This is an archive of the discontinued LLVM Phabricator instance.

Differential D20068

[Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ].
ClosedPublic

Authored by m_zuckerman on May 9 2016, 7:33 AM.

Details

Reviewers
AsafBadouh
delena
igorb
Commits
rGcb8567747118: [Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ].
rC269649: [Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ].
rL269649: [Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ].

Diff Detail

Event Timeline

m_zuckerman updated this revision to Diff 56576.May 9 2016, 7:33 AM
m_zuckerman retitled this revision from to [Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ]..
m_zuckerman updated this object.
m_zuckerman added reviewers: AsafBadouh, igorb, delena.
m_zuckerman added a subscriber: llvm-commits.
delena accepted this revision.May 15 2016, 5:33 AM
delena edited edge metadata.
This revision is now accepted and ready to land.May 15 2016, 5:33 AM
Closed by commit rL269649: [Clang][AVX512] completing missing intrinsics [vsqrt|vrsqrt|vrcp14 ]. (authored by mzuckerm). · Explain WhyMay 16 2016, 4:48 AM
This revision was automatically updated to reflect the committed changes.
craig.topper added a subscriber: craig.topper.May 16 2016, 8:49 AM

Please send clang reviews to cfe-commits not llvm-commits in the future.

Revision Contents

PathSize
lib/
Headers/
avx512fintrin.h
avx512fintrin.h (revision 268910)
106 lines
test/
CodeGen/
avx512f-builtins.c
avx512f-builtins.c (revision 268910)
84 lines

Diff 56576

lib/Headers/avx512fintrin.h

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,137 Lines • ▼ Show 20 Lines
_mm512_sqrt_pd(__m512d __a) _mm512_sqrt_pd(__m512d __a)
{ {
return (__m512d)__builtin_ia32_sqrtpd512_mask((__v8df)__a, return (__m512d)__builtin_ia32_sqrtpd512_mask((__v8df)__a,
(__v8df) _mm512_setzero_pd (), (__v8df) _mm512_setzero_pd (),
(__mmask8) -1, (__mmask8) -1,
_MM_FROUND_CUR_DIRECTION); _MM_FROUND_CUR_DIRECTION);
} }
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
(__v8df) __W,
(__mmask8) __U,
_MM_FROUND_CUR_DIRECTION);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
(__v8df)
_mm512_setzero_pd (),
(__mmask8) __U,
_MM_FROUND_CUR_DIRECTION);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_sqrt_ps(__m512 __a) _mm512_sqrt_ps(__m512 __a)
{ {
return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)__a, return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)__a,
(__v16sf) _mm512_setzero_ps (), (__v16sf) _mm512_setzero_ps (),
(__mmask16) -1, (__mmask16) -1,
_MM_FROUND_CUR_DIRECTION); _MM_FROUND_CUR_DIRECTION);
} }
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_sqrt_ps(_m512 __W, __mmask8 __U, __m512 __A)
{
return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)__a,
(__v16sf) __W,
(__mmask16) __U,
_MM_FROUND_CUR_DIRECTION);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_sqrt_ps( __mmask8 __U, __m512 __A)
{
return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)__a,
(__v16sf) _mm512_setzero_ps (),
(__mmask16) __U,
_MM_FROUND_CUR_DIRECTION);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_rsqrt14_pd(__m512d __A) _mm512_rsqrt14_pd(__m512d __A)
{ {
return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A, return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
(__v8df) (__v8df)
_mm512_setzero_pd (), _mm512_setzero_pd (),
(__mmask8) -1);} (__mmask8) -1);}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_rsqrt14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
(__v8df) __W,
(__mmask8) __U);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_rsqrt14_pd (__mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
(__v8df)
_mm512_setzero_pd (),
(__mmask8) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_rsqrt14_ps(__m512 __A) _mm512_rsqrt14_ps(__m512 __A)
{ {
return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A, return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
(__v16sf) (__v16sf)
_mm512_setzero_ps (), _mm512_setzero_ps (),
(__mmask16) -1); (__mmask16) -1);
} }
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_rsqrt14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
(__v16sf) __W,
(__mmask16) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_rsqrt14_ps (__mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
(__v16sf)
_mm512_setzero_ps (),
(__mmask16) __U);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rsqrt14_ss(__m128 __A, __m128 __B) _mm_rsqrt14_ss(__m128 __A, __m128 __B)
{ {
return (__m128) __builtin_ia32_rsqrt14ss_mask ((__v4sf) __A, return (__m128) __builtin_ia32_rsqrt14ss_mask ((__v4sf) __A,
(__v4sf) __B, (__v4sf) __B,
(__v4sf) (__v4sf)
_mm_setzero_ps (), _mm_setzero_ps (),
(__mmask8) -1); (__mmask8) -1);
▲ Show 20 Lines • Show All 49 Lines • ▼ Show 20 Lines
_mm512_rcp14_pd(__m512d __A) _mm512_rcp14_pd(__m512d __A)
{ {
return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A, return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
(__v8df) (__v8df)
_mm512_setzero_pd (), _mm512_setzero_pd (),
(__mmask8) -1); (__mmask8) -1);
} }
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_rcp14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
(__v8df) __W,
(__mmask8) __U);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_rcp14_pd (__mmask8 __U, __m512d __A)
{
return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
(__v8df)
_mm512_setzero_pd (),
(__mmask8) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_rcp14_ps(__m512 __A) _mm512_rcp14_ps(__m512 __A)
{ {
return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A, return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
(__v16sf) (__v16sf)
_mm512_setzero_ps (), _mm512_setzero_ps (),
(__mmask16) -1); (__mmask16) -1);
} }
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_rcp14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
(__v16sf) __W,
(__mmask16) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_rcp14_ps (__mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
(__v16sf)
_mm512_setzero_ps (),
(__mmask16) __U);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rcp14_ss(__m128 __A, __m128 __B) _mm_rcp14_ss(__m128 __A, __m128 __B)
{ {
return (__m128) __builtin_ia32_rcp14ss_mask ((__v4sf) __A, return (__m128) __builtin_ia32_rcp14ss_mask ((__v4sf) __A,
(__v4sf) __B, (__v4sf) __B,
(__v4sf) (__v4sf)
_mm_setzero_ps (), _mm_setzero_ps (),
(__mmask8) -1); (__mmask8) -1);
▲ Show 20 Lines • Show All 7,319 Lines • Show Last 20 Lines

