PR target/61878
* config/i386/avx512fintrin.h (_mm512_mask_cmpge_epi32_mask): New.
(_mm512_mask_cmpge_epu32_mask): Ditto.
(_mm512_cmpge_epu32_mask): Ditto.
(_mm512_mask_cmpge_epi64_mask): Ditto.
(_mm512_cmpge_epi64_mask): Ditto.
(_mm512_mask_cmpge_epu64_mask): Ditto.
(_mm512_cmpge_epu64_mask): Ditto.
(_mm512_mask_cmple_epi32_mask): Ditto.
(_mm512_cmple_epi32_mask): Ditto.
(_mm512_mask_cmple_epu32_mask): Ditto.
(_mm512_cmple_epu32_mask): Ditto.
(_mm512_mask_cmple_epi64_mask): Ditto.
(_mm512_cmple_epi64_mask): Ditto.
(_mm512_mask_cmple_epu64_mask): Ditto.
(_mm512_cmple_epu64_mask): Ditto.
(_mm512_mask_cmplt_epi32_mask): Ditto.
(_mm512_cmplt_epi32_mask): Ditto.
(_mm512_mask_cmplt_epu32_mask): Ditto.
(_mm512_cmplt_epu32_mask): Ditto.
(_mm512_mask_cmplt_epi64_mask): Ditto.
(_mm512_cmplt_epi64_mask): Ditto.
(_mm512_mask_cmplt_epu64_mask): Ditto.
(_mm512_cmplt_epu64_mask): Ditto.
(_mm512_mask_cmpneq_epi32_mask): Ditto.
(_mm512_mask_cmpneq_epu32_mask): Ditto.
(_mm512_cmpneq_epu32_mask): Ditto.
(_mm512_mask_cmpneq_epi64_mask): Ditto.
(_mm512_cmpneq_epi64_mask): Ditto.
(_mm512_mask_cmpneq_epu64_mask): Ditto.
(_mm512_cmpneq_epu64_mask): Ditto.
(_mm512_castpd_ps): Ditto.
(_mm512_castpd_si512): Ditto.
(_mm512_castps_pd): Ditto.
(_mm512_castps_si512): Ditto.
(_mm512_castsi512_ps): Ditto.
(_mm512_castsi512_pd): Ditto.
(_mm512_castpd512_pd128): Ditto.
(_mm512_castps512_ps128): Ditto.
(_mm512_castsi512_si128): Ditto.
(_mm512_castpd512_pd256): Ditto.
(_mm512_castps512_ps256): Ditto.
(_mm512_castsi512_si256): Ditto.
(_mm512_castpd128_pd512): Ditto.
(_mm512_castps128_ps512): Ditto.
(_mm512_castsi128_si512): Ditto.
(_mm512_castpd256_pd512): Ditto.
(_mm512_castps256_ps512): Ditto.
(_mm512_castsi256_si512): Ditto.
(_mm512_cmpeq_epu32_mask): Ditto.
(_mm512_mask_cmpeq_epu32_mask): Ditto.
(_mm512_mask_cmpeq_epu64_mask): Ditto.
(_mm512_cmpeq_epu64_mask): Ditto.
(_mm512_cmpgt_epu32_mask): Ditto.
(_mm512_mask_cmpgt_epu32_mask): Ditto.
(_mm512_mask_cmpgt_epu64_mask): Ditto.
(_mm512_cmpgt_epu64_mask): Ditto.
* config/i386/i386-builtin-types.def: Add V16SF_FTYPE_V8SF,
V16SI_FTYPE_V8SI, V16SI_FTYPE_V4SI, V8DF_FTYPE_V2DF.
* config/i386/i386.c (enum ix86_builtins): Add
IX86_BUILTIN_SI512_SI256, IX86_BUILTIN_PD512_PD256,
IX86_BUILTIN_PS512_PS256, IX86_BUILTIN_SI512_SI,
IX86_BUILTIN_PD512_PD, IX86_BUILTIN_PS512_PS.
