@@ -25,13 +25,17 @@ typedef union
{
__m512 zmm;
__m512h zmmh;
+ __m512i zmmi;
__m256 ymm[2];
__m256h ymmh[2];
__m256i ymmi[2];
__m128h xmmh[4];
__m128 xmm[4];
+ __m128i xmmi[4];
unsigned short u16[32];
unsigned int u32[16];
+ long long s64[8];
+ unsigned long long u64[8];
float f32[16];
_Float16 f16[32];
} V512;
@@ -162,9 +166,9 @@ init_src()
int i;
for (i = 0; i < AVX512F_MAX_ELEM; i++) {
- v1.f32[i] = -i + 1;
+ v1.f32[i] = i + 1;
v2.f32[i] = i * 0.5f;
- v3.f32[i] = i * 2.5f;
+ v3.f32[i] = i * 1.5f;
v4.f32[i] = i - 0.5f;
src3.u32[i] = (i + 1) * 10;
@@ -217,30 +221,45 @@ init_dest(V512 * res, V512 * exp)
#if AVX512F_LEN == 256
#undef HF
#undef SF
+#undef SI
+#undef H_HF
#undef NET_MASK
-#undef MASK_VALUE
+#undef MASK_VALUE
+#undef HALF_MASK
#undef ZMASK_VALUE
#define NET_MASK 0xffff
#define MASK_VALUE 0xcccc
#define ZMASK_VALUE 0xfcc1
+#define HALF_MASK 0xcc
#define HF(x) x.ymmh[0]
+#define H_HF(x) x.xmmh[0]
#define SF(x) x.ymm[0]
+#define SI(x) x.ymmi[0]
#elif AVX512F_LEN == 128
#undef HF
#undef SF
+#undef SI
+#undef H_HF
#undef NET_MASK
#undef MASK_VALUE
#undef ZMASK_VALUE
+#undef HALF_MASK
#define NET_MASK 0xff
#define MASK_VALUE 0xcc
+#define HALF_MASK MASK_VALUE
#define ZMASK_VALUE 0xc1
#define HF(x) x.xmmh[0]
#define SF(x) x.xmm[0]
+#define SI(x) x.xmmi[0]
+#define H_HF(x) x.xmmh[0]
#else
#define NET_MASK 0xffffffff
#define MASK_VALUE 0xcccccccc
#define ZMASK_VALUE 0xfcc1fcc1
+#define HALF_MASK 0xcccc
#define HF(x) x.zmmh
#define SF(x) x.zmm
+#define SI(x) x.zmmi
+#define H_HF(x) x.ymmh[0]
#endif
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\{rn-sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\{rz-sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m256h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epi32 (x1);
+ res1 = _mm512_mask_cvtph_epi32 (res, m16, x2);
+ res2 = _mm512_maskz_cvtph_epi32 (m16, x3);
+ res = _mm512_cvt_roundph_epi32 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epi32 (res, m16, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epi32 (m16, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,79 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_d) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ v5.u32[i] = 0;
+ }
+ else {
+ v5.u32[i] = dest->u32[i];
+ }
+ }
+ else {
+ v5.u32[i] = v1.f32[i];
+
+ }
+ }
+ *dest = v5;
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epi32) (H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epi32);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epi32) (SI(res), HALF_MASK, H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epi32);
+
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epi32) (HALF_MASK, H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epi32);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epi32) (H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epi32);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epi32) (SI(res), HALF_MASK, H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epi32);
+
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epi32) (HALF_MASK, H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epi32);
+#endif
+
+ if (n_errs != 0)
+ abort ();
+}
+
+
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\{rz-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m128h x1, x2, x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epi64 (x1);
+ res1 = _mm512_mask_cvtph_epi64 (res, m8, x2);
+ res2 = _mm512_maskz_cvtph_epi64 (m8, x3);
+ res = _mm512_cvt_roundph_epi64 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epi64 (res, m8, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epi64 (m8, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,78 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_q) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 8; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ v5.u64[i] = 0;
+ }
+ else {
+ v5.u64[i] = dest->u64[i];
+ }
+ }
+ else {
+ v5.u64[i] = v1.f32[i];
+ }
+ }
+ *dest = v5;
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epi64) (src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epi64);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epi64) (SI(res), 0xcc, src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epi64);
+
+ EMULATE(cvtph2_q)(&exp, src1, 0xfa, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epi64) (0xfa, src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epi64);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epi64) (src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epi64);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epi64) (SI(res), 0xcc, src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epi64);
+
+ EMULATE(cvtph2_q)(&exp, src1, 0xfa, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epi64) (0xfa, src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epi64);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\{rn-sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\{rz-sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m256h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epu32 (x1);
