new file mode 100644
@@ -0,0 +1,19 @@
+/* { dg-do compile} */
+/* { dg-options "-O2 -mavx512fp16" } */
+
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\r\]*%zmm0" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\r\]*%xmm0" 1 } } */
+
+#include<immintrin.h>
+
+_Float16
+neghf (_Float16 a)
+{
+ return -a;
+}
+
+__m512h
+neghf512 (__m512h a)
+{
+ return -a;
+}
new file mode 100644
@@ -0,0 +1,33 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+static void
+test_512 (void)
+{
+ V512 v1, v2, v3, v4, exp, res;
+ int i;
+ init_src();
+
+ unpack_ph_2twops(src1, &v1, &v2);
+ v1.f32[0] = -v1.f32[0];
+ exp = pack_twops_2ph(v1, v2);
+ res.zmmh = src1.zmmh;
+ res.f16[0] = -res.f16[0];
+ check_results(&res, &exp, 32, "neg");
+
+ unpack_ph_2twops(src1, &v1, &v2);
+ for (i=0; i<16; i++)
+ {
+ v1.f32[i] = -v1.f32[i];
+ v2.f32[i] = -v2.f32[i];
+ }
+ exp = pack_twops_2ph(v1, v2);
+ res.zmmh = -src1.zmmh;
+ check_results(&res, &exp, 32, "neg");
+ if (n_errs != 0) {
+ abort ();
+ }
+}
new file mode 100644
@@ -0,0 +1,31 @@
+/* { dg-do compile } */
+/* { dg-options "-Ofast -mavx512fp16" } */
+
+_Float16
+f1 (_Float16 x)
+{
+ return __builtin_fabsf16 (x);
+}
+
+_Float16
+f2 (_Float16 x, _Float16 y)
+{
+ return __builtin_copysignf16 (x, y);
+}
+
+_Float16
+f3 (_Float16 x)
+{
+ return -x;
+}
+
+_Float16
+f4 (_Float16 x, _Float16 y)
+{
+ return x * __builtin_copysignf16 (1, y);
+}
+
+
+/* { dg-final { scan-assembler-times "vandps\[^\n\r\]*xmm\[0-9\]" 4 } } */
+/* { dg-final { scan-assembler-times "vorps\[^\n\r\]*xmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[^\n\r\]*xmm\[0-9\]" 2 } } */
new file mode 100644
@@ -0,0 +1,82 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-Ofast -mavx512fp16 -mavx512dq" } */
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+void NOINLINE
+emulate_absneg_ph (V512 * dest, V512 op1, int abs)
+{
+ V512 v1, v2, v3, v4;
+ int i;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(*dest, &v3, &v4);
+
+ for (i = 0; i != 16; i++) {
+ if (abs) {
+ v3.f32[i] = __builtin_fabsf (v1.f32[i]);
+ v4.f32[i] = __builtin_fabsf (v2.f32[i]);
+ }
+ else {
+ v3.f32[i] = -v1.f32[i];
+ v4.f32[i] = -v2.f32[i];
+ }
+ }
+ *dest = pack_twops_2ph(v3, v4);
+}
+
+void NOINLINE
+emulate_copysign_ph (V512 * dest, V512 op1, V512 op2, int xorsign)
+{
+ V512 v1, v2, v3, v4, v5, v6;
+ int i;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v5, &v6);
+
+ for (i = 0; i != 16; i++) {
+ if (xorsign) {
+ v5.f32[i] = v1.f32[i] * __builtin_copysignf (1, v3.f32[i]);
+ v6.f32[i] = v2.f32[i] * __builtin_copysignf (1, v4.f32[i]);
+ }
+ else {
+ v5.f32[i] = __builtin_copysignf (v1.f32[i], v3.f32[i]);
+ v6.f32[i] = __builtin_copysignf (v2.f32[i], v4.f32[i]);
+ }
+ }
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+void
+test_512 (void)
+{
+ V512 res, exp;
+
+ init_src ();
+
+ /* Abs for float16. */
+ emulate_absneg_ph (&exp, src1, 1);
+ res.f16[0] = __builtin_fabsf16 (src1.f16[0]);
+ check_results (&res, &exp, 1, "abs_float16");
+
+ /* Neg for float16. */
+ emulate_absneg_ph (&exp, src1, 0);
+ res.f16[0] = -(src1.f16[0]);
+ check_results (&res, &exp, 1, "neg_float16");
+
+ /* Copysign for float16. */
+ emulate_copysign_ph (&exp, src1, src2, 0);
+ res.f16[0] = __builtin_copysignf16 (src1.f16[0], src2.f16[0]);
+ check_results (&res, &exp, 1, "copysign_float16");
+
+ /* Xorsign for float16. */
+ emulate_copysign_ph (&exp, src1, src2, 1);
