This is an archive of the discontinued LLVM Phabricator instance.

Differential D31034

[X86][AVX512][Clang][Intrinsics] Adding missing intrinsics to Clang .
ClosedPublic

Authored by igorb on Mar 16 2017, 8:18 AM.

Details

Reviewers
zvi
guyblank
eladcohen
craig.topper
Commits
rGf050b797aca0: [X86][AVX512][Clang][Intrinsics] Adding missing intrinsics to Clang .
rC298208: [X86][AVX512][Clang][Intrinsics] Adding missing intrinsics to Clang .
rL298208: [X86][AVX512][Clang][Intrinsics] Adding missing intrinsics to Clang .
Summary

Adding missing intrinsics :

_mm512_set_epi16, 
_mm512_set_epi8,
_mm512_permutevar_epi32
_mm512_mask_permutevar_epi32

Diff Detail

Repository
rL LLVM

Event Timeline

igorb created this revision.Mar 16 2017, 8:18 AM
craig.topper added a subscriber: craig.topper.Mar 16 2017, 9:52 AM
craig.topper added inline comments.
lib/Headers/avx512fintrin.h
9672

Isn't the convention to prefix all variable names in intrinsics with 2 leading underscores except when they are macros?

craig.topper added a reviewer: craig.topper.Mar 16 2017, 3:10 PM
igorb updated this revision to Diff 92263.Mar 19 2017, 12:47 AM
igorb marked an inline comment as done.

Thanks for the comments.

craig.topper accepted this revision.Mar 19 2017, 12:59 AM

LGTM

This revision is now accepted and ready to land.Mar 19 2017, 12:59 AM
Closed by commit rL298208: [X86][AVX512][Clang][Intrinsics] Adding missing intrinsics to Clang . (authored by ibreger). · Explain WhyMar 19 2017, 1:39 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Headers/
43 lines
test/
CodeGen/
127 lines