(bdesc_args): Add __builtin_ia32_si512_256si,
__builtin_ia32_ps512_256ps, __builtin_ia32_pd512_256pd,
__builtin_ia32_si512_si, __builtin_ia32_ps512_ps,
__builtin_ia32_pd512_pd.
(ix86_expand_args_builtin): Handle new FTYPEs.
* config/i386/sse.md (castmode): Add 512-bit modes.
(AVX512MODE2P): New.
(avx512f_<castmode><avxsizesuffix>_<castmode): New.
(avx512f_<castmode><avxsizesuffix>_256<castmode): Ditto.
gcc/testsuite/
PR target/61878
* gcc.target/i386/avx512f-typecast-1.c: New test.
* gcc.target/i386/avx512f-vpcmpequd-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpequd-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpequq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpequq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpged-1.c: Add new intrinsic.
* gcc.target/i386/avx512f-vpcmpged-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeud-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeud-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeuq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgeuq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgtud-1.c: New test.
* gcc.target/i386/avx512f-vpcmpgtud-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgtuq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpgtuq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpled-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpled-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleud-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleud-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleuq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpleuq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltd-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltd-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltud-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltud-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltuq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpltuq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpneqd-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpneqd-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpneqq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpneqq-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpnequd-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpnequd-2.c: Ditto.
* gcc.target/i386/avx512f-vpcmpnequq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpnequq-2.c: Ditto.
From-SVN: r214019
+2014-08-15 Ilya Tocar <tocarip@gmail.com>
+
+ PR target/61878
+ * config/i386/avx512fintrin.h (_mm512_mask_cmpge_epi32_mask): New.
+ (_mm512_mask_cmpge_epu32_mask): Ditto.
+ (_mm512_cmpge_epu32_mask): Ditto.
+ (_mm512_mask_cmpge_epi64_mask): Ditto.
+ (_mm512_cmpge_epi64_mask): Ditto.
+ (_mm512_mask_cmpge_epu64_mask): Ditto.
+ (_mm512_cmpge_epu64_mask): Ditto.
+ (_mm512_mask_cmple_epi32_mask): Ditto.
+ (_mm512_cmple_epi32_mask): Ditto.
+ (_mm512_mask_cmple_epu32_mask): Ditto.
+ (_mm512_cmple_epu32_mask): Ditto.
+ (_mm512_mask_cmple_epi64_mask): Ditto.
+ (_mm512_cmple_epi64_mask): Ditto.
+ (_mm512_mask_cmple_epu64_mask): Ditto.
+ (_mm512_cmple_epu64_mask): Ditto.
+ (_mm512_mask_cmplt_epi32_mask): Ditto.
+ (_mm512_cmplt_epi32_mask): Ditto.
+ (_mm512_mask_cmplt_epu32_mask): Ditto.
+ (_mm512_cmplt_epu32_mask): Ditto.
+ (_mm512_mask_cmplt_epi64_mask): Ditto.
+ (_mm512_cmplt_epi64_mask): Ditto.
+ (_mm512_mask_cmplt_epu64_mask): Ditto.
+ (_mm512_cmplt_epu64_mask): Ditto.
+ (_mm512_mask_cmpneq_epi32_mask): Ditto.
+ (_mm512_mask_cmpneq_epu32_mask): Ditto.
+ (_mm512_cmpneq_epu32_mask): Ditto.
+ (_mm512_mask_cmpneq_epi64_mask): Ditto.
+ (_mm512_cmpneq_epi64_mask): Ditto.
+ (_mm512_mask_cmpneq_epu64_mask): Ditto.
+ (_mm512_cmpneq_epu64_mask): Ditto.
+ (_mm512_castpd_ps): Ditto.
+ (_mm512_castpd_si512): Ditto.
+ (_mm512_castps_pd): Ditto.
+ (_mm512_castps_si512): Ditto.
+ (_mm512_castsi512_ps): Ditto.
+ (_mm512_castsi512_pd): Ditto.
+ (_mm512_castpd512_pd128): Ditto.
+ (_mm512_castps512_ps128): Ditto.