+ res.f16[0] = src1.f16[0] * __builtin_copysignf16 (1, src2.f16[0]);
+ check_results (&res, &exp, 1, "xorsign_float16");
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
new file mode 100644
@@ -0,0 +1,124 @@
+/* { dg-do compile } */
+/* { dg-options "-Ofast -mavx512vl -mavx512fp16" } */
+
+#include<immintrin.h>
+__m128h
+f1 (__m128h x)
+{
+ int i = 0;
+ __m128h y;
+ for (; i != 8; i++)
+ y[i] = __builtin_fabsf16 (x[i]);
+ return y;
+}
+
+__m256h
+f2 (__m256h x)
+{
+ int i = 0;
+ __m256h y;
+ for (; i != 16; i++)
+ y[i] = __builtin_fabsf16 (x[i]);
+ return y;
+}
+
+__m512h
+f3 (__m512h x)
+{
+ int i = 0;
+ __m512h y;
+ for (; i != 32; i++)
+ y[i] = __builtin_fabsf16 (x[i]);
+ return y;
+}
+
+__m128h
+f4 (__m128h x)
+{
+ return -x;
+}
+
+__m256h
+f5 (__m256h x)
+{
+ return -x;
+}
+
+__m512h
+f6 (__m512h x)
+{
+ return -x;
+}
+
+__m128h
+f7 (__m128h x, __m128h y)
+{
+ int i = 0;
+ __m128h z;
+ for (; i != 8; i++)
+ z[i] = __builtin_copysignf16 (x[i], y[i]);
+ return z;
+}
+
+__m256h
+f8 (__m256h x, __m256h y)
+{
+ int i = 0;
+ __m256h z;
+ for (; i != 16; i++)
+ z[i] = __builtin_copysignf16 (x[i], y[i]);
+ return z;
+}
+
+__m512h
+f9 (__m512h x, __m512h y)
+{
+ int i = 0;
+ __m512h z;
+ for (; i != 32; i++)
+ z[i] = __builtin_copysignf16 (x[i], y[i]);
+ return z;
+}
+
+__m128h
+f10 (__m128h x, __m128h y)
+{
+ int i = 0;
+ __m128h z;
+ for (; i != 8; i++)
+ z[i] = x[i] * __builtin_copysignf16 (1, y[i]);
+ return z;
+}
+
+__m256h
+f11 (__m256h x, __m256h y)
+{
+ int i = 0;
+ __m256h z;
+ for (; i != 16; i++)
+ z[i] = x[i] * __builtin_copysignf16 (1, y[i]);
+ return z;
+}
+
+__m512h
+f12 (__m512h x, __m512h y)
+{
+ int i = 0;
+ __m512h z;
+ for (; i != 32; i++)
+ z[i] = x[i] * __builtin_copysignf16 (1, y[i]);
+ return z;
+}
+
+/* { dg-final { scan-assembler "vandps\[^\n\r\]*xmm0" } } */
+/* { dg-final { scan-assembler "vandps\[^\n\r\]*ymm0" } } */
+/* { dg-final { scan-assembler "vpandd\[^\n\r\]*zmm0" } } */
+/* { dg-final { scan-assembler-times "vxorps\[^\n\r\]*xmm0" 2 } } */
+/* { dg-final { scan-assembler-times "vxorps\[^\n\r\]*ymm0" 2 } } */
+/* { dg-final { scan-assembler-times "vpxord\[^\n\r\]*zmm0" 2 } } */
+/* { dg-final { scan-assembler-times "vorps\[^\n\r\]*xmm0" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[^\n\r\]*ymm0" 1 } } */
+/* { dg-final { scan-assembler-times "vpord\[^\n\r\]*zmm0" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[^\n\r\]*xmm0" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[^\n\r\]*ymm0" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[^\n\r\]*zmm0" 1 } } */
new file mode 100644
@@ -0,0 +1,119 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-Ofast -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+void NOINLINE
+emulate_absneg_ph (V512 * dest, V512 op1, int abs)
+{
+ V512 v1, v2, v3, v4;
+ int i;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(*dest, &v3, &v4);
+
+ for (i = 0; i != 16; i++) {
+ if (abs) {
+ v3.f32[i] = __builtin_fabsf (v1.f32[i]);
+ v4.f32[i] = __builtin_fabsf (v2.f32[i]);
+ }
+ else {
+ v3.f32[i] = -v1.f32[i];
+ v4.f32[i] = -v2.f32[i];
+ }
+ }
+ *dest = pack_twops_2ph(v3, v4);
+}
+
+void NOINLINE
+emulate_copysign_ph (V512 * dest, V512 op1, V512 op2, int xorsign)
+{
+ V512 v1, v2, v3, v4, v5, v6;
+ int i;
+
+ unpack_ph_2twops(op1, &v1, &v2);
+ unpack_ph_2twops(op2, &v3, &v4);
+ unpack_ph_2twops(*dest, &v5, &v6);
+
+ for (i = 0; i != 16; i++) {
+ if (xorsign) {
+ v5.f32[i] = v1.f32[i] * __builtin_copysignf (1, v3.f32[i]);
