gcc/testsuite/ChangeLog: * gcc.target/i386/avx512fp16-vcvttph2dq-1a.c: New test. * gcc.target/i386/avx512fp16-vcvttph2dq-1b.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2qq-1a.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2qq-1b.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2udq-1a.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2udq-1b.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2uqq-1a.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2uqq-1b.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2uw-1a.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2uw-1b.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2w-1a.c: Ditto. * gcc.target/i386/avx512fp16-vcvttph2w-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2dq-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2dq-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2qq-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2qq-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2udq-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2udq-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2uqq-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2uqq-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2uw-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2uw-1b.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2w-1a.c: Ditto. * gcc.target/i386/avx512fp16vl-vcvttph2w-1b.c: Ditto. --- .../i386/avx512fp16-vcvttph2dq-1a.c | 24 ++++++ .../i386/avx512fp16-vcvttph2dq-1b.c | 79 +++++++++++++++++ .../i386/avx512fp16-vcvttph2qq-1a.c | 24 ++++++ .../i386/avx512fp16-vcvttph2qq-1b.c | 78 +++++++++++++++++ .../i386/avx512fp16-vcvttph2udq-1a.c | 24 ++++++ .../i386/avx512fp16-vcvttph2udq-1b.c | 79 +++++++++++++++++ .../i386/avx512fp16-vcvttph2uqq-1a.c | 24 ++++++ .../i386/avx512fp16-vcvttph2uqq-1b.c | 78 +++++++++++++++++ .../i386/avx512fp16-vcvttph2uw-1a.c | 24 ++++++ .../i386/avx512fp16-vcvttph2uw-1b.c | 84 +++++++++++++++++++ .../gcc.target/i386/avx512fp16-vcvttph2w-1a.c | 24 ++++++ .../gcc.target/i386/avx512fp16-vcvttph2w-1b.c | 83 ++++++++++++++++++ .../i386/avx512fp16vl-vcvttph2dq-1a.c | 27 ++++++ .../i386/avx512fp16vl-vcvttph2dq-1b.c | 15 ++++ .../i386/avx512fp16vl-vcvttph2qq-1a.c | 27 ++++++ .../i386/avx512fp16vl-vcvttph2qq-1b.c | 15 ++++ .../i386/avx512fp16vl-vcvttph2udq-1a.c | 27 ++++++ .../i386/avx512fp16vl-vcvttph2udq-1b.c | 15 ++++ .../i386/avx512fp16vl-vcvttph2uqq-1a.c | 27 ++++++ .../i386/avx512fp16vl-vcvttph2uqq-1b.c | 15 ++++ .../i386/avx512fp16vl-vcvttph2uw-1a.c | 29 +++++++ .../i386/avx512fp16vl-vcvttph2uw-1b.c | 15 ++++ .../i386/avx512fp16vl-vcvttph2w-1a.c | 29 +++++++ .../i386/avx512fp16vl-vcvttph2w-1b.c | 15 ++++ 24 files changed, 881 insertions(+) create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1b.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1a.c create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1b.c
diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1a.c new file mode 100644 index 00000000000..0e44aaf1bb5 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\{sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\{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_cvttph_epi32 (x1); + res1 = _mm512_mask_cvttph_epi32 (res, m16, x2); + res2 = _mm512_maskz_cvttph_epi32 (m16, x3); + res = _mm512_cvtt_roundph_epi32 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epi32 (res, m16, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epi32 (m16, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1b.c new file mode 100644 index 00000000000..c18fefbf206 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2dq-1b.c @@ -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 (_cvttph_epi32) (H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epi32); + + init_dest(&res, &exp); + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0); + SI(res) = INTRINSIC (_mask_cvttph_epi32) (SI(res), HALF_MASK, H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epi32); + + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epi32) (HALF_MASK, H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epi32); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epi32) (H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epi32); + + init_dest(&res, &exp); + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epi32) (SI(res), HALF_MASK, H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epi32); + + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epi32) (HALF_MASK, H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epi32); +#endif + + if (n_errs != 0) + abort (); +} + + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1a.c new file mode 100644 index 00000000000..124169467ee --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\{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_cvttph_epi64 (x1); + res1 = _mm512_mask_cvttph_epi64 (res, m8, x2); + res2 = _mm512_maskz_cvttph_epi64 (m8, x3); + res = _mm512_cvtt_roundph_epi64 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epi64 (res, m8, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epi64 (m8, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1b.c new file mode 100644 index 00000000000..2a9a2ca26f9 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2qq-1b.c @@ -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 (_cvttph_epi64) (src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epi64); + + init_dest(&res, &exp); + EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0); + SI(res) = INTRINSIC (_mask_cvttph_epi64) (SI(res), 0xcc, src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epi64); + + EMULATE(cvtph2_q)(&exp, src1, 0xfa, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epi64) (0xfa, src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epi64); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epi64) (src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epi64); + + init_dest(&res, &exp); + EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epi64) (SI(res), 0xcc, src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epi64); + + EMULATE(cvtph2_q)(&exp, src1, 0xfa, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epi64) (0xfa, src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epi64); +#endif + + if (n_errs != 0) { + abort (); + } +} + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1a.c new file mode 100644 