+ res1 = _mm512_mask_cvtph_epu32 (res, m16, x2);
+ res2 = _mm512_maskz_cvtph_epu32 (m16, x3);
+ res = _mm512_cvt_roundph_epu32 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epu32 (res, m16, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epu32 (m16, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,79 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_d) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ v5.u32[i] = 0;
+ }
+ else {
+ v5.u32[i] = dest->u32[i];
+ }
+ }
+ else {
+ v5.u32[i] = v1.f32[i];
+
+ }
+ }
+ *dest = v5;
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epu32) (H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epu32);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epu32) (SI(res), HALF_MASK, H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epu32);
+
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epu32) (HALF_MASK, H_HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epu32);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epu32) (H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epu32);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epu32) (SI(res), HALF_MASK, H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epu32);
+
+ EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epu32) (HALF_MASK, H_HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epu32);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\{rz-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m128h x1, x2, x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epu64 (x1);
+ res1 = _mm512_mask_cvtph_epu64 (res, m8, x2);
+ res2 = _mm512_maskz_cvtph_epu64 (m8, x3);
+ res = _mm512_cvt_roundph_epu64 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epu64 (res, m8, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epu64 (m8, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,78 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_q) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 8; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ v5.u64[i] = 0;
+ }
+ else {
+ v5.u64[i] = dest->u64[i];
+ }
+ }
+ else {
+ v5.u64[i] = v1.f32[i];
+ }
+ }
+ *dest = v5;
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epu64) (src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epu64);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epu64) (SI(res), 0xcc, src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epu64);
+
+ EMULATE(cvtph2_q)(&exp, src1, 0xfc, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epu64) (0xfc, src1.xmmh[0]);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epu64);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epu64) (src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epu64);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epu64) (SI(res), 0xcc, src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epu64);
+
+ EMULATE(cvtph2_q)(&exp, src1, 0xfc, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epu64) (0xfc, src1.xmmh[0], _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epu64);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask32 m32;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epu16 (x1);
+ res1 = _mm512_mask_cvtph_epu16 (res, m32, x2);
+ res2 = _mm512_maskz_cvtph_epu16 (m32, x3);
+ res = _mm512_cvt_roundph_epu16 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epu16 (res, m32, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epu16 (m32, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,84 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_w) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+ m2 = (k >> 16) & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ dest->u16[i] = 0;
+ }
+ }
+ else {
+ dest->u16[i] = v1.f32[i];
+
+ }
+
+ if (((1 << i) & m2) == 0) {
+ if (zero_mask) {
+ dest->u16[i+16] = 0;
+ }
+ }
+ else {
+ dest->u16[i+16] = v2.f32[i];
+ }
+ }
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epu16) (HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epu16);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epu16) (SI(res), MASK_VALUE, HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epu16);
+
+ EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epu16) (ZMASK_VALUE, HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epu16);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epu16) (HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epu16);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epu16) (SI(res), MASK_VALUE, HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epu16);
+
+ EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epu16) (ZMASK_VALUE, HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epu16);
+#endif
+
+ if (n_errs != 0)
+ abort ();
+}
+
new file mode 100644
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask32 m32;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_cvtph_epi16 (x1);
+ res1 = _mm512_mask_cvtph_epi16 (res, m32, x2);
+ res2 = _mm512_maskz_cvtph_epi16 (m32, x3);
+ res = _mm512_cvt_roundph_epi16 (x1, 4);
+ res1 = _mm512_mask_cvt_roundph_epi16 (res, m32, x2, 8);
+ res2 = _mm512_maskz_cvt_roundph_epi16 (m32, x3, 11);
