This is an archive of the discontinued LLVM Phabricator instance.

Differential D11564

[X86][AVX512bw] add convert i16 to i8 and unpack intrinsics
ClosedPublic

Authored by AsafBadouh on Jul 28 2015, 8:46 AM.

Details

Reviewers
delena
gbenyei
mkuper
igorb
Commits
rG1998eb207762: [X86][AVX512BW] add convert i16 to i8 and unpack intrinsics
rC243514: [X86][AVX512BW] add convert i16 to i8 and unpack intrinsics
rL243514: [X86][AVX512BW] add convert i16 to i8 and unpack intrinsics
Summary

tests included

Diff Detail

Repository
rL LLVM

Event Timeline

AsafBadouh updated this revision to Diff 30822.Jul 28 2015, 8:46 AM
AsafBadouh retitled this revision from to [X86][AVX512bw] add convert i16 to i8 and unpack intrinsics.
AsafBadouh updated this object.
AsafBadouh added reviewers: delena, igorb, gbenyei, mkuper.
AsafBadouh added a subscriber: llvm-commits.
mkuper added inline comments.Jul 28 2015, 10:54 AM
lib/Headers/avx512bwintrin.h
1297 ↗(On Diff #30822)

Can these fixes be a separate commit?

1369 ↗(On Diff #30822)

Sometimes the call is on the same line as the cast, sometimes not. Looks odd, style-wise.

1391 ↗(On Diff #30822)

Extra space before the ()

1412 ↗(On Diff #30822)

Same here.

1421 ↗(On Diff #30822)

Same here.

1448 ↗(On Diff #30822)

Same here.
There are more below, can you just find them all and remove the extra spaces?

AsafBadouh updated this revision to Diff 30883.Jul 29 2015, 12:11 AM
AsafBadouh marked an inline comment as done.

Fixed according to Michael Kuperstein comments

mkuper added inline comments.Jul 29 2015, 12:44 AM
lib/Headers/avx512bwintrin.h
1403 ↗(On Diff #30883)

Sorry for the nitpicking, but, for consistency, line-break before the __M?

1484 ↗(On Diff #30883)

Have a tab here, perhaps?

Closed by commit rL243514: [X86][AVX512BW] add convert i16 to i8 and unpack intrinsics (authored by abadouh). · Explain WhyJul 29 2015, 5:34 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
cfe/
trunk/
include/
clang/
Basic/
7 lines
lib/
Headers/
163 lines
test/
CodeGen/
127 lines

Diff 30899

cfe/trunk/include/clang/Basic/BuiltinsX86.def

Show First 20 Lines • Show All 1,404 Lines • ▼ Show 20 Lines
BUILTIN(__builtin_ia32_vpermt2varps128_mask, "V4fV4iV4fV4fUc", "") BUILTIN(__builtin_ia32_vpermt2varps128_mask, "V4fV4iV4fV4fUc", "")
BUILTIN(__builtin_ia32_vpermt2varps128_maskz, "V4fV4iV4fV4fUc", "") BUILTIN(__builtin_ia32_vpermt2varps128_maskz, "V4fV4iV4fV4fUc", "")
BUILTIN(__builtin_ia32_vpermt2varps256_mask, "V8fV8iV8fV8fUc", "") BUILTIN(__builtin_ia32_vpermt2varps256_mask, "V8fV8iV8fV8fUc", "")
BUILTIN(__builtin_ia32_vpermt2varps256_maskz, "V8fV8iV8fV8fUc", "") BUILTIN(__builtin_ia32_vpermt2varps256_maskz, "V8fV8iV8fV8fUc", "")
BUILTIN(__builtin_ia32_vpermt2varq128_mask, "V2LLiV2LLiV2LLiV2LLiUc", "") BUILTIN(__builtin_ia32_vpermt2varq128_mask, "V2LLiV2LLiV2LLiV2LLiUc", "")
BUILTIN(__builtin_ia32_vpermt2varq128_maskz, "V2LLiV2LLiV2LLiV2LLiUc", "") BUILTIN(__builtin_ia32_vpermt2varq128_maskz, "V2LLiV2LLiV2LLiV2LLiUc", "")
BUILTIN(__builtin_ia32_vpermt2varq256_mask, "V4LLiV4LLiV4LLiV4LLiUc", "") BUILTIN(__builtin_ia32_vpermt2varq256_mask, "V4LLiV4LLiV4LLiV4LLiUc", "")
BUILTIN(__builtin_ia32_vpermt2varq256_maskz, "V4LLiV4LLiV4LLiV4LLiUc", "") BUILTIN(__builtin_ia32_vpermt2varq256_maskz, "V4LLiV4LLiV4LLiV4LLiUc", "")
BUILTIN(__builtin_ia32_pmovswb512_mask, "V32cV32sV32cUi", "")
BUILTIN(__builtin_ia32_pmovuswb512_mask, "V32cV32sV32cUi", "")
BUILTIN(__builtin_ia32_pmovwb512_mask, "V32cV32sV32cUi", "")
BUILTIN(__builtin_ia32_punpckhbw512_mask, "V64cV64cV64cV64cULLi", "")
BUILTIN(__builtin_ia32_punpckhwd512_mask, "V32sV32sV32sV32sUi", "")
BUILTIN(__builtin_ia32_punpcklbw512_mask, "V64cV64cV64cV64cULLi", "")
BUILTIN(__builtin_ia32_punpcklwd512_mask, "V32sV32sV32sV32sUi", "")
#undef BUILTIN #undef BUILTIN

