This is an archive of the discontinued LLVM Phabricator instance.

Differential D20096

[Clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm] instruction set.
ClosedPublic

Authored by m_zuckerman on May 10 2016, 4:04 AM.

Details

Reviewers
AsafBadouh
delena
igorb
Commits
rG5e2c6b6200e1: [clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm]…
rC269170: [clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm]…
rL269170: [clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm]…

Diff Detail

Repository
rL LLVM

Event Timeline

m_zuckerman updated this revision to Diff 56688.May 10 2016, 4:04 AM
m_zuckerman retitled this revision from to [Clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm] instruction set..
m_zuckerman updated this object.
m_zuckerman added reviewers: AsafBadouh, igorb, delena.
m_zuckerman added a subscriber: llvm-commits.
AsafBadouh added inline comments.May 10 2016, 5:24 AM
lib/Headers/avx512fintrin.h
2664 ↗(On Diff #56688)

test is missing

5616 ↗(On Diff #56688)

test is missing

5647 ↗(On Diff #56688)

test is missing

m_zuckerman updated this revision to Diff 56710.May 10 2016, 6:57 AM
m_zuckerman marked 3 inline comments as done.
AsafBadouh accepted this revision.May 10 2016, 7:14 AM
AsafBadouh edited edge metadata.

LGTM

This revision is now accepted and ready to land.May 10 2016, 7:14 AM
Closed by commit rL269170: [clang][AVX512] completing missing intrinsics for [vpermt2d|vptestm]… (authored by mzuckerm). · Explain WhyMay 11 2016, 4:27 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
cfe/
trunk/
lib/
Headers/
141 lines
test/
CodeGen/
56 lines

