This is an archive of the discontinued LLVM Phabricator instance.

Differential D20815

[Clang][Intrinsics][avx512] Adding round roundscale to clang
ClosedPublic

Authored by m_zuckerman on May 31 2016, 7:29 AM.

Details

Reviewers
AsafBadouh
delena
igorb
Commits
rGc301c194ecf5: [Clang][Intrinsics][avx512] Adding round roundscale to clang
rC271368: [Clang][Intrinsics][avx512] Adding round roundscale to clang
rL271368: [Clang][Intrinsics][avx512] Adding round roundscale to clang

Diff Detail

Repository
rL LLVM

Event Timeline

m_zuckerman updated this revision to Diff 59064.May 31 2016, 7:29 AM
m_zuckerman retitled this revision from to [Clang][Intrinsics][avx512] Continue Adding round roundscale to clang.
m_zuckerman updated this object.
m_zuckerman added reviewers: AsafBadouh, igorb, delena.
m_zuckerman added a subscriber: cfe-commits.
m_zuckerman retitled this revision from [Clang][Intrinsics][avx512] Continue Adding round roundscale to clang to [Clang][Intrinsics][avx512] Adding round roundscale to clang.May 31 2016, 7:36 AM
delena accepted this revision.May 31 2016, 7:50 AM
delena edited edge metadata.
This revision is now accepted and ready to land.May 31 2016, 7:50 AM
Closed by commit rL271368: [Clang][Intrinsics][avx512] Adding round roundscale to clang (authored by mzuckerm). · Explain WhyJun 1 2016, 12:42 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
cfe/
trunk/
lib/
Headers/
36 lines
test/
CodeGen/
42 lines