Diff 92263

lib/Headers/avx512fintrin.h

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 8,842 Lines • ▼ Show 20 Lines
_mm512_permutexvar_epi32 (__m512i __X, __m512i __Y) _mm512_permutexvar_epi32 (__m512i __X, __m512i __Y)
{ {
return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y, return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y,
(__v16si) __X, (__v16si) __X,
(__v16si) _mm512_undefined_epi32 (), (__v16si) _mm512_undefined_epi32 (),
(__mmask16) -1); (__mmask16) -1);
} }
#define _mm512_permutevar_epi32 _mm512_permutexvar_epi32
static __inline__ __m512i __DEFAULT_FN_ATTRS static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X, _mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X,
__m512i __Y) __m512i __Y)
{ {
return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y, return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y,
(__v16si) __X, (__v16si) __X,
(__v16si) __W, (__v16si) __W,
__M); __M);
} }
#define _mm512_mask_permutevar_epi32 _mm512_mask_permutexvar_epi32
static __inline__ __mmask16 __DEFAULT_FN_ATTRS static __inline__ __mmask16 __DEFAULT_FN_ATTRS
_mm512_kand (__mmask16 __A, __mmask16 __B) _mm512_kand (__mmask16 __A, __mmask16 __B)
{ {
return (__mmask16) __builtin_ia32_kandhi ((__mmask16) __A, (__mmask16) __B); return (__mmask16) __builtin_ia32_kandhi ((__mmask16) __A, (__mmask16) __B);
} }
static __inline__ __mmask16 __DEFAULT_FN_ATTRS static __inline__ __mmask16 __DEFAULT_FN_ATTRS
_mm512_kandn (__mmask16 __A, __mmask16 __B) _mm512_kandn (__mmask16 __A, __mmask16 __B)
▲ Show 20 Lines • Show All 751 Lines • ▼ Show 20 Lines
static __inline__ __m512i __DEFAULT_FN_ATTRS static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_set1_epi64 (__m512i __O, __mmask8 __M, long long __A) _mm512_mask_set1_epi64 (__m512i __O, __mmask8 __M, long long __A)
{ {
return (__m512i) __builtin_ia32_pbroadcastq512_gpr_mask (__A, (__v8di) __O, return (__m512i) __builtin_ia32_pbroadcastq512_gpr_mask (__A, (__v8di) __O,
__M); __M);
} }
#endif #endif
static __inline __m512i __DEFAULT_FN_ATTRS static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_set_epi8 (char __e63, char __e62, char __e61, char __e60, char __e59,
char __e58, char __e57, char __e56, char __e55, char __e54, char __e53,
char __e52, char __e51, char __e50, char __e49, char __e48, char __e47,
char __e46, char __e45, char __e44, char __e43, char __e42, char __e41,
char __e40, char __e39, char __e38, char __e37, char __e36, char __e35,
char __e34, char __e33, char __e32, char __e31, char __e30, char __e29,
char __e28, char __e27, char __e26, char __e25, char __e24, char __e23,
char __e22, char __e21, char __e20, char __e19, char __e18, char __e17,
char __e16, char __e15, char __e14, char __e13, char __e12, char __e11,
char __e10, char __e9, char __e8, char __e7, char __e6, char __e5,
char __e4, char __e3, char __e2, char __e1, char __e0) {
return __extension__ (__m512i)(__v64qi)
{__e0, __e1, __e2, __e3, __e4, __e5, __e6, __e7,
__e8, __e9, __e10, __e11, __e12, __e13, __e14, __e15,
__e16, __e17, __e18, __e19, __e20, __e21, __e22, __e23,
__e24, __e25, __e26, __e27, __e28, __e29, __e30, __e31,
__e32, __e33, __e34, __e35, __e36, __e37, __e38, __e39,
__e40, __e41, __e42, __e43, __e44, __e45, __e46, __e47,
__e48, __e49, __e50, __e51, __e52, __e53, __e54, __e55,
__e56, __e57, __e58, __e59, __e60, __e61, __e62, __e63};
}
static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_set_epi16(short __e31, short __e30, short __e29, short __e28,
short __e27, short __e26, short __e25, short __e24, short __e23,
short __e22, short __e21, short __e20, short __e19, short __e18,
short __e17, short __e16, short __e15, short __e14, short __e13,
short __e12, short __e11, short __e10, short __e9, short __e8,
short __e7, short __e6, short __e5, short __e4, short __e3,
short __e2, short __e1, short __e0) {
return __extension__ (__m512i)(__v32hi)
{__e0, __e1, __e2, __e3, __e4, __e5, __e6, __e7,
__e8, __e9, __e10, __e11, __e12, __e13, __e14, __e15,
__e16, __e17, __e18, __e19, __e20, __e21, __e22, __e23,
__e24, __e25, __e26, __e27, __e28, __e29, __e30, __e31 };
}
static __inline __m512i __DEFAULT_FN_ATTRS
_mm512_set_epi32 (int __A, int __B, int __C, int __D, _mm512_set_epi32 (int __A, int __B, int __C, int __D,
craig.topperUnsubmitted
Done
ReplyInline Actions

Isn't the convention to prefix all variable names in intrinsics with 2 leading underscores except when they are macros?

craig.topper: Isn't the convention to prefix all variable names in intrinsics with 2 leading underscores…
int __E, int __F, int __G, int __H, int __E, int __F, int __G, int __H,
int __I, int __J, int __K, int __L, int __I, int __J, int __K, int __L,
int __M, int __N, int __O, int __P) int __M, int __N, int __O, int __P)
{ {
return __extension__ (__m512i)(__v16si) return __extension__ (__m512i)(__v16si)
{ __P, __O, __N, __M, __L, __K, __J, __I, { __P, __O, __N, __M, __L, __K, __J, __I,
__H, __G, __F, __E, __D, __C, __B, __A }; __H, __G, __F, __E, __D, __C, __B, __A };
} }
▲ Show 20 Lines • Show All 597 Lines • Show Last 20 Lines