+ (_mm512_castsi512_si128): Ditto.
+ (_mm512_castpd512_pd256): Ditto.
+ (_mm512_castps512_ps256): Ditto.
+ (_mm512_castsi512_si256): Ditto.
+ (_mm512_castpd128_pd512): Ditto.
+ (_mm512_castps128_ps512): Ditto.
+ (_mm512_castsi128_si512): Ditto.
+ (_mm512_castpd256_pd512): Ditto.
+ (_mm512_castps256_ps512): Ditto.
+ (_mm512_castsi256_si512): Ditto.
+ (_mm512_cmpeq_epu32_mask): Ditto.
+ (_mm512_mask_cmpeq_epu32_mask): Ditto.
+ (_mm512_mask_cmpeq_epu64_mask): Ditto.
+ (_mm512_cmpeq_epu64_mask): Ditto.
+ (_mm512_cmpgt_epu32_mask): Ditto.
+ (_mm512_mask_cmpgt_epu32_mask): Ditto.
+ (_mm512_mask_cmpgt_epu64_mask): Ditto.
+ (_mm512_cmpgt_epu64_mask): Ditto.
+ * config/i386/i386-builtin-types.def: Add V16SF_FTYPE_V8SF,
+ V16SI_FTYPE_V8SI, V16SI_FTYPE_V4SI, V8DF_FTYPE_V2DF.
+ * config/i386/i386.c (enum ix86_builtins): Add
+ IX86_BUILTIN_SI512_SI256, IX86_BUILTIN_PD512_PD256,
+ IX86_BUILTIN_PS512_PS256, IX86_BUILTIN_SI512_SI,
+ IX86_BUILTIN_PD512_PD, IX86_BUILTIN_PS512_PS.
+ (bdesc_args): Add __builtin_ia32_si512_256si,
+ __builtin_ia32_ps512_256ps, __builtin_ia32_pd512_256pd,
+ __builtin_ia32_si512_si, __builtin_ia32_ps512_ps,
+ __builtin_ia32_pd512_pd.
+ (ix86_expand_args_builtin): Handle new FTYPEs.
+ * config/i386/sse.md (castmode): Add 512-bit modes.
+ (AVX512MODE2P): New.
+ (avx512f_<castmode><avxsizesuffix>_<castmode): New.
+ (avx512f_<castmode><avxsizesuffix>_256<castmode): Ditto.
+
2014-08-15 Richard Biener <rguenther@suse.de>
* fold-const.c (tree_swap_operands_p): Put all constants
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpge_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 5,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpge_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 5,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpge_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 5,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epi64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpge_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 5,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmple_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 2,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmple_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 2,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmple_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 2,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmple_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 2,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmplt_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 1,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmplt_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 1,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmplt_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 1,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmplt_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 1,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpneq_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 4,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpneq_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
+ (__v16si) __Y, 4,
+ (__mmask16) __M);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu32_mask (__m512i __X, __m512i __Y)
{
return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpneq_epi64_mask (__mmask16 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 4,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epi64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpneq_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
+ (__v8di) __Y, 4,
+ (__mmask8) __M);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu64_mask (__m512i __X, __m512i __Y)
{