test/CodeGen/avx512f-builtins.c

// RUN: %clang_cc1 %s -triple=x86_64-apple-darwin -target-feature +avx512f -emit-llvm -o - -Werror | FileCheck %s // RUN: %clang_cc1 %s -triple=x86_64-apple-darwin -target-feature +avx512f -emit-llvm -o - -Werror | FileCheck %s
// Don't include mm_malloc.h, it's system specific. // Don't include mm_malloc.h, it's system specific.
#define __MM_MALLOC_H #define __MM_MALLOC_H
#include <immintrin.h> #include <immintrin.h>
__m512d test_mm512_sqrt_pd(__m512d a) __m512d test_mm512_sqrt_pd(__m512d a)
{ {
// CHECK-LABEL: @test_mm512_sqrt_pd // CHECK-LABEL: @test_mm512_sqrt_pd
// CHECK: @llvm.x86.avx512.mask.sqrt.pd.512 // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
return _mm512_sqrt_pd(a); return _mm512_sqrt_pd(a);
} }
__m512d test_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_mask_sqrt_pd
// CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
return _mm512_mask_sqrt_pd (__W,__U,__A);
}
__m512d test_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_maskz_sqrt_pd
// CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
return _mm512_maskz_sqrt_pd (__U,__A);
}
__m512 test_mm512_sqrt_ps(__m512 a) __m512 test_mm512_sqrt_ps(__m512 a)
{ {
// CHECK-LABEL: @test_mm512_sqrt_ps // CHECK-LABEL: @test_mm512_sqrt_ps
// CHECK: @llvm.x86.avx512.mask.sqrt.ps.512 // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
return _mm512_sqrt_ps(a); return _mm512_sqrt_ps(a);
} }
__m512 test_mm512_mask_sqrt_ps(_m512 __W, __mmask8 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_mask_sqrt_ps
// CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
return _mm512_mask_sqrt_ps( __W, __U, __A);
}
__m512 test_mm512_maskz_sqrt_ps( __mmask8 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_maskz_sqrt_ps
// CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
return _mm512_maskz_sqrt_ps(__U ,__A);
}
__m512d test_mm512_rsqrt14_pd(__m512d a) __m512d test_mm512_rsqrt14_pd(__m512d a)
{ {
// CHECK-LABEL: @test_mm512_rsqrt14_pd // CHECK-LABEL: @test_mm512_rsqrt14_pd
// CHECK: @llvm.x86.avx512.rsqrt14.pd.512 // CHECK: @llvm.x86.avx512.rsqrt14.pd.512
return _mm512_rsqrt14_pd(a); return _mm512_rsqrt14_pd(a);
} }
__m512d test_mm512_mask_rsqrt14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_mask_rsqrt14_pd
// CHECK: @llvm.x86.avx512.rsqrt14.pd.512
return _mm512_mask_rsqrt14_pd (__W,__U,__A);
}
__m512d test_mm512_maskz_rsqrt14_pd (__mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_maskz_rsqrt14_pd
// CHECK: @llvm.x86.avx512.rsqrt14.pd.512
return _mm512_maskz_rsqrt14_pd (__U,__A);
}
__m512 test_mm512_rsqrt14_ps(__m512 a) __m512 test_mm512_rsqrt14_ps(__m512 a)
{ {
// CHECK-LABEL: @test_mm512_rsqrt14_ps // CHECK-LABEL: @test_mm512_rsqrt14_ps
// CHECK: @llvm.x86.avx512.rsqrt14.ps.512 // CHECK: @llvm.x86.avx512.rsqrt14.ps.512