Diff 59175

cfe/trunk/lib/Headers/avx512fintrin.h

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 2,376 Lines • ▼ Show 20 Lines (__m512)__builtin_ia32_rndscaleps_mask((__v16sf)(__m512)(C), (int)(imm), \
_MM_FROUND_CUR_DIRECTION); }) _MM_FROUND_CUR_DIRECTION); })
#define _mm512_maskz_roundscale_ps(A, B, imm) __extension__ ({\ #define _mm512_maskz_roundscale_ps(A, B, imm) __extension__ ({\
(__m512)__builtin_ia32_rndscaleps_mask((__v16sf)(__m512)(B), (int)(imm), \ (__m512)__builtin_ia32_rndscaleps_mask((__v16sf)(__m512)(B), (int)(imm), \
(__v16sf)_mm512_setzero_ps(), \ (__v16sf)_mm512_setzero_ps(), \
(__mmask16)(A), \ (__mmask16)(A), \
_MM_FROUND_CUR_DIRECTION); }) _MM_FROUND_CUR_DIRECTION); })
#define _mm512_mask_roundscale_round_ps( __A, __B, __C, __imm, __R) __extension__ ({ \
(__m512)__builtin_ia32_rndscaleps_mask ((__v16sf)( __C), (int)__imm,\
(__v16sf)( __A),\
(__mmask16)( __B),(int) __R);\
})
#define _mm512_maskz_roundscale_round_ps( __A, __B, __imm,__R) __extension__ ({ \
(__m512)__builtin_ia32_rndscaleps_mask ((__v16sf)( __B), (int)__imm,\
(__v16sf)_mm512_setzero_ps (),\
(__mmask16)( __A),(int) __R);\
})
#define _mm512_roundscale_round_ps( __A, __imm, __R) __extension__ ({ \
(__m512)__builtin_ia32_rndscaleps_mask ((__v16sf)( __A),(int) __imm,\
(__v16sf) _mm512_undefined_ps (),\
(__mmask16) -1,(int) __R);\
})
#define _mm512_roundscale_pd(A, B) __extension__ ({ \ #define _mm512_roundscale_pd(A, B) __extension__ ({ \
(__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(A), (int)(B), \ (__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(A), (int)(B), \
(__v8df)(__m512d)(A), (__mmask8)-1, \ (__v8df)(__m512d)(A), (__mmask8)-1, \
_MM_FROUND_CUR_DIRECTION); }) _MM_FROUND_CUR_DIRECTION); })
#define _mm512_mask_roundscale_pd(A, B, C, imm) __extension__ ({\ #define _mm512_mask_roundscale_pd(A, B, C, imm) __extension__ ({\
(__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(C), (int)(imm), \ (__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(C), (int)(imm), \
(__v8df)(__m512d)(A), (__mmask8)(B), \ (__v8df)(__m512d)(A), (__mmask8)(B), \
_MM_FROUND_CUR_DIRECTION); }) _MM_FROUND_CUR_DIRECTION); })
#define _mm512_maskz_roundscale_pd(A, B, imm) __extension__ ({\ #define _mm512_maskz_roundscale_pd(A, B, imm) __extension__ ({\
(__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(B), (int)(imm), \ (__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(__m512d)(B), (int)(imm), \
(__v8df)_mm512_setzero_pd(), \ (__v8df)_mm512_setzero_pd(), \
(__mmask8)(A), \ (__mmask8)(A), \
_MM_FROUND_CUR_DIRECTION); }) _MM_FROUND_CUR_DIRECTION); })
#define _mm512_mask_roundscale_round_pd( __A, __B, __C, __imm ,__R) __extension__ ({ \
(__m512d)__builtin_ia32_rndscalepd_mask ((__v8df)( __C),(int)__imm,\
(__v8df)( __A),\
(__mmask8)( __B),(int)__R);\
})
#define _mm512_maskz_roundscale_round_pd( __A, __B, __imm, __R) __extension__ ({ \
(__m512d)__builtin_ia32_rndscalepd_mask ((__v8df)( __B),(int)__imm,\
(__v8df)_mm512_setzero_pd (),\
(__mmask8)( __A),(int) __R);\
})
#define _mm512_roundscale_round_pd( __A, __imm , __R) __extension__ ({ \
(__m512d)__builtin_ia32_rndscalepd_mask ((__v8df)( __A),(int) __imm,\
(__v8df)_mm512_undefined_pd (),\
(__mmask8) -1,(int) __R);\
})
#define _mm512_fmadd_round_pd(A, B, C, R) __extension__ ({ \ #define _mm512_fmadd_round_pd(A, B, C, R) __extension__ ({ \
(__m512d)__builtin_ia32_vfmaddpd512_mask((__v8df)(__m512d)(A), \ (__m512d)__builtin_ia32_vfmaddpd512_mask((__v8df)(__m512d)(A), \
(__v8df)(__m512d)(B), \ (__v8df)(__m512d)(B), \
(__v8df)(__m512d)(C), (__mmask8)-1, \ (__v8df)(__m512d)(C), (__mmask8)-1, \
(int)(R)); }) (int)(R)); })
#define _mm512_mask_fmadd_round_pd(A, U, B, C, R) __extension__ ({ \ #define _mm512_mask_fmadd_round_pd(A, U, B, C, R) __extension__ ({ \
▲ Show 20 Lines • Show All 6,844 Lines • Show Last 20 Lines