return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
return __builtin_ia32_kmov16 (__A);
}
+extern __inline __m512
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd_ps (__m512d __A)
+{
+ return (__m512) (__A);
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd_si512 (__m512d __A)
+{
+ return (__m512i) (__A);
+}
+
+extern __inline __m512d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps_pd (__m512 __A)
+{
+ return (__m512d) (__A);
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps_si512 (__m512 __A)
+{
+ return (__m512i) (__A);
+}
+
+extern __inline __m512
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi512_ps (__m512i __A)
+{
+ return (__m512) (__A);
+}
+
+extern __inline __m512d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi512_pd (__m512i __A)
+{
+ return (__m512d) (__A);
+}
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd512_pd128 (__m512d __A)
+{
+ return (__m128d)_mm512_extractf32x4_ps((__m512)__A, 0);
+}
+
+extern __inline __m128
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps512_ps128 (__m512 __A)
+{
+ return _mm512_extractf32x4_ps(__A, 0);
+}
+
+extern __inline __m128i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi512_si128 (__m512i __A)
+{
+ return (__m128i)_mm512_extracti32x4_epi32((__m512i)__A, 0);
+}
+
+extern __inline __m256d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd512_pd256 (__m512d __A)
+{
+ return _mm512_extractf64x4_pd(__A, 0);
+}
+
+extern __inline __m256
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps512_ps256 (__m512 __A)
+{
+ return (__m256)_mm512_extractf64x4_pd((__m512d)__A, 0);
+}
+
+extern __inline __m256i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi512_si256 (__m512i __A)
+{
+ return (__m256i)_mm512_extractf64x4_pd((__m512d)__A, 0);
+}
+
+extern __inline __m512d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd128_pd512 (__m128d __A)
+{
+ return (__m512d) __builtin_ia32_pd512_pd((__m128d)__A);
+}
+
+extern __inline __m512
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps128_ps512 (__m128 __A)
+{
+ return (__m512) __builtin_ia32_ps512_ps((__m128)__A);
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi128_si512 (__m128i __A)
+{
+ return (__m512i) __builtin_ia32_si512_si((__v4si)__A);
+}
+
+extern __inline __m512d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castpd256_pd512 (__m256d __A)
+{
+ return __builtin_ia32_pd512_256pd (__A);
+}
+
+extern __inline __m512
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castps256_ps512 (__m256 __A)
+{
+ return __builtin_ia32_ps512_256ps (__A);
+}
+
+extern __inline __m512i
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_castsi256_si512 (__m256i __A)
+{
+ return (__m512i)__builtin_ia32_si512_256si ((__v8si)__A);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_cmpeq_epu32_mask (__m512i __A, __m512i __B)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
+ (__v16si) __B, 0,
+ (__mmask16) -1);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpeq_epu32_mask (__mmask16 __U, __m512i __A, __m512i __B)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
+ (__v16si) __B, 0, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpeq_epu64_mask (__mmask8 __U, __m512i __A, __m512i __B)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
+ (__v8di) __B, 0, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_cmpeq_epu64_mask (__m512i __A, __m512i __B)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
+ (__v8di) __B, 0,
+ (__mmask8) -1);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_cmpgt_epu32_mask (__m512i __A, __m512i __B)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
+ (__v16si) __B, 6,
+ (__mmask16) -1);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpgt_epu32_mask (__mmask16 __U, __m512i __A, __m512i __B)
