for (i = 0; i < AVX512F_MAX_ELEM; i++) {
v1.f32[i] = i + 1;
- v2.f32[i] = i * 0.5f;
+ v2.f32[i] = (i + 2) * 0.5f;
v3.f32[i] = i * 1.5f;
- v4.f32[i] = i - 0.5f;
+ v4.f32[i] = i - 1.5f;
src3.u32[i] = (i + 1) * 10;
}
#undef DF
#undef H_HF
#undef NET_MASK
+#undef NET_CMASK
#undef MASK_VALUE
#undef HALF_MASK
#undef ZMASK_VALUE
#define NET_MASK 0xffff
+#define NET_CMASK 0xff
#define MASK_VALUE 0xcccc
#define ZMASK_VALUE 0xfcc1
#define HALF_MASK 0xcc
#undef SI
#undef H_HF
#undef NET_MASK
+#undef NET_CMASK
#undef MASK_VALUE
#undef ZMASK_VALUE
#undef HALF_MASK
#define NET_MASK 0xff
+#define NET_CMASK 0xff
#define MASK_VALUE 0xcc
#define HALF_MASK MASK_VALUE
#define ZMASK_VALUE 0xc1
#define H_HF(x) x.xmmh[0]
#else
#define NET_MASK 0xffffffff
+#define NET_CMASK 0xffff
#define MASK_VALUE 0xcccccccc
#define ZMASK_VALUE 0xfcc1fcc1
#define HALF_MASK 0xcccc
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512h res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_fcmadd_pch (x1, x2, x3);
+ res1 = _mm512_mask_fcmadd_pch (res1, m16, x1, x2);
+ res1 = _mm512_mask3_fcmadd_pch (res1, x1, x2, m16);
+ res2 = _mm512_maskz_fcmadd_pch (m16, x1, x2, x3);
+ res = _mm512_fcmadd_round_pch (x1, x2, x3, 8);
+ res1 = _mm512_mask_fcmadd_round_pch (res1, m16, x1, x2, 8);
+ res1 = _mm512_mask3_fcmadd_round_pch (res1, x1, x2, m16, 8);
+ res2 = _mm512_maskz_fcmadd_round_pch (m16, x1, x2, x3, 11);
+}
--- /dev/null
+/* { 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(c_fmadd_pch) (V512 * dest, V512 op1, V512 op2,
+ __mmask16 k, int zero_mask, int c_flag,
+ int is_mask3)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ int invert = 1;
+ if (c_flag == 1)
+ invert = -1;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v7, &v8);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << (i / 2)) & k) == 0) {
+ if (zero_mask) {
+ v5.f32[i] = 0;
+ }
+ else {
+ v5.u32[i] = is_mask3 ? v3.u32[i] : v7.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v5.f32[i] = v1.f32[i] * v7.f32[i]
+ - invert * (v1.f32[i+1] * v7.f32[i+1]) + v3.f32[i];
+ }
+ else {
+ v5.f32[i] = v1.f32[i-1] * v7.f32[i]
+ + invert * (v1.f32[i] * v7.f32[i-1]) + v3.f32[i];
+
+ }
+ }
+ if (((1 << (i / 2 + 8)) & k) == 0) {
+ if (zero_mask) {
+ v6.f32[i] = 0;
+ }
+ else {
+ v6.u32[i] = is_mask3 ? v4.u32[i] : v8.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v6.f32[i] = v2.f32[i] * v8.f32[i]
+ - invert * (v2.f32[i+1] * v8.f32[i+1]) + v4.f32[i];
+ }
+ else {
+ v6.f32[i] = v2.f32[i-1] * v8.f32[i]
+ + invert * (v2.f32[i] * v8.f32[i-1]) + v4.f32[i];
+ }
+
+ }
+ }
+
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, NET_CMASK, 0, 1, 0);
+ HF(res) = INTRINSIC (_fcmadd_pch) (HF(res), HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fcmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 1, 0);
+ HF(res) = INTRINSIC (_mask_fcmadd_pch) (HF(res) ,HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 1, 1);
+ HF(res) = INTRINSIC (_mask3_fcmadd_pch) (HF(res), HF(src1),
+ HF(src2), HALF_MASK);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fcmadd_pch);
+
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 1, 1, 0);
+ HF(res) = INTRINSIC (_maskz_fcmadd_pch) (HALF_MASK, HF(res),
+ HF(src1), HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmadd_pch);
+
+#if AVX512F_LEN == 512
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, NET_CMASK, 0, 1, 0);
+ HF(res) = INTRINSIC (_fcmadd_round_pch) (HF(res), HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fcmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 1, 0);
+ HF(res) = INTRINSIC (_mask_fcmadd_round_pch) (HF(res) ,HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 1, 1);
+ HF(res) = INTRINSIC (_mask3_fcmadd_round_pch) (HF(res), HF(src1),
+ HF(src2), HALF_MASK, _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fcmadd_pch);
+
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 1, 1, 0);