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 7,706 Lines • ▼ Show 20 Lines
__m512i test_mm512_mask_set1_epi32 (__m512i __O, __mmask16 __M, int __A) __m512i test_mm512_mask_set1_epi32 (__m512i __O, __mmask16 __M, int __A)
{ {
//CHECK-LABEL: @test_mm512_mask_set1_epi32 //CHECK-LABEL: @test_mm512_mask_set1_epi32
//CHECK: @llvm.x86.avx512.mask.pbroadcast.d.gpr.512 //CHECK: @llvm.x86.avx512.mask.pbroadcast.d.gpr.512
return _mm512_mask_set1_epi32 ( __O, __M, __A); return _mm512_mask_set1_epi32 ( __O, __M, __A);
} }
__m512i test_mm512_set_epi8(char e63, char e62, char e61, char e60, char e59,
char e58, char e57, char e56, char e55, char e54, char e53, char e52,
char e51, char e50, char e49, char e48, char e47, char e46, char e45,
char e44, char e43, char e42, char e41, char e40, char e39, char e38,
char e37, char e36, char e35, char e34, char e33, char e32, char e31,
char e30, char e29, char e28, char e27, char e26, char e25, char e24,
char e23, char e22, char e21, char e20, char e19, char e18, char e17,
char e16, char e15, char e14, char e13, char e12, char e11, char e10,
char e9, char e8, char e7, char e6, char e5, char e4, char e3, char e2,
char e1, char e0) {
//CHECK-LABEL: @test_mm512_set_epi8
//CHECK: load i8, i8* %e63.addr, align 1
//CHECK: load i8, i8* %e62.addr, align 1
//CHECK: load i8, i8* %e61.addr, align 1
//CHECK: load i8, i8* %e60.addr, align 1
//CHECK: load i8, i8* %e59.addr, align 1
//CHECK: load i8, i8* %e58.addr, align 1
//CHECK: load i8, i8* %e57.addr, align 1
//CHECK: load i8, i8* %e56.addr, align 1
//CHECK: load i8, i8* %e55.addr, align 1
//CHECK: load i8, i8* %e54.addr, align 1
//CHECK: load i8, i8* %e53.addr, align 1
//CHECK: load i8, i8* %e52.addr, align 1
//CHECK: load i8, i8* %e51.addr, align 1
//CHECK: load i8, i8* %e50.addr, align 1
//CHECK: load i8, i8* %e49.addr, align 1
//CHECK: load i8, i8* %e48.addr, align 1
//CHECK: load i8, i8* %e47.addr, align 1
//CHECK: load i8, i8* %e46.addr, align 1
//CHECK: load i8, i8* %e45.addr, align 1
//CHECK: load i8, i8* %e44.addr, align 1
//CHECK: load i8, i8* %e43.addr, align 1
//CHECK: load i8, i8* %e42.addr, align 1
//CHECK: load i8, i8* %e41.addr, align 1
//CHECK: load i8, i8* %e40.addr, align 1
//CHECK: load i8, i8* %e39.addr, align 1
//CHECK: load i8, i8* %e38.addr, align 1
//CHECK: load i8, i8* %e37.addr, align 1
//CHECK: load i8, i8* %e36.addr, align 1
//CHECK: load i8, i8* %e35.addr, align 1
//CHECK: load i8, i8* %e34.addr, align 1
//CHECK: load i8, i8* %e33.addr, align 1
//CHECK: load i8, i8* %e32.addr, align 1
//CHECK: load i8, i8* %e31.addr, align 1
//CHECK: load i8, i8* %e30.addr, align 1
//CHECK: load i8, i8* %e29.addr, align 1
//CHECK: load i8, i8* %e28.addr, align 1
//CHECK: load i8, i8* %e27.addr, align 1
//CHECK: load i8, i8* %e26.addr, align 1
//CHECK: load i8, i8* %e25.addr, align 1
//CHECK: load i8, i8* %e24.addr, align 1
//CHECK: load i8, i8* %e23.addr, align 1
//CHECK: load i8, i8* %e22.addr, align 1
//CHECK: load i8, i8* %e21.addr, align 1
//CHECK: load i8, i8* %e20.addr, align 1
//CHECK: load i8, i8* %e19.addr, align 1
//CHECK: load i8, i8* %e18.addr, align 1
//CHECK: load i8, i8* %e17.addr, align 1
//CHECK: load i8, i8* %e16.addr, align 1
//CHECK: load i8, i8* %e15.addr, align 1