Diff 56878

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 2,824 Lines • ▼ Show 20 Lines
_mm512_permutex2var_epi32(__m512i __A, __m512i __I, __m512i __B) _mm512_permutex2var_epi32(__m512i __A, __m512i __I, __m512i __B)
{ {
return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I
/* idx */ , /* idx */ ,
(__v16si) __A, (__v16si) __A,
(__v16si) __B, (__v16si) __B,
(__mmask16) -1); (__mmask16) -1);
} }
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_permutex2var_epi32 (__m512i __A, __mmask16 __U,
__m512i __I, __m512i __B)
{
return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I
/* idx */ ,
(__v16si) __A,
(__v16si) __B,
(__mmask16) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_epi32 (__mmask16 __U, __m512i __A,
__m512i __I, __m512i __B)
{
return (__m512i) __builtin_ia32_vpermt2vard512_maskz ((__v16si) __I
/* idx */ ,
(__v16si) __A,
(__v16si) __B,
(__mmask16) __U);
}
static __inline __m512i __DEFAULT_FN_ATTRS static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_permutex2var_epi64(__m512i __A, __m512i __I, __m512i __B) _mm512_permutex2var_epi64(__m512i __A, __m512i __I, __m512i __B)
{ {
return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I
/* idx */ , /* idx */ ,
(__v8di) __A, (__v8di) __A,
(__v8di) __B, (__v8di) __B,
(__mmask8) -1); (__mmask8) -1);
} }
static __inline __m512d __DEFAULT_FN_ATTRS static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_permutex2var_pd(__m512d __A, __m512i __I, __m512d __B) _mm512_mask_permutex2var_epi64 (__m512i __A, __mmask8 __U, __m512i __I,
__m512i __B)
{ {
return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I
/* idx */ , /* idx */ ,
(__v8df) __A, (__v8di) __A,
(__v8df) __B, (__v8di) __B,
(__mmask8) -1); (__mmask8) __U);
} }
static __inline __m512 __DEFAULT_FN_ATTRS
_mm512_permutex2var_ps(__m512 __A, __m512i __I, __m512 __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_epi64 (__mmask8 __U, __m512i __A,
__m512i __I, __m512i __B)
{ {
return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I return (__m512i) __builtin_ia32_vpermt2varq512_maskz ((__v8di) __I
/* idx */ , /* idx */ ,
(__v16sf) __A, (__v8di) __A,
(__v16sf) __B, (__v8di) __B,
(__mmask16) -1); (__mmask8) __U);
} }
#define _mm512_alignr_epi64(A, B, I) __extension__ ({ \ #define _mm512_alignr_epi64(A, B, I) __extension__ ({ \
(__m512i)__builtin_ia32_alignq512_mask((__v8di)(__m512i)(A), \ (__m512i)__builtin_ia32_alignq512_mask((__v8di)(__m512i)(A), \
(__v8di)(__m512i)(B), \ (__v8di)(__m512i)(B), \
(I), (__v8di)_mm512_setzero_si512(), \ (I), (__v8di)_mm512_setzero_si512(), \
(__mmask8)-1); }) (__mmask8)-1); })
▲ Show 20 Lines • Show All 593 Lines • ▼ Show 20 Lines
static __inline __mmask16 __DEFAULT_FN_ATTRS static __inline __mmask16 __DEFAULT_FN_ATTRS
_mm512_test_epi32_mask(__m512i __A, __m512i __B) _mm512_test_epi32_mask(__m512i __A, __m512i __B)
{ {
return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A, return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A,
(__v16si) __B, (__v16si) __B,
(__mmask16) -1); (__mmask16) -1);
} }
static __inline__ __mmask16 __DEFAULT_FN_ATTRS
_mm512_mask_test_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A,
(__v16si) __B, __U);
}
static __inline __mmask8 __DEFAULT_FN_ATTRS static __inline __mmask8 __DEFAULT_FN_ATTRS
_mm512_test_epi64_mask(__m512i __A, __m512i __B) _mm512_test_epi64_mask(__m512i __A, __m512i __B)
{ {
return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A, return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A,
(__v8di) __B, (__v8di) __B,
(__mmask8) -1); (__mmask8) -1);
} }
static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm512_mask_test_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A, (__v8di) __B, __U);