+{
+ return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
+ (__v16si) __B, 6, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmpgt_epu64_mask (__mmask8 __U, __m512i __A, __m512i __B)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
+ (__v8di) __B, 6, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_cmpgt_epu64_mask (__m512i __A, __m512i __B)
+{
+ return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
+ (__v8di) __B, 6,
+ (__mmask8) -1);
+}
+
#ifdef __DISABLE_AVX512F__
#undef __DISABLE_AVX512F__
#pragma GCC pop_options
DEF_FUNCTION_TYPE (V8DF, DOUBLE)
DEF_FUNCTION_TYPE (V8DI, INT64)
DEF_FUNCTION_TYPE (V16SF, V4SF)
+DEF_FUNCTION_TYPE (V16SF, V8SF)
DEF_FUNCTION_TYPE (V8DF, V4DF)
+DEF_FUNCTION_TYPE (V8DF, V2DF)
DEF_FUNCTION_TYPE (V8DI, V4DI)
DEF_FUNCTION_TYPE (V16QI, V8DI)
DEF_FUNCTION_TYPE (UINT, V4SF)
DEF_FUNCTION_TYPE (UINT, V2DF)
DEF_FUNCTION_TYPE (UINT64, V2DF)
DEF_FUNCTION_TYPE (V16SI, V16SI)
+DEF_FUNCTION_TYPE (V16SI, V4SI)
+DEF_FUNCTION_TYPE (V16SI, V8SI)
DEF_FUNCTION_TYPE (V16SI, V16SI, V16SI, HI)
DEF_FUNCTION_TYPE (V8DI, V8DI)
DEF_FUNCTION_TYPE (V8DI, V8DI, V8DI, QI)
IX86_BUILTIN_GATHERDIV8SI,
/* AVX512F */
+ IX86_BUILTIN_SI512_SI256,
+ IX86_BUILTIN_PD512_PD256,
+ IX86_BUILTIN_PS512_PS256,
+ IX86_BUILTIN_SI512_SI,
+ IX86_BUILTIN_PD512_PD,
+ IX86_BUILTIN_PS512_PS,
IX86_BUILTIN_ADDPD512,
IX86_BUILTIN_ADDPS512,
IX86_BUILTIN_ADDSD_ROUND,
{ OPTION_MASK_ISA_BMI2, CODE_FOR_bmi2_pext_di3, "__builtin_ia32_pext_di", IX86_BUILTIN_PEXT64, UNKNOWN, (int) UINT64_FTYPE_UINT64_UINT64 },
/* AVX512F */
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_si512_256si, "__builtin_ia32_si512_256si", IX86_BUILTIN_SI512_SI256, UNKNOWN, (int) V16SI_FTYPE_V8SI },
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_ps512_256ps, "__builtin_ia32_ps512_256ps", IX86_BUILTIN_PS512_PS256, UNKNOWN, (int) V16SF_FTYPE_V8SF },
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_pd512_256pd, "__builtin_ia32_pd512_256pd", IX86_BUILTIN_PD512_PD256, UNKNOWN, (int) V8DF_FTYPE_V4DF },
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_si512_si, "__builtin_ia32_si512_si", IX86_BUILTIN_SI512_SI, UNKNOWN, (int) V16SI_FTYPE_V4SI },
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_ps512_ps, "__builtin_ia32_ps512_ps", IX86_BUILTIN_PS512_PS, UNKNOWN, (int) V16SF_FTYPE_V4SF },
+ { OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_pd512_pd, "__builtin_ia32_pd512_pd", IX86_BUILTIN_PD512_PD, UNKNOWN, (int) V8DF_FTYPE_V2DF },
{ OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_alignv16si_mask, "__builtin_ia32_alignd512_mask", IX86_BUILTIN_ALIGND512, UNKNOWN, (int) V16SI_FTYPE_V16SI_V16SI_INT_V16SI_HI },
{ OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_alignv8di_mask, "__builtin_ia32_alignq512_mask", IX86_BUILTIN_ALIGNQ512, UNKNOWN, (int) V8DI_FTYPE_V8DI_V8DI_INT_V8DI_QI },
{ OPTION_MASK_ISA_AVX512F, CODE_FOR_avx512f_blendmv16si, "__builtin_ia32_blendmd_512_mask", IX86_BUILTIN_BLENDMD512, UNKNOWN, (int) V16SI_FTYPE_V16SI_V16SI_HI },
case V16SI_FTYPE_V16SI:
case V16SI_FTYPE_INT:
case V16SF_FTYPE_FLOAT:
+ case V16SF_FTYPE_V8SF:
+ case V16SI_FTYPE_V8SI:
case V16SF_FTYPE_V4SF:
+ case V16SI_FTYPE_V4SI:
case V16SF_FTYPE_V16SF:
case V8HI_FTYPE_V8DI:
case V8UHI_FTYPE_V8UHI:
case V8DI_FTYPE_V8DI:
case V8DF_FTYPE_DOUBLE:
case V8DF_FTYPE_V4DF:
+ case V8DF_FTYPE_V2DF:
case V8DF_FTYPE_V8DF:
case V8DF_FTYPE_V8SI:
nargs = 1;
[(V16QI "7") (V8HI "15") (V4SI "31") (V2DI "63")])
;; Mapping of mode to cast intrinsic name
-(define_mode_attr castmode [(V8SI "si") (V8SF "ps") (V4DF "pd")])
+(define_mode_attr castmode
+ [(V8SI "si") (V8SF "ps") (V4DF "pd")
+ (V16SI "si") (V16SF "ps") (V8DF "pd")])
;; Instruction suffix for sign and zero extensions.