index 00000000000..0fd60f56777 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\{sae\}\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\{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_cvttph_epu32 (x1); + res1 = _mm512_mask_cvttph_epu32 (res, m16, x2); + res2 = _mm512_maskz_cvttph_epu32 (m16, x3); + res = _mm512_cvtt_roundph_epu32 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epu32 (res, m16, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epu32 (m16, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1b.c new file mode 100644 index 00000000000..98bce374753 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2udq-1b.c @@ -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 (_cvttph_epu32) (H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epu32); + + init_dest(&res, &exp); + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0); + SI(res) = INTRINSIC (_mask_cvttph_epu32) (SI(res), HALF_MASK, H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epu32); + + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epu32) (HALF_MASK, H_HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epu32); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_d)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epu32) (H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epu32); + + init_dest(&res, &exp); + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epu32) (SI(res), HALF_MASK, H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epu32); + + EMULATE(cvtph2_d)(&exp, src1, HALF_MASK, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epu32) (HALF_MASK, H_HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epu32); +#endif + + if (n_errs != 0) { + abort (); + } +} + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1a.c new file mode 100644 index 00000000000..04fee2936c8 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\{sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\{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_cvttph_epu64 (x1); + res1 = _mm512_mask_cvttph_epu64 (res, m8, x2); + res2 = _mm512_maskz_cvttph_epu64 (m8, x3); + res = _mm512_cvtt_roundph_epu64 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epu64 (res, m8, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epu64 (m8, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1b.c new file mode 100644 index 00000000000..31879ef8983 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uqq-1b.c @@ -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 (_cvttph_epu64) (src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epu64); + + init_dest(&res, &exp); + EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0); + SI(res) = INTRINSIC (_mask_cvttph_epu64) (SI(res), 0xcc, src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epu64); + + EMULATE(cvtph2_q)(&exp, src1, 0xfc, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epu64) (0xfc, src1.xmmh[0]); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epu64); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_q)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epu64) (src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epu64); + + init_dest(&res, &exp); + EMULATE(cvtph2_q)(&exp, src1, 0xcc, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epu64) (SI(res), 0xcc, src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epu64); + + EMULATE(cvtph2_q)(&exp, src1, 0xfc, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epu64) (0xfc, src1.xmmh[0], _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epu64); +#endif + + if (n_errs != 0) { + abort (); + } +} + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1a.c new file mode 100644 index 00000000000..b31af8441a9 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\{sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\{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_cvttph_epu16 (x1); + res1 = _mm512_mask_cvttph_epu16 (res, m32, x2); + res2 = _mm512_maskz_cvttph_epu16 (m32, x3); + res = _mm512_cvtt_roundph_epu16 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epu16 (res, m32, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epu16 (m32, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1b.c new file mode 100644 index 00000000000..34e94e8e549 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2uw-1b.c @@ -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 (_cvttph_epu16) (HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epu16); + + init_dest(&res, &exp); + EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0); + SI(res) = INTRINSIC (_mask_cvttph_epu16) (SI(res), MASK_VALUE, HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epu16); + + EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epu16) (ZMASK_VALUE, HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epu16); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epu16) (HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epu16); + + init_dest(&res, &exp); + EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epu16) (SI(res), MASK_VALUE, HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epu16); + + EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epu16) (ZMASK_VALUE, HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epu16); +#endif + + if (n_errs != 0) + abort (); +} + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1a.c new file mode 100644 index 00000000000..a918594d0d8 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1a.