+ HF(res) = INTRINSIC (_maskz_fcmadd_round_pch) (HALF_MASK, HF(res),
+ HF(src1), HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmadd_pch);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512h res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_fcmul_pch (x1, x2);
+ res1 = _mm512_mask_fcmul_pch (res1, m16, x1, x2);
+ res2 = _mm512_maskz_fcmul_pch (m16, x1, x2);
+ res = _mm512_fcmul_round_pch (x1, x2, 8);
+ res1 = _mm512_mask_fcmul_round_pch (res1, m16, x1, x2, 8);
+ res2 = _mm512_maskz_fcmul_round_pch (m16, x1, x2, 11);
+}
--- /dev/null
+/* { 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(c_fmul_pch) (V512 * dest, V512 op1, V512 op2,
+ __mmask16 k, int zero_mask, int c_flag)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ int invert = 1;
+ if (c_flag == 1)
+ invert = -1;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v7, &v8);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << (i / 2)) & k) == 0) {
+ if (zero_mask) {
+ v5.f32[i] = 0;
+ }
+ else {
+ v5.u32[i] = v7.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v5.f32[i] = v1.f32[i] * v3.f32[i]
+ - invert * (v1.f32[i+1] * v3.f32[i+1]);
+ }
+ else {
+ v5.f32[i] = v1.f32[i] * v3.f32[i-1]
+ + invert * (v1.f32[i-1] * v3.f32[i]);
+
+ }
+ }
+ if (((1 << (i / 2 + 8)) & k) == 0) {
+ if (zero_mask) {
+ v6.f32[i] = 0;
+ }
+ else {
+ v6.u32[i] = v8.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v6.f32[i] = v2.f32[i] * v4.f32[i]
+ - invert * (v2.f32[i+1] * v4.f32[i+1]);
+ }
+ else {
+ v6.f32[i] = v2.f32[i] * v4.f32[i-1]
+ + invert * (v2.f32[i-1] * v4.f32[i]);
+ }
+
+ }
+ }
+
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 1);
+ HF(res) = INTRINSIC (_fcmul_pch) (HF(src1), HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fcmul_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 1);
+ HF(res) = INTRINSIC (_mask_fcmul_pch) (HF(res) ,HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmul_pch);
+
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 1);
+ HF(res) = INTRINSIC (_maskz_fcmul_pch) ( HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmul_pch);
+
+#if AVX512F_LEN == 512
+ EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 1);
+ HF(res) = INTRINSIC (_fcmul_round_pch) (HF(src1), HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fcmul_round_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 1);
+ HF(res) = INTRINSIC (_mask_fcmul_round_pch) (HF(res) ,HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmul_round_pch);
+
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 1);
+ HF(res) = INTRINSIC (_maskz_fcmul_round_pch) ( HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmul_round_pch);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512h res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_fmadd_pch (x1, x2, x3);
+ res1 = _mm512_mask_fmadd_pch (res1, m16, x1, x2);
+ res1 = _mm512_mask3_fmadd_pch (res1, x1, x2, m16);
+ res2 = _mm512_maskz_fmadd_pch (m16, x1, x2, x3);
+ res = _mm512_fmadd_round_pch (x1, x2, x3, 8);
+ res1 = _mm512_mask_fmadd_round_pch (res1, m16, x1, x2, 8);
+ res1 = _mm512_mask3_fmadd_round_pch (res1, x1, x2, m16, 8);
+ res2 = _mm512_maskz_fmadd_round_pch (m16, x1, x2, x3, 11);
+}
--- /dev/null
+/* { 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(c_fmadd_pch) (V512 * dest, V512 op1, V512 op2,
+ __mmask16 k, int zero_mask, int c_flag,
+ int is_mask3)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ int invert = 1;
+ if (c_flag == 1)
+ invert = -1;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v7, &v8);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << (i / 2)) & k) == 0) {
+ if (zero_mask) {
+ v5.f32[i] = 0;
+ }
+ else {
+ v5.u32[i] = is_mask3 ? v3.u32[i] : v7.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v5.f32[i] = v1.f32[i] * v7.f32[i]
+ - invert * (v1.f32[i+1] * v7.f32[i+1]) + v3.f32[i];
+ }