//CHECK: load i8, i8* %e14.addr, align 1
//CHECK: load i8, i8* %e13.addr, align 1
//CHECK: load i8, i8* %e12.addr, align 1
//CHECK: load i8, i8* %e11.addr, align 1
//CHECK: load i8, i8* %e10.addr, align 1
//CHECK: load i8, i8* %e9.addr, align 1
//CHECK: load i8, i8* %e8.addr, align 1
//CHECK: load i8, i8* %e7.addr, align 1
//CHECK: load i8, i8* %e6.addr, align 1
//CHECK: load i8, i8* %e5.addr, align 1
//CHECK: load i8, i8* %e4.addr, align 1
//CHECK: load i8, i8* %e3.addr, align 1
//CHECK: load i8, i8* %e2.addr, align 1
//CHECK: load i8, i8* %e1.addr, align 1
//CHECK: load i8, i8* %e0.addr, align 1
return _mm512_set_epi8(e63, e62, e61, e60, e59, e58, e57, e56, e55, e54,
e53, e52, e51, e50, e49, e48,e47, e46, e45, e44, e43, e42, e41, e40,
e39, e38, e37, e36, e35, e34, e33, e32,e31, e30, e29, e28, e27, e26,
e25, e24, e23, e22, e21, e20, e19, e18, e17, e16, e15, e14, e13, e12,
e11, e10, e9, e8, e7, e6, e5, e4, e3, e2, e1, e0);
}
__m512i test_mm512_set_epi16(short e31, short e30, short e29, short e28,
short e27, short e26, short e25, short e24, short e23, short e22,
short e21, short e20, short e19, short e18, short e17,
short e16, short e15, short e14, short e13, short e12,
short e11, short e10, short e9, short e8, short e7,
short e6, short e5, short e4, short e3, short e2, short e1, short e0) {
//CHECK-LABEL: @test_mm512_set_epi16
//CHECK: insertelement{{.*}}i32 0
//CHECK: insertelement{{.*}}i32 1
//CHECK: insertelement{{.*}}i32 2
//CHECK: insertelement{{.*}}i32 3
//CHECK: insertelement{{.*}}i32 4
//CHECK: insertelement{{.*}}i32 5
//CHECK: insertelement{{.*}}i32 6
//CHECK: insertelement{{.*}}i32 7
//CHECK: insertelement{{.*}}i32 8
//CHECK: insertelement{{.*}}i32 9
//CHECK: insertelement{{.*}}i32 10
//CHECK: insertelement{{.*}}i32 11
//CHECK: insertelement{{.*}}i32 12
//CHECK: insertelement{{.*}}i32 13
//CHECK: insertelement{{.*}}i32 14
//CHECK: insertelement{{.*}}i32 15
//CHECK: insertelement{{.*}}i32 16
//CHECK: insertelement{{.*}}i32 17
//CHECK: insertelement{{.*}}i32 18
//CHECK: insertelement{{.*}}i32 19
//CHECK: insertelement{{.*}}i32 20
//CHECK: insertelement{{.*}}i32 21
//CHECK: insertelement{{.*}}i32 22
//CHECK: insertelement{{.*}}i32 23
//CHECK: insertelement{{.*}}i32 24
//CHECK: insertelement{{.*}}i32 25
//CHECK: insertelement{{.*}}i32 26
//CHECK: insertelement{{.*}}i32 27
//CHECK: insertelement{{.*}}i32 28
//CHECK: insertelement{{.*}}i32 29
//CHECK: insertelement{{.*}}i32 30
//CHECK: insertelement{{.*}}i32 31
return _mm512_set_epi16(e31, e30, e29, e28, e27, e26, e25, e24, e23, e22,
e21, e20, e19, e18, e17, e16, e15, e14, e13, e12, e11, e10, e9, e8, e7,
e6, e5, e4, e3, e2, e1, e0);
}
__m512i test_mm512_set_epi32 (int __A, int __B, int __C, int __D, __m512i test_mm512_set_epi32 (int __A, int __B, int __C, int __D,
int __E, int __F, int __G, int __H, int __E, int __F, int __G, int __H,
int __I, int __J, int __K, int __L, int __I, int __J, int __K, int __L,
int __M, int __N, int __O, int __P) int __M, int __N, int __O, int __P)
{ {
//CHECK-LABEL: @test_mm512_set_epi32 //CHECK-LABEL: @test_mm512_set_epi32
//CHECK: insertelement{{.*}}i32 0 //CHECK: insertelement{{.*}}i32 0
//CHECK: insertelement{{.*}}i32 1 //CHECK: insertelement{{.*}}i32 1
▲ Show 20 Lines • Show All 528 Lines • Show Last 20 Lines