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\{sae\}\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\{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_cvttph_epi16 (x1); + res1 = _mm512_mask_cvttph_epi16 (res, m32, x2); + res2 = _mm512_maskz_cvttph_epi16 (m32, x3); + res = _mm512_cvtt_roundph_epi16 (x1, 4); + res1 = _mm512_mask_cvtt_roundph_epi16 (res, m32, x2, 8); + res2 = _mm512_maskz_cvtt_roundph_epi16 (m32, x3, 8); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1b.c new file mode 100644 index 00000000000..23bc8e680c5 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-vcvttph2w-1b.c @@ -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 (_cvttph_epi16) (HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvttph_epi16); + + init_dest(&res, &exp); + EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0); + SI(res) = INTRINSIC (_mask_cvttph_epi16) (SI(res), MASK_VALUE, HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvttph_epi16); + + EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1); + SI(res) = INTRINSIC (_maskz_cvttph_epi16) (ZMASK_VALUE, HF(src1)); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvttph_epi16); + +#if AVX512F_LEN == 512 + EMULATE(cvtph2_w)(&exp, src1, NET_MASK, 0); + SI(res) = INTRINSIC (_cvtt_roundph_epi16) (HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _cvtt_roundph_epi16); + + init_dest(&res, &exp); + EMULATE(cvtph2_w)(&exp, src1, MASK_VALUE, 0); + SI(res) = INTRINSIC (_mask_cvtt_roundph_epi16) (SI(res), MASK_VALUE, HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _mask_cvtt_roundph_epi16); + + EMULATE(cvtph2_w)(&exp, src1, ZMASK_VALUE, 1); + SI(res) = INTRINSIC (_maskz_cvtt_roundph_epi16) (ZMASK_VALUE, HF(src1), _ROUND_NINT); + CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_cvtt_roundph_epi16); +#endif + + if (n_errs != 0) { + abort (); + } +} + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1a.c new file mode 100644 index 00000000000..b4c084020ac --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1a.c @@ -0,0 +1,27 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2dq\[ \\t\]+%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_cvttph_epi32 (x3); + res1 = _mm256_mask_cvttph_epi32 (res1, m8, x3); + res1 = _mm256_maskz_cvttph_epi32 (m8, x3); + + res2 = _mm_cvttph_epi32 (x3); + res2 = _mm_mask_cvttph_epi32 (res2, m8, x3); + res2 = _mm_maskz_cvttph_epi32 (m8, x3); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1b.c new file mode 100644 index 00000000000..f9d82f92f4d --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2dq-1b.c @@ -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-vcvttph2dq-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2dq-1b.c" + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1a.c new file mode 100644 index 00000000000..421c688ee29 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1a.c @@ -0,0 +1,27 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2qq\[ \\t\]+%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_cvttph_epi64 (x3); + res1 = _mm256_mask_cvttph_epi64 (res1, m8, x3); + res1 = _mm256_maskz_cvttph_epi64 (m8, x3); + + res2 = _mm_cvttph_epi64 (x3); + res2 = _mm_mask_cvttph_epi64 (res2, m8, x3); + res2 = _mm_maskz_cvttph_epi64 (m8, x3); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1b.c new file mode 100644 index 00000000000..323ab74fa05 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2qq-1b.c @@ -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-vcvttph2qq-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2qq-1b.c" + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1a.c new file mode 100644 index 00000000000..60f43189d61 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1a.c @@ -0,0 +1,27 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2udq\[ \\t\]+%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_cvttph_epu32 (x3); + res1 = _mm256_mask_cvttph_epu32 (res1, m8, x3); + res1 = _mm256_maskz_cvttph_epu32 (m8, x3); + + res2 = _mm_cvttph_epu32 (x3); + res2 = _mm_mask_cvttph_epu32 (res2, m8, x3); + res2 = _mm_maskz_cvttph_epu32 (m8, x3); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1b.c new file mode 100644 index 00000000000..61365d456c2 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2udq-1b.c @@ -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-vcvttph2udq-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2udq-1b.c" + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1a.c new file mode 100644 index 00000000000..37008f9d9e9 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1a.c @@ -0,0 +1,27 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uqq\[ \\t\]+%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_cvttph_epu64 (x3); + res1 = _mm256_mask_cvttph_epu64 (res1, m8, x3); + res1 = _mm256_maskz_cvttph_epu64 (m8, x3); + + res2 = _mm_cvttph_epu64 (x3); + res2 = _mm_mask_cvttph_epu64 (res2, m8, x3); + res2 = _mm_maskz_cvttph_epu64 (m8, x3); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1b.c new file mode 100644 index 00000000000..6360402e6d6 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uqq-1b.c @@ -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-vcvttph2uqq-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2uqq-1b.c" + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1a.c new file mode 100644 index 00000000000..eafa31a786b --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1a.c @@ -0,0 +1,29 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2uw\[ \\t\]+%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_cvttph_epu16 (x3); + res1 = _mm256_mask_cvttph_epu16 (res1, m16, x3); + res1 = _mm256_maskz_cvttph_epu16 (m16, x3); + + res2 = _mm_cvttph_epu16 (x4); + res2 = _mm_mask_cvttph_epu16 (res2, m8, x4); + res2 = _mm_maskz_cvttph_epu16 (m8, x4); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1b.c new file mode 100644 index 00000000000..dd5ed9d5b38 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2uw-1b.c @@ -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-vcvttph2uw-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2uw-1b.c" + diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1a.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1a.c new file mode 100644 index 00000000000..7476d3c1160 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1a.c @@ -0,0 +1,29 @@ +/* { dg-do compile } */ +/* { dg-options "-mavx512fp16 -mavx512vl -O2" } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */ +/* { dg-final { scan-assembler-times "vcvttph2w\[ \\t\]+%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_cvttph_epi16 (x3); + res1 = _mm256_mask_cvttph_epi16 (res1, m16, x3); + res1 = _mm256_maskz_cvttph_epi16 (m16, x3); + + res2 = _mm_cvttph_epi16 (x4); + res2 = _mm_mask_cvttph_epi16 (res2, m8, x4); + res2 = _mm_maskz_cvttph_epi16 (m8, x4); +} diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1b.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1b.c new file mode 100644 index 00000000000..7a04a6a8ebc --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-vcvttph2w-1b.c @@ -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-vcvttph2w-1b.c" + +#undef AVX512F_LEN +#undef AVX512F_LEN_HALF + +#define AVX512F_LEN 128 +#define AVX512F_LEN_HALF 128 +#include "avx512fp16-vcvttph2w-1b.c" + -- 2.18.1