(define_code_attr extsuffix [(sign_extend "sx") (zero_extend "zx")])
;; Mix-n-match
(define_mode_iterator AVX256MODE2P [V8SI V8SF V4DF])
+(define_mode_iterator AVX512MODE2P [V16SI V16SF V8DF])
;; Mapping for dbpsabbw modes
(define_mode_attr dbpsadbwmode
[(set_attr "type" "sselog1")
(set_attr "length_immediate" "1")
(set_attr "mode" "TI")])
+
+(define_insn_and_split "avx512f_<castmode><avxsizesuffix>_<castmode>"
+ [(set (match_operand:AVX512MODE2P 0 "nonimmediate_operand" "=x,m")
+ (unspec:AVX512MODE2P
+ [(match_operand:<ssequartermode> 1 "nonimmediate_operand" "xm,x")]
+ UNSPEC_CAST))]
+ "TARGET_AVX512F"
+ "#"
+ "&& reload_completed"
+ [(const_int 0)]
+{
+ rtx op0 = operands[0];
+ rtx op1 = operands[1];
+ if (REG_P (op0))
+ op0 = gen_rtx_REG (<ssequartermode>mode, REGNO (op0));
+ else
+ op1 = gen_rtx_REG (<MODE>mode, REGNO (op1));
+ emit_move_insn (op0, op1);
+ DONE;
+})
+
+(define_insn_and_split "avx512f_<castmode><avxsizesuffix>_256<castmode>"
+ [(set (match_operand:AVX512MODE2P 0 "nonimmediate_operand" "=x,m")
+ (unspec:AVX512MODE2P
+ [(match_operand:<ssehalfvecmode> 1 "nonimmediate_operand" "xm,x")]
+ UNSPEC_CAST))]
+ "TARGET_AVX512F"
+ "#"
+ "&& reload_completed"
+ [(const_int 0)]
+{
+ rtx op0 = operands[0];
+ rtx op1 = operands[1];
+ if (REG_P (op0))
+ op0 = gen_rtx_REG (<ssehalfvecmode>mode, REGNO (op0));
+ else
+ op1 = gen_rtx_REG (<MODE>mode, REGNO (op1));
+ emit_move_insn (op0, op1);
+ DONE;
+})
+2014-08-15 Ilya Tocar <tocarip@gmail.com>
+
+ PR target/61878
+ * gcc.target/i386/avx512f-typecast-1.c: New test.
+ * gcc.target/i386/avx512f-vpcmpequd-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpequd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpequq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpequq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpged-1.c: Add new intrinsic.
+ * gcc.target/i386/avx512f-vpcmpged-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeud-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeud-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeuq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgeuq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgtud-1.c: New test.
+ * gcc.target/i386/avx512f-vpcmpgtud-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgtuq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpgtuq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpled-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpled-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleud-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleud-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleuq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpleuq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltd-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltud-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltud-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltuq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpltuq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpneqd-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpneqd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpneqq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpneqq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpnequd-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpnequd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpnequq-1.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpnequq-2.c: Ditto.