+ v6.f32[i] = v2.f32[i] * __builtin_copysignf (1, v4.f32[i]);
+ }
+ else {
+ v5.f32[i] = __builtin_copysignf (v1.f32[i], v3.f32[i]);
+ v6.f32[i] = __builtin_copysignf (v2.f32[i], v4.f32[i]);
+ }
+ }
+ *dest = pack_twops_2ph(v5, v6);
+}
+
+
+void
+test_512 (void)
+{
+ V512 res, exp;
+
+ init_src ();
+
+ /* Abs for vector float16. */
+ emulate_absneg_ph (&exp, src1, 1);
+ for (int i = 0; i != 8; i++)
+ res.f16[i] = __builtin_fabsf16 (src1.f16[i]);
+ check_results (&res, &exp, 8, "abs_m128h");
+
+ for (int i = 0; i != 16; i++)
+ res.f16[i] = __builtin_fabsf16 (src1.f16[i]);
+ check_results (&res, &exp, 16, "abs_m256h");
+
+ for (int i = 0; i != 32; i++)
+ res.f16[i] = __builtin_fabsf16 (src1.f16[i]);
+ check_results (&res, &exp, 32, "abs_m512h");
+
+ /* Neg for vector float16. */
+ emulate_absneg_ph (&exp, src1, 0);
+ for (int i = 0; i != 8; i++)
+ res.f16[i] = -(src1.f16[i]);
+ check_results (&res, &exp, 8, "neg_m128h");
+
+ for (int i = 0; i != 16; i++)
+ res.f16[i] = -(src1.f16[i]);
+ check_results (&res, &exp, 16, "neg_m256h");
+
+ for (int i = 0; i != 32; i++)
+ res.f16[i] = -(src1.f16[i]);
+ check_results (&res, &exp, 32, "neg_m512h");
+
+ /* Copysign for vector float16. */
+ emulate_copysign_ph (&exp, src1, src2, 0);
+ for (int i = 0; i != 8; i++)
+ res.f16[i] = __builtin_copysignf16 (src1.f16[i], src2.f16[i]);
+ check_results (&res, &exp, 8, "copysign_m128h");
+
+ for (int i = 0; i != 16; i++)
+ res.f16[i] = __builtin_copysignf16 (src1.f16[i], src2.f16[i]);
+ check_results (&res, &exp, 16, "copysign_m256h");
+
+ for (int i = 0; i != 32; i++)
+ res.f16[i] = __builtin_copysignf16 (src1.f16[i], src2.f16[i]);
+ check_results (&res, &exp, 32, "copysign_m512h");
+
+ /* Xorsign for vector float16. */
+ emulate_copysign_ph (&exp, src1, src2, 1);
+ for (int i = 0; i != 8; i++)
+ res.f16[i] = src1.f16[i] * __builtin_copysignf16 (1, src2.f16[i]);
+ check_results (&res, &exp, 8, "xorsign_m128h");
+
+ for (int i = 0; i != 16; i++)
+ res.f16[i] = src1.f16[i] * __builtin_copysignf16 (1, src2.f16[i]);
+ check_results (&res, &exp, 16, "xorsign_m256h");
+
+ for (int i = 0; i != 32; i++)
+ res.f16[i] = src1.f16[i] * __builtin_copysignf16 (1, src2.f16[i]);
+ check_results (&res, &exp, 32, "xorsign_m512h");
+
+ if (n_errs != 0) {
+ abort ();
+ }
+}
new file mode 100644
@@ -0,0 +1,18 @@
+/* { dg-do compile} */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl" } */
+
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\r\]*%xmm0" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\r\]*%ymm0" 1 } } */
+#include<immintrin.h>
+
+__m128h
+neghf128 (__m128h a)
+{
+ return -a;
+}
+
+__m256h
+neghf256 (__m256h a)
+{
+ return -a;
+}
new file mode 100644
@@ -0,0 +1,33 @@
+/* { dg-do run { target avx512fp16 } } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mavx512dq" } */
+
+#define AVX512FP16
+#include "avx512fp16-helper.h"
+
+static void
+test_512 (void)
+{
+ V512 v1, v2, v3, v4, exp, res;
+ int i;
+ init_src();
+
+ unpack_ph_2twops(src1, &v1, &v2);
+ v1.f32[0] = -v1.f32[0];
+ exp = pack_twops_2ph(v1, v2);
+ res.zmmh = src1.zmmh;
+ res.f16[0] = -res.f16[0];
+ check_results(&res, &exp, 32, "neg");
+
+ unpack_ph_2twops(src1, &v1, &v2);
+ for (i=0; i<16; i++)
+ {
+ v1.f32[i] = -v1.f32[i];
+ v2.f32[i] = -v2.f32[i];
+ }
+ exp = pack_twops_2ph(v1, v2);
+ res.zmmh = -src1.zmmh;
+ check_results(&res, &exp, 32, "neg");
+ if (n_errs != 0) {
+ abort ();
+ }
+}