+ else {
+ v5.f32[i] = v1.f32[i-1] * v7.f32[i]
+ + invert * (v1.f32[i] * v7.f32[i-1]) + v3.f32[i];
+
+ }
+ }
+ if (((1 << (i / 2 + 8)) & k) == 0) {
+ if (zero_mask) {
+ v6.f32[i] = 0;
+ }
+ else {
+ v6.u32[i] = is_mask3 ? v4.u32[i] : v8.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v6.f32[i] = v2.f32[i] * v8.f32[i]
+ - invert * (v2.f32[i+1] * v8.f32[i+1]) + v4.f32[i];
+ }
+ else {
+ v6.f32[i] = v2.f32[i-1] * v8.f32[i]
+ + invert * (v2.f32[i] * v8.f32[i-1]) + v4.f32[i];
+ }
+
+ }
+ }
+
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, NET_CMASK, 0, 0, 0);
+ HF(res) = INTRINSIC (_fmadd_pch) (HF(res), HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 0, 0);
+ HF(res) = INTRINSIC (_mask_fmadd_pch) (HF(res), HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 0, 1);
+ HF(res) = INTRINSIC (_mask3_fmadd_pch) (HF(res), HF(src1), HF(src2),
+ HALF_MASK);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 1, 0, 0);
+ HF(res) = INTRINSIC (_maskz_fmadd_pch) (HALF_MASK, HF(res), HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmadd_pch);
+
+#if AVX512F_LEN == 512
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, NET_CMASK, 0, 0, 0);
+ HF(res) = INTRINSIC (_fmadd_round_pch) (HF(res), HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 0, 0);
+ HF(res) = INTRINSIC (_mask_fmadd_round_pch) (HF(res), HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 0, 0, 1);
+ HF(res) = INTRINSIC (_mask3_fmadd_round_pch) (HF(res), HF(src1), HF(src2),
+ HALF_MASK, _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmadd_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmadd_pch)(&exp, src1, src2, HALF_MASK, 1, 0, 0);
+ HF(res) = INTRINSIC (_maskz_fmadd_round_pch) (HALF_MASK, HF(res), HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmadd_pch);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2" } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\{rn-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\{rz-sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512h res, res1, res2;
+volatile __m512h x1, x2, x3;
+volatile __mmask16 m16;
+
+void extern
+avx512f_test (void)
+{
+ res = _mm512_fmul_pch (x1, x2);
+ res1 = _mm512_mask_fmul_pch (res1, m16, x1, x2);
+ res2 = _mm512_maskz_fmul_pch (m16, x1, x2);
+ res = _mm512_fmul_round_pch (x1, x2, 8);
+ res1 = _mm512_mask_fmul_round_pch (res1, m16, x1, x2, 8);
+ res2 = _mm512_maskz_fmul_round_pch (m16, x1, x2, 11);
+}
--- /dev/null
+/* { 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(c_fmul_pch) (V512 * dest, V512 op1, V512 op2,
+ __mmask16 k, int zero_mask, int c_flag)
+{
+ V512 v1, v2, v3, v4, v5, v6, v7, v8;
+ int i;
+ int invert = 1;
+ if (c_flag == 1)
+ invert = -1;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v7, &v8);
+
+ for (i = 0; i < 16; i++) {
+ if (((1 << (i / 2)) & k) == 0) {
+ if (zero_mask) {
+ v5.f32[i] = 0;
+ }
+ else {
+ v5.u32[i] = v7.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v5.f32[i] = v1.f32[i] * v3.f32[i]
+ - invert * (v1.f32[i+1] * v3.f32[i+1]);
+ }
+ else {
+ v5.f32[i] = v1.f32[i-1] * v3.f32[i]
+ + invert * (v1.f32[i] * v3.f32[i-1]);
+
+ }
+ }
+ if (((1 << (i / 2 + 8)) & k) == 0) {
+ if (zero_mask) {
+ v6.f32[i] = 0;
+ }
+ else {
+ v6.u32[i] = v8.u32[i];
+ }
+ }
+ else {
+ if ((i % 2) == 0) {
+ v6.f32[i] = v2.f32[i] * v4.f32[i]
+ - invert * (v2.f32[i+1] * v4.f32[i+1]);
+ }
+ else {
+ v6.f32[i] = v2.f32[i-1] * v4.f32[i]
+ + invert * (v2.f32[i] * v4.f32[i-1]);
+ }
+
+ }
+ }
+
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+void
+TEST (void)
+{
+ V512 res;
+ V512 exp;
+
+ init_src();
+
+ EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 0);
+ HF(res) = INTRINSIC (_fmul_pch) (HF(src1), HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fmul_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 0);