+
2014-08-15 Jakub Jelinek <jakub@redhat.com>
Tobias Burnus <burnus@net-b.de>
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-require-effective-target avx512f } */
+
+#include "avx512f-check.h"
+
+
+void
+avx512f_test (void)
+{
+ union512i_d a, ad;
+ union512 b, bd;
+ union512d c, cd;
+ union256i_d d, dd;
+ union256 e, ed;
+ union256d f, fd;
+ union128i_d g, gd;
+ union128 h, hd;
+ union128d k, kd;
+ int i;
+
+ for (i = 0; i < 16; i++)
+ {
+ a.a[i] = 7146908634 + i;
+ b.a[i] = 45.12f + i;
+ }
+
+ for (i = 0; i < 8; i++)
+ {
+ c.a[i] = 41234512513451345.0905 + i;
+ d.a[i] = 109534 + i;
+ e.a[i] = 85034.095f + i;
+ }
+
+ for (i = 0; i < 4; i++)
+ {
+ f.a[i] = 41234512451345.0905 + i;
+ g.a[i] = 71469086341 + i;
+ h.a[i] = 45.1264f + i;
+ }
+
+ for (i = 0; i < 2; i++)
+ {
+ k.a[i] = 7146908634.576 + i;
+ }
+
+ bd.x = _mm512_castpd_ps (c.x);
+ if (memcmp(bd.a, c.a, 64))
+ abort ();
+
+ ad.x = _mm512_castpd_si512 (c.x);
+ if (memcmp(ad.a, c.a, 64))
+ abort ();
+
+ cd.x = _mm512_castps_pd (b.x);
+ if (memcmp(cd.a, b.a, 64))
+ abort ();
+
+ ad.x = _mm512_castps_si512 (b.x);
+ if (memcmp(ad.a, b.a, 64))
+ abort ();
+
+ bd.x = _mm512_castsi512_ps (a.x);
+ if (memcmp(bd.a, a.a, 64))
+ abort ();
+
+ cd.x = _mm512_castsi512_pd (a.x);
+ if (memcmp(cd.a, a.a, 64))
+ abort ();
+
+ kd.x = _mm512_castpd512_pd128 (c.x);
+ if (memcmp(kd.a, c.a, 16))
+ abort ();
+
+ hd.x = _mm512_castps512_ps128 (b.x);
+ if (memcmp(hd.a, b.a, 16))
+ abort ();
+
+ gd.x = _mm512_castsi512_si128 (a.x);
+ if (memcmp(gd.a, a.a, 16))
+ abort ();
+
+ fd.x = _mm512_castpd512_pd256 (c.x);
+ if (memcmp(fd.a, c.a, 32))
+ abort ();
+
+ ed.x = _mm512_castps512_ps256 (b.x);
+ if (memcmp(ed.a, b.a, 32))
+ abort ();
+
+ dd.x = _mm512_castsi512_si256 (a.x);
+ if (memcmp(dd.a, a.a, 32))
+ abort ();
+
+ cd.x = _mm512_castpd128_pd512 (k.x);
+ if (memcmp(cd.a, k.a, 16))
+ abort ();
+
+ bd.x = _mm512_castps128_ps512 (h.x);
+ if (memcmp(bd.a, h.a, 16))
+ abort ();
+
+ ad.x = _mm512_castsi128_si512 (g.x);
+ if (memcmp(ad.a, g.a, 16))
+ abort ();
+
+ cd.x = _mm512_castpd256_pd512 (f.x);
+ if (memcmp(cd.a, f.a, 32))
+ abort ();
+
+ bd.x = _mm512_castps256_ps512 (e.x);
+ if (memcmp(bd.a, e.a, 32))
+ abort ();
+
+ ad.x = _mm512_castsi256_si512 (d.x);
+ if (memcmp(ad.a, d.a, 32))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -O2" } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __mmask16 m;
+
+void extern
+avx512f_test (void)
+{
+ m = _mm512_cmpeq_epu32_mask (x, x);
+ m = _mm512_mask_cmpeq_epu32_mask (3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-require-effective-target avx512f } */
+
+#define AVX512F
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+static void
+CALC (MASK_TYPE *r, unsigned int *s1, unsigned int *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] == s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+ res1 = 0;
+ res2 = 0;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpeq_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpeq_epu32_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -O2" } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __mmask8 m;
+
+void extern
+avx512f_test (void)
+{
+ m = _mm512_cmpeq_epu64_mask (x, x);
+ m = _mm512_mask_cmpeq_epu64_mask (3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-require-effective-target avx512f } */
+
+#define AVX512F
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+static void
+CALC (MASK_TYPE *r, unsigned long long *s1, unsigned long long *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] == s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
+ MASK_TYPE res1, res2, res_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ res1 = 0;
+ res2 = 0;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpeq_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpeq_epu64_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpge_epi32_mask (x, x);
+ m = _mm512_mask_cmpge_epi32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpge_epi32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpge_epi32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpge_epi64_mask (x, x);
+ m = _mm512_mask_cmpge_epi64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpge_epi64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpge_epi64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpge_epu32_mask (x, x);
+ m = _mm512_mask_cmpge_epu32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpge_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpge_epu32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpge_epu64_mask (x, x);
+ m = _mm512_mask_cmpge_epu64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpge_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpge_epu64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -O2" } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __mmask16 m;
+
+void extern
+avx512f_test (void)
+{
+ m = _mm512_cmpgt_epu32_mask (x, x);