cfe/trunk/test/CodeGen/avx512f-builtins.c

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 6,647 Lines • ▼ Show 20 Lines
__m512 test_mm512_maskz_roundscale_ps(__mmask16 __U, __m512 __A) __m512 test_mm512_maskz_roundscale_ps(__mmask16 __U, __m512 __A)
{ {
// CHECK-LABEL: @test_mm512_maskz_roundscale_ps // CHECK-LABEL: @test_mm512_maskz_roundscale_ps
// CHECK: @llvm.x86.avx512.mask.rndscale.ps.512 // CHECK: @llvm.x86.avx512.mask.rndscale.ps.512
return _mm512_maskz_roundscale_ps(__U,__A, 1); return _mm512_maskz_roundscale_ps(__U,__A, 1);
} }
__m512 test_mm512_mask_roundscale_round_ps(__m512 __A,__mmask16 __U,__m512 __C)
{
// CHECK-LABEL: @test_mm512_mask_roundscale_round_ps
// CHECK: @llvm.x86.avx512.mask.rndscale.ps.512
return _mm512_mask_roundscale_round_ps(__A,__U,__C,3,_MM_FROUND_CUR_DIRECTION);
}
__m512 test_mm512_maskz_roundscale_round_ps(__m512 __A,__mmask16 __U)
{
// CHECK-LABEL: @test_mm512_maskz_roundscale_round_ps
// CHECK: @llvm.x86.avx512.mask.rndscale.ps.512
return _mm512_maskz_roundscale_round_ps(__U,__A,3,_MM_FROUND_CUR_DIRECTION);
}
__m512 test_mm512_roundscale_round_ps(__m512 __A)
{
// CHECK-LABEL: @test_mm512_roundscale_round_ps
// CHECK: @llvm.x86.avx512.mask.rndscale.ps.512
return _mm512_roundscale_round_ps(__A,3,_MM_FROUND_CUR_DIRECTION);
}
__m512d test_mm512_mask_roundscale_pd(__m512d __W, __mmask8 __U, __m512d __A) __m512d test_mm512_mask_roundscale_pd(__m512d __W, __mmask8 __U, __m512d __A)
{ {
// CHECK-LABEL: @test_mm512_mask_roundscale_pd // CHECK-LABEL: @test_mm512_mask_roundscale_pd
// CHECK: @llvm.x86.avx512.mask.rndscale.pd.512 // CHECK: @llvm.x86.avx512.mask.rndscale.pd.512
return _mm512_mask_roundscale_pd(__W,__U,__A, 1); return _mm512_mask_roundscale_pd(__W,__U,__A, 1);
} }
__m512d test_mm512_maskz_roundscale_pd(__mmask8 __U, __m512d __A) __m512d test_mm512_maskz_roundscale_pd(__mmask8 __U, __m512d __A)
{ {
// CHECK-LABEL: @test_mm512_maskz_roundscale_pd // CHECK-LABEL: @test_mm512_maskz_roundscale_pd
// CHECK: @llvm.x86.avx512.mask.rndscale.pd.512 // CHECK: @llvm.x86.avx512.mask.rndscale.pd.512
return _mm512_maskz_roundscale_pd(__U,__A, 1); return _mm512_maskz_roundscale_pd(__U,__A, 1);
} }
__m512d test_mm512_mask_roundscale_round_pd(__m512d __A,__mmask8 __U,__m512d __C)
{
// CHECK-LABEL: @test_mm512_mask_roundscale_round_pd
// CHECK: @llvm.x86.avx512.mask.rndscale.pd.512
return _mm512_mask_roundscale_round_pd(__A,__U,__C,3,_MM_FROUND_CUR_DIRECTION);
}
__m512d test_mm512_maskz_roundscale_round_pd(__m512d __A,__mmask8 __U)
{
// CHECK-LABEL: @test_mm512_maskz_roundscale_round_pd
// CHECK: @llvm.x86.avx512.mask.rndscale.pd.512
return _mm512_maskz_roundscale_round_pd(__U,__A,3,_MM_FROUND_CUR_DIRECTION);
}
__m512d test_mm512_roundscale_round_pd(__m512d __A)
{
// CHECK-LABEL: @test_mm512_roundscale_round_pd
// CHECK: @llvm.x86.avx512.mask.rndscale.pd.512
return _mm512_roundscale_round_pd(__A,3,_MM_FROUND_CUR_DIRECTION);
}
__m512i test_mm512_mask_max_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B) __m512i test_mm512_mask_max_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{ {
// CHECK-LABEL: @test_mm512_mask_max_epi32 // CHECK-LABEL: @test_mm512_mask_max_epi32
// CHECK: @llvm.x86.avx512.mask.pmaxs.d.512 // CHECK: @llvm.x86.avx512.mask.pmaxs.d.512
return _mm512_mask_max_epi32 (__W,__M,__A,__B); return _mm512_mask_max_epi32 (__W,__M,__A,__B);
} }
__m512i test_mm512_maskz_max_epi32 (__mmask16 __M, __m512i __A, __m512i __B) __m512i test_mm512_maskz_max_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
▲ Show 20 Lines • Show All 359 Lines • Show Last 20 Lines