return _mm512_rsqrt14_ps(a); return _mm512_rsqrt14_ps(a);
} }
__m512 test_mm512_mask_rsqrt14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_mask_rsqrt14_ps
// CHECK: @llvm.x86.avx512.rsqrt14.ps.512
return _mm512_mask_rsqrt14_ps (__W,__U,__A);
}
__m512 test_mm512_maskz_rsqrt14_ps (__mmask16 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_maskz_rsqrt14_ps
// CHECK: @llvm.x86.avx512.rsqrt14.ps.512
return _mm512_maskz_rsqrt14_ps (__U,__A);
}
__m512 test_mm512_add_ps(__m512 a, __m512 b) __m512 test_mm512_add_ps(__m512 a, __m512 b)
{ {
// CHECK-LABEL: @test_mm512_add_ps // CHECK-LABEL: @test_mm512_add_ps
// CHECK: fadd <16 x float> // CHECK: fadd <16 x float>
return _mm512_add_ps(a, b); return _mm512_add_ps(a, b);
} }
__m512d test_mm512_add_pd(__m512d a, __m512d b) __m512d test_mm512_add_pd(__m512d a, __m512d b)
▲ Show 20 Lines • Show All 4,601 Lines • ▼ Show 20 Lines
} }
__m128 test_mm_maskz_rsqrt14_ss(__mmask8 __U, __m128 __A, __m128 __B){ __m128 test_mm_maskz_rsqrt14_ss(__mmask8 __U, __m128 __A, __m128 __B){
// CHECK-LABEL: @test_mm_maskz_rsqrt14_ss // CHECK-LABEL: @test_mm_maskz_rsqrt14_ss
// CHECK: @llvm.x86.avx512.rsqrt14.ss // CHECK: @llvm.x86.avx512.rsqrt14.ss
return _mm_maskz_rsqrt14_ss(__U, __A, __B); return _mm_maskz_rsqrt14_ss(__U, __A, __B);
} }
__m512d test_mm512_mask_rcp14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_mask_rcp14_pd
// CHECK: @llvm.x86.avx512.rcp14.pd.512
return _mm512_mask_rcp14_pd (__W,__U,__A);
}
__m512d test_mm512_maskz_rcp14_pd (__mmask8 __U, __m512d __A)
{
// CHECK-LABEL: @test_mm512_maskz_rcp14_pd
// CHECK: @llvm.x86.avx512.rcp14.pd.512
return _mm512_maskz_rcp14_pd (__U,__A);
}
__m512 test_mm512_mask_rcp14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_mask_rcp14_ps
// CHECK: @llvm.x86.avx512.rcp14.ps.512
return _mm512_mask_rcp14_ps (__W,__U,__A);
}
__m512 test_mm512_maskz_rcp14_ps (__mmask16 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_maskz_rcp14_ps
// CHECK: @llvm.x86.avx512.rcp14.ps.512
return _mm512_maskz_rcp14_ps (__U,__A);
}
__m128d test_mm_mask_rcp14_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){ __m128d test_mm_mask_rcp14_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B){
// CHECK-LABEL: @test_mm_mask_rcp14_sd // CHECK-LABEL: @test_mm_mask_rcp14_sd
// CHECK: @llvm.x86.avx512.rcp14.sd // CHECK: @llvm.x86.avx512.rcp14.sd
return _mm_mask_rcp14_sd(__W, __U, __A, __B); return _mm_mask_rcp14_sd(__W, __U, __A, __B);
} }
__m128d test_mm_maskz_rcp14_sd(__mmask8 __U, __m128d __A, __m128d __B){ __m128d test_mm_maskz_rcp14_sd(__mmask8 __U, __m128d __A, __m128d __B){
// CHECK-LABEL: @test_mm_maskz_rcp14_sd // CHECK-LABEL: @test_mm_maskz_rcp14_sd
▲ Show 20 Lines • Show All 1,366 Lines • Show Last 20 Lines