}
/* SIMD load ops */ /* SIMD load ops */
static __inline __m512i __DEFAULT_FN_ATTRS static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_loadu_si512 (void const *__P) _mm512_loadu_si512 (void const *__P)
{ {
return (__m512i) __builtin_ia32_loaddqusi512_mask ((const __v16si *) __P, return (__m512i) __builtin_ia32_loaddqusi512_mask ((const __v16si *) __P,
(__v16si) (__v16si)
_mm512_setzero_si512 (), _mm512_setzero_si512 (),
▲ Show 20 Lines • Show All 2,413 Lines • ▼ Show 20 Lines
{ {
return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A, return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
(__v16si) __C, (__v16si) __C,
(__v16sf) (__v16sf)
_mm512_setzero_ps (), _mm512_setzero_ps (),
(__mmask16) __U); (__mmask16) __U);
} }
static __inline__ __m512i __DEFAULT_FN_ATTRS static __inline __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_epi32 (__mmask16 __U, __m512i __A, _mm512_permutex2var_pd(__m512d __A, __m512i __I, __m512d __B)
__m512i __I, __m512i __B)
{ {
return (__m512i) __builtin_ia32_vpermt2vard512_maskz ((__v16si) __I return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I
/* idx */ , /* idx */ ,
(__v16si) __A, (__v8df) __A,
(__v16si) __B, (__v8df) __B,
(__mmask16) __U); (__mmask8) -1);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_permutex2var_pd (__m512d __A, __mmask8 __U, __m512i __I, __m512d __B)
{
return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I
/* idx */ ,
(__v8df) __A,
(__v8df) __B,
(__mmask8) __U);
} }
static __inline__ __m512d __DEFAULT_FN_ATTRS static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_pd (__mmask8 __U, __m512d __A, __m512i __I, _mm512_maskz_permutex2var_pd (__mmask8 __U, __m512d __A, __m512i __I,
__m512d __B) __m512d __B)
{ {
return (__m512d) __builtin_ia32_vpermt2varpd512_maskz ((__v8di) __I return (__m512d) __builtin_ia32_vpermt2varpd512_maskz ((__v8di) __I
/* idx */ , /* idx */ ,
(__v8df) __A, (__v8df) __A,
(__v8df) __B, (__v8df) __B,
(__mmask8) __U); (__mmask8) __U);
} }
static __inline __m512 __DEFAULT_FN_ATTRS
_mm512_permutex2var_ps(__m512 __A, __m512i __I, __m512 __B)
{
return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I
/* idx */ ,
(__v16sf) __A,
(__v16sf) __B,
(__mmask16) -1);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_ps (__mmask16 __U, __m512 __A, __m512i __I, _mm512_mask_permutex2var_ps (__m512 __A, __mmask16 __U, __m512i __I, __m512 __B)
__m512 __B)
{ {
return (__m512) __builtin_ia32_vpermt2varps512_maskz ((__v16si) __I return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I
/* idx */ , /* idx */ ,
(__v16sf) __A, (__v16sf) __A,
(__v16sf) __B, (__v16sf) __B,
(__mmask16) __U); (__mmask16) __U);
} }
static __inline__ __m512i __DEFAULT_FN_ATTRS static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_permutex2var_epi64 (__mmask8 __U, __m512i __A, _mm512_maskz_permutex2var_ps (__mmask16 __U, __m512 __A, __m512i __I,
__m512i __I, __m512i __B) __m512 __B)
{ {
return (__m512i) __builtin_ia32_vpermt2varq512_maskz ((__v8di) __I return (__m512) __builtin_ia32_vpermt2varps512_maskz ((__v16si) __I
/* idx */ , /* idx */ ,
(__v8di) __A, (__v16sf) __A,
(__v8di) __B, (__v16sf) __B,
(__mmask8) __U); (__mmask16) __U);
} }
static __inline__ __mmask16 __DEFAULT_FN_ATTRS static __inline__ __mmask16 __DEFAULT_FN_ATTRS
_mm512_testn_epi32_mask (__m512i __A, __m512i __B) _mm512_testn_epi32_mask (__m512i __A, __m512i __B)
{ {
return (__mmask16) __builtin_ia32_ptestnmd512 ((__v16si) __A, return (__mmask16) __builtin_ia32_ptestnmd512 ((__v16si) __A,
(__v16si) __B, (__v16si) __B,
(__mmask16) -1); (__mmask16) -1);
▲ Show 20 Lines • Show All 2,950 Lines • Show Last 20 Lines

