i < 8; i++)
v5.f32[i] = v1.f32[i];
*dest = pack_twops_2ph(v5, v6);
}
void
test_512 (void)
{
V512 res;
V512 exp;
init_src();
emulate_vcvtss2sh(&exp, src1, src2, 0x1, 0);
res.xmmh[0] = _mm_cvt_roundss_sh(src1.xmmh[0], src2.xmm[0],
_ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "_mm_cvt_roundss_sh");
init_dest(&res, &exp);
emulate_vcvtss2sh(&exp, src1, src2, 0x1, 0);
res.xmmh[0] = _mm_mask_cvt_roundss_sh(res.xmmh[0], 0x1, src1.xmmh[0],
src2.xmm[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_mask_cvt_roundss_sh");
emulate_vcvtss2sh(&exp, src1, src2, 0x2, 1);
res.xmmh[0] = _mm_maskz_cvt_roundss_sh(0x2, src1.xmmh[0],
src2.xmm[0], _ROUND_NINT);
check_results(&res, &exp, N_ELEMS, "mm_maskz_cvt_roundss_sh");
if (n_errs != 0) {
abort ();
}
}