+ m = _mm512_mask_cmpgt_epu32_mask (3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-require-effective-target avx512f } */
+
+#define AVX512F
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+static void
+CALC (MASK_TYPE *r, unsigned int *s1, unsigned int *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] > s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+ res1 = 0;
+ res2 = 0;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpgt_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpgt_epu32_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512f -O2" } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __mmask8 m;
+
+void extern
+avx512f_test (void)
+{
+ m = _mm512_cmpgt_epu64_mask (x, x);
+ m = _mm512_mask_cmpgt_epu64_mask (3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-require-effective-target avx512f } */
+
+#define AVX512F
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+static void
+CALC (MASK_TYPE *r, unsigned long long *s1, unsigned long long *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] > s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
+ MASK_TYPE res1, res2, res_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ res1 = 0;
+ res2 = 0;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpgt_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpgt_epu64_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmple_epi32_mask (x, x);
+ m = _mm512_mask_cmple_epi32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmple_epi32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmple_epi32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmple_epi64_mask (x, x);
+ m = _mm512_mask_cmple_epi64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmple_epi64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmple_epi64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmple_epu32_mask (x, x);
+ m = _mm512_mask_cmple_epu32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmple_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmple_epu32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmple_epu64_mask (x, x);
+ m = _mm512_mask_cmple_epu64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmple_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmple_epu64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmplt_epi32_mask (x, x);
+ m = _mm512_mask_cmplt_epi32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmplt_epi32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmplt_epi32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmplt_epi64_mask (x, x);
+ m = _mm512_mask_cmplt_epi64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmplt_epi64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmplt_epi64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmplt_epu32_mask (x, x);
+ m = _mm512_mask_cmplt_epu32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmplt_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmplt_epu32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmplt_epu64_mask (x, x);
+ m = _mm512_mask_cmplt_epu64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmplt_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmplt_epu64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpneq_epi32_mask (x, x);
+ m = _mm512_mask_cmpneq_epi32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpneq_epi32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpneq_epi32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpneq_epi64_mask (x, x);
+ m = _mm512_mask_cmpneq_epi64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpneq_epi64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpneq_epi64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+ if (res2 != res_ref)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpneq_epu32_mask (x, x);
+ m = _mm512_mask_cmpneq_epu32_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_d) src1, src2;
- MASK_TYPE res_ref, res1;
+ MASK_TYPE res_ref, res1, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpneq_epu32_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpneq_epu32_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res_ref != res1)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res_ref != res2)
+ abort ();
}
/* { dg-do compile } */
/* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[0-9\]\{" 1 } } */
#include <immintrin.h>
avx512f_test (void)
{
m = _mm512_cmpneq_epu64_mask (x, x);
+ m = _mm512_mask_cmpneq_epu64_mask (m, x, x);
}
{
int i;
UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
- MASK_TYPE res1, res_ref;
+ MASK_TYPE res1, res_ref, res2;
res1 = 0;
for (i = 0; i < SIZE / 2; i++)
}
res1 = INTRINSIC (_cmpneq_epu64_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpneq_epu64_mask) (MASK_VALUE, src1.x, src2.x);
CALC (&res_ref, src1.a, src2.a);
if (res1 != res_ref)
abort ();
+
+ res_ref &= MASK_VALUE;
+
+ if (res2 != res_ref)
+ abort ();
}
projects / platform / upstream / gcc.git / commitdiff