+}
new file mode 100644
@@ -0,0 +1,83 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+#define N_ELEMS (AVX512F_LEN / 16)
+
+void NOINLINE
+EMULATE(cvtph2_w) (V512 * dest, V512 op1,
+ __mmask32 k, int zero_mask)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ __mmask16 m1, m2;
+
+ m1 = k & 0xffff;
+ m2 = (k >> 16) & 0xffff;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << i) & m1) == 0) {
+ if (zero_mask) {
+ dest->u16[i] = 0;
+ }
+ }
+ else {
+ dest->u16[i] = v1.f32[i];
+
+ }
+
+ if (((1 << i) & m2) == 0) {
+ if (zero_mask) {
+ dest->u16[i+16] = 0;
+ }
+ }
+ else {
+ dest->u16[i+16] = v2.f32[i];
+ }
+ }
+}
+
+void
+TEST (void)
+{
+ V512 res, exp;
+
+ init_src();
+
+ EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvtph_epi16) (HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvtph_epi16);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0);
+ SI(res) = INTRINSIC (_mask_cvtph_epi16) (SI(res), MASK_VALUE, HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtph_epi16);
+
+ EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1);
+ SI(res) = INTRINSIC (_maskz_cvtph_epi16) (ZMASK_VALUE, HF(src1));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtph_epi16);
+
+#if AVX512F_LEN == 512
+ EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0);
+ SI(res) = INTRINSIC (_cvt_roundph_epi16) (HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _cvt_roundph_epi16);
+
+ init_dest(&res, &exp);
+ EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0);
+ SI(res) = INTRINSIC (_mask_cvt_roundph_epi16) (SI(res), MASK_VALUE, HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvt_roundph_epi16);
+
+ EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1);
+ SI(res) = INTRINSIC (_maskz_cvt_roundph_epi16) (ZMASK_VALUE, HF(src1), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvt_roundph_epi16);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
new file mode 100644
@@ -0,0 +1,27 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m128h x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epi32 (x3);
+ res1 = _mm256_mask_cvtph_epi32 (res1, m8, x3);
+ res1 = _mm256_maskz_cvtph_epi32 (m8, x3);
+
+ res2 = _mm_cvtph_epi32 (x3);
+ res2 = _mm_mask_cvtph_epi32 (res2, m8, x3);
+ res2 = _mm_maskz_cvtph_epi32 (m8, x3);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2dq-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2dq-1b.c"
+
new file mode 100644
@@ -0,0 +1,27 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m128h x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epi64 (x3);
+ res1 = _mm256_mask_cvtph_epi64 (res1, m8, x3);
+ res1 = _mm256_maskz_cvtph_epi64 (m8, x3);
+
+ res2 = _mm_cvtph_epi64 (x3);
+ res2 = _mm_mask_cvtph_epi64 (res2, m8, x3);
+ res2 = _mm_maskz_cvtph_epi64 (m8, x3);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2qq-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2qq-1b.c"
+
new file mode 100644
@@ -0,0 +1,27 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m128h x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epu32 (x3);
+ res1 = _mm256_mask_cvtph_epu32 (res1, m8, x3);
+ res1 = _mm256_maskz_cvtph_epu32 (m8, x3);
+
+ res2 = _mm_cvtph_epu32 (x3);
+ res2 = _mm_mask_cvtph_epu32 (res2, m8, x3);
+ res2 = _mm_maskz_cvtph_epu32 (m8, x3);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2udq-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2udq-1b.c"
+
new file mode 100644
@@ -0,0 +1,27 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m128h x3;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epu64 (x3);
+ res1 = _mm256_mask_cvtph_epu64 (res1, m8, x3);
+ res1 = _mm256_maskz_cvtph_epu64 (m8, x3);
+
+ res2 = _mm_cvtph_epu64 (x3);
+ res2 = _mm_mask_cvtph_epu64 (res2, m8, x3);
+ res2 = _mm_maskz_cvtph_epu64 (m8, x3);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2uqq-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2uqq-1b.c"
+
new file mode 100644
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2uw\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m256h x3;
+volatile __m128h x4;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epu16 (x3);
+ res1 = _mm256_mask_cvtph_epu16 (res1, m16, x3);
+ res1 = _mm256_maskz_cvtph_epu16 (m16, x3);
+
+ res2 = _mm_cvtph_epu16 (x4);
+ res2 = _mm_mask_cvtph_epu16 (res2, m8, x4);
+ res2 = _mm_maskz_cvtph_epu16 (m8, x4);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2uw-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2uw-1b.c"
+
new file mode 100644
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2w\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __m256h x3;
+volatile __m128h x4;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_cvtph_epi16 (x3);
+ res1 = _mm256_mask_cvtph_epi16 (res1, m16, x3);
+ res1 = _mm256_maskz_cvtph_epi16 (m16, x3);
+
+ res2 = _mm_cvtph_epi16 (x4);
+ res2 = _mm_mask_cvtph_epi16 (res2, m8, x4);
+ res2 = _mm_maskz_cvtph_epi16 (m8, x4);
+}
new file mode 100644
@@ -0,0 +1,15 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2w-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vcvtph2w-1b.c"
+