+ HF(res) = INTRINSIC (_mask_fmul_pch) (HF(res),HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmul_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 0);
+ HF(res) = INTRINSIC (_maskz_fmul_pch) (HALF_MASK, HF(src1),
+ HF(src2));
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmul_pch);
+
+#if AVX512F_LEN == 512
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 0);
+ HF(res) = INTRINSIC (_fmul_round_pch) (HF(src1), HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _fmul_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 0);
+ HF(res) = INTRINSIC (_mask_fmul_round_pch) (HF(res),HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmul_pch);
+
+ init_dest(&res, &exp);
+ EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 0);
+ HF(res) = INTRINSIC (_maskz_fmul_round_pch) (HALF_MASK, HF(src1),
+ HF(src2), _ROUND_NINT);
+ CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmul_pch);
+#endif
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]+%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 __m256h res1;
+volatile __m128h res2;
+volatile __m256h x1, x2, x3;
+volatile __m128h x4, x5, x6;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_fcmadd_pch (x1, x2, x3);
+ res1 = _mm256_mask_fcmadd_pch (res1, m8, x1, x2);
+ res1 = _mm256_mask3_fcmadd_pch (res1, x1, x2, m8);
+ res1 = _mm256_maskz_fcmadd_pch (m8, x1, x2, x3);
+
+ res2 = _mm_fcmadd_pch (x4, x5, x6);
+ res2 = _mm_mask_fcmadd_pch (res2, m8, x4, x5);
+ res2 = _mm_mask3_fcmadd_pch (res2, x4, x5, m8);
+ res2 = _mm_maskz_fcmadd_pch (m8, x4, x5, x6);
+}
--- /dev/null
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfcmaddcph-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfcmaddcph-1b.c"
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]+%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 __m256h res1;
+volatile __m128h res2;
+volatile __m256h x1, x2, x3;
+volatile __m128h x4, x5, x6;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_fcmul_pch (x1, x2);
+ res1 = _mm256_mask_fcmul_pch (res1, m8, x1, x2);
+ res1 = _mm256_maskz_fcmul_pch (m8, x1, x2);
+
+ res2 = _mm_fcmul_pch (x4, x5);
+ res2 = _mm_mask_fcmul_pch (res2, m8, x4, x5);
+ res2 = _mm_maskz_fcmul_pch (m8, x4, x5);
+}
--- /dev/null
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfcmulcph-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfcmulcph-1b.c"
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]+%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 __m256h res1;
+volatile __m128h res2;
+volatile __m256h x1, x2, x3;
+volatile __m128h x4, x5, x6;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_fmadd_pch (x1, x2, x3);
+ res1 = _mm256_mask_fmadd_pch (res1, m8, x1, x2);
+ res1 = _mm256_mask3_fmadd_pch (res1, x1, x2, m8);
+ res1 = _mm256_maskz_fmadd_pch (m8, x1, x2, x3);
+
+ res2 = _mm_fmadd_pch (x4, x5, x6);
+ res2 = _mm_mask_fmadd_pch (res2, m8, x4, x5);
+ res2 = _mm_mask3_fmadd_pch (res2, x4, x5, m8);
+ res2 = _mm_maskz_fmadd_pch (m8, x4, x5, x6);
+}
--- /dev/null
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfmaddcph-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfmaddcph-1b.c"
+
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -mavx512fp16 -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]+%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 __m256h res1;
+volatile __m128h res2;
+volatile __m256h x1, x2, x3;
+volatile __m128h x4, x5, x6;
+volatile __mmask8 m8;
+
+void extern
+avx512f_test (void)
+{
+ res1 = _mm256_fmul_pch (x1, x2);
+ res1 = _mm256_mask_fmul_pch (res1, m8, x1, x2);
+ res1 = _mm256_maskz_fmul_pch (m8, x1, x2);
+
+ res2 = _mm_fmul_pch (x4, x5);
+ res2 = _mm_mask_fmul_pch (res2, m8, x4, x5);
+ res2 = _mm_maskz_fmul_pch (m8, x4, x5);
+}
--- /dev/null
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfmulcph-1b.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512fp16-vfmulcph-1b.c"
+
projects / thirdparty / gcc.git / commitdiff