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

Show First 20 Lines • Show All 3,103 Lines • ▼ Show 20 Lines
} }
__m512i test_mm512_maskz_permutex2var_epi32(__mmask16 __U, __m512i __A, __m512i __I, __m512i __B) { __m512i test_mm512_maskz_permutex2var_epi32(__mmask16 __U, __m512i __A, __m512i __I, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_permutex2var_epi32 // CHECK-LABEL: @test_mm512_maskz_permutex2var_epi32
// CHECK: @llvm.x86.avx512.maskz.vpermt2var.d.512 // CHECK: @llvm.x86.avx512.maskz.vpermt2var.d.512
return _mm512_maskz_permutex2var_epi32(__U, __A, __I, __B); return _mm512_maskz_permutex2var_epi32(__U, __A, __I, __B);
} }
__m512i test_mm512_mask_permutex2var_epi32 (__m512i __A, __mmask16 __U, __m512i __I, __m512i __B)
{
// CHECK-LABEL: @test_mm512_mask_permutex2var_epi32
// CHECK: @llvm.x86.avx512.mask.vpermt2var.d.512
return _mm512_mask_permutex2var_epi32 (__A,__U,__I,__B);
}
__m512d test_mm512_permutex2var_pd (__m512d __A, __m512i __I, __m512d __B)
{
// CHECK-LABEL: @test_mm512_permutex2var_pd
// CHECK: @llvm.x86.avx512.mask.vpermt2var.pd.512
return _mm512_permutex2var_pd (__A, __I,__B);
}
__m512d test_mm512_mask_permutex2var_pd (__m512d __A, __mmask8 __U, __m512i __I, __m512d __B)
{
// CHECK-LABEL: @test_mm512_mask_permutex2var_pd
// CHECK: @llvm.x86.avx512.mask.vpermt2var.pd.512
return _mm512_mask_permutex2var_pd (__A,__U,__I,__B);
}
__m512d test_mm512_maskz_permutex2var_pd(__mmask8 __U, __m512d __A, __m512i __I, __m512d __B) { __m512d test_mm512_maskz_permutex2var_pd(__mmask8 __U, __m512d __A, __m512i __I, __m512d __B) {
// CHECK-LABEL: @test_mm512_maskz_permutex2var_pd // CHECK-LABEL: @test_mm512_maskz_permutex2var_pd
// CHECK: @llvm.x86.avx512.maskz.vpermt2var.pd.512 // CHECK: @llvm.x86.avx512.maskz.vpermt2var.pd.512
return _mm512_maskz_permutex2var_pd(__U, __A, __I, __B); return _mm512_maskz_permutex2var_pd(__U, __A, __I, __B);
} }
__m512 test_mm512_permutex2var_ps (__m512 __A, __m512i __I, __m512 __B)
{
// CHECK-LABEL: @test_mm512_permutex2var_ps
// CHECK: @llvm.x86.avx512.mask.vpermt2var.ps.512
return _mm512_permutex2var_ps (__A, __I, __B);
}
__m512 test_mm512_mask_permutex2var_ps (__m512 __A, __mmask16 __U, __m512i __I, __m512 __B)
{
// CHECK-LABEL: @test_mm512_mask_permutex2var_ps
// CHECK: @llvm.x86.avx512.mask.vpermt2var.ps.512
return _mm512_mask_permutex2var_ps (__A,__U,__I,__B);
}
__m512 test_mm512_maskz_permutex2var_ps(__mmask16 __U, __m512 __A, __m512i __I, __m512 __B) { __m512 test_mm512_maskz_permutex2var_ps(__mmask16 __U, __m512 __A, __m512i __I, __m512 __B) {
// CHECK-LABEL: @test_mm512_maskz_permutex2var_ps // CHECK-LABEL: @test_mm512_maskz_permutex2var_ps
// CHECK: @llvm.x86.avx512.maskz.vpermt2var.ps.512 // CHECK: @llvm.x86.avx512.maskz.vpermt2var.ps.512
return _mm512_maskz_permutex2var_ps(__U, __A, __I, __B); return _mm512_maskz_permutex2var_ps(__U, __A, __I, __B);
} }
__m512i test_mm512_mask_permutex2var_epi64 (__m512i __A, __mmask8 __U, __m512i __I, __m512i __B){
// CHECK-LABEL: @test_mm512_mask_permutex2var_epi64
// CHECK: @llvm.x86.avx512.mask.vpermt2var.q.512
return _mm512_mask_permutex2var_epi64(__A, __U, __I, __B);
}
__m512i test_mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i __A, __m512i __I, __m512i __B) { __m512i test_mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i __A, __m512i __I, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_permutex2var_epi64 // CHECK-LABEL: @test_mm512_maskz_permutex2var_epi64
// CHECK: @llvm.x86.avx512.maskz.vpermt2var.q.512 // CHECK: @llvm.x86.avx512.maskz.vpermt2var.q.512
return _mm512_maskz_permutex2var_epi64(__U, __A, __I, __B); return _mm512_maskz_permutex2var_epi64(__U, __A, __I, __B);
} }
__mmask16 test_mm512_testn_epi32_mask(__m512i __A, __m512i __B) { __mmask16 test_mm512_testn_epi32_mask(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_testn_epi32_mask // CHECK-LABEL: @test_mm512_testn_epi32_mask
// CHECK: @llvm.x86.avx512.ptestnm.d.512 // CHECK: @llvm.x86.avx512.ptestnm.d.512
Show All 12 Lines__mmask8 test_mm512_testn_epi64_mask(__m512i __A, __m512i __B) {
return _mm512_testn_epi64_mask(__A, __B); return _mm512_testn_epi64_mask(__A, __B);
} }
__mmask8 test_mm512_mask_testn_epi64_mask(__mmask8 __U, __m512i __A, __m512i __B) { __mmask8 test_mm512_mask_testn_epi64_mask(__mmask8 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_testn_epi64_mask // CHECK-LABEL: @test_mm512_mask_testn_epi64_mask
// CHECK: @llvm.x86.avx512.ptestnm.q.512 // CHECK: @llvm.x86.avx512.ptestnm.q.512
return _mm512_mask_testn_epi64_mask(__U, __A, __B); return _mm512_mask_testn_epi64_mask(__U, __A, __B);
} }
__mmask16 test_mm512_mask_test_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
// CHECK-LABEL: @test_mm512_mask_test_epi32_mask
// CHECK: @llvm.x86.avx512.ptestm.d.512
return _mm512_mask_test_epi32_mask (__U,__A,__B);
}
__mmask8 test_mm512_mask_test_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
// CHECK-LABEL: @test_mm512_mask_test_epi64_mask
// CHECK: @llvm.x86.avx512.ptestm.q.512
return _mm512_mask_test_epi64_mask (__U,__A,__B);
}
__m512i test_mm512_maskz_unpackhi_epi32(__mmask16 __U, __m512i __A, __m512i __B) { __m512i test_mm512_maskz_unpackhi_epi32(__mmask16 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_unpackhi_epi32 // CHECK-LABEL: @test_mm512_maskz_unpackhi_epi32
// CHECK: @llvm.x86.avx512.mask.punpckhd.q.512 // CHECK: @llvm.x86.avx512.mask.punpckhd.q.512
return _mm512_maskz_unpackhi_epi32(__U, __A, __B); return _mm512_maskz_unpackhi_epi32(__U, __A, __B);
} }
__m512i test_mm512_unpackhi_epi64(__m512i __A, __m512i __B) { __m512i test_mm512_unpackhi_epi64(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_unpackhi_epi64 // CHECK-LABEL: @test_mm512_unpackhi_epi64
▲ Show 20 Lines • Show All 3,169 Lines • Show Last 20 Lines