cfe/trunk/lib/Headers/avx512bwintrin.h

Show First 20 Lines • Show All 1,342 Lines • ▼ Show 20 Lines
_mm512_maskz_madd_epi16 (__mmask16 __U, __m512i __A, __m512i __B) { _mm512_maskz_madd_epi16 (__mmask16 __U, __m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_pmaddwd512_mask ((__v32hi) __A, return (__m512i) __builtin_ia32_pmaddwd512_mask ((__v32hi) __A,
(__v32hi) __B, (__v32hi) __B,
(__v16si) (__v16si)
_mm512_setzero_si512 (), _mm512_setzero_si512 (),
(__mmask16) __U); (__mmask16) __U);
} }
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_cvtsepi16_epi8 (__m512i __A) {
return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
(__v32qi)_mm256_setzero_si256(),
(__mmask32) -1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_mask_cvtsepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
(__v32qi)__O,
__M);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_maskz_cvtsepi16_epi8 (__mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
(__v32qi) _mm256_setzero_si256(),
__M);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_cvtusepi16_epi8 (__m512i __A) {
return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
(__v32qi) _mm256_setzero_si256(),
(__mmask32) -1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_mask_cvtusepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
(__v32qi) __O,
__M);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_maskz_cvtusepi16_epi8 (__mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
(__v32qi) _mm256_setzero_si256(),
__M);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_cvtepi16_epi8 (__m512i __A) {
return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
(__v32qi) _mm256_setzero_si256(),
(__mmask32) -1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_mask_cvtepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
(__v32qi) __O,
__M);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm512_maskz_cvtepi16_epi8 (__mmask32 __M, __m512i __A) {
return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
(__v32qi) _mm256_setzero_si256(),
__M);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_unpackhi_epi8 (__m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) _mm512_setzero_qi(),
(__mmask64) -1);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_unpackhi_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
__m512i __B) {
return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) __W,
(__mmask64) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_unpackhi_epi8 (__mmask64 __U, __m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) _mm512_setzero_qi(),
(__mmask64) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_unpackhi_epi16 (__m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) _mm512_setzero_hi(),
(__mmask32) -1);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_unpackhi_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
__m512i __B) {
return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) __W,
(__mmask32) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_unpackhi_epi16 (__mmask32 __U, __m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) _mm512_setzero_hi(),
(__mmask32) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_unpacklo_epi8 (__m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) _mm512_setzero_qi(),
(__mmask64) -1);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_unpacklo_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
__m512i __B) {
return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) __W,
(__mmask64) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_unpacklo_epi8 (__mmask64 __U, __m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
(__v64qi) __B,
(__v64qi) _mm512_setzero_qi(),
(__mmask64) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_unpacklo_epi16 (__m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) _mm512_setzero_hi(),
(__mmask32) -1);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_unpacklo_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
__m512i __B) {
return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) __W,
(__mmask32) __U);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_unpacklo_epi16 (__mmask32 __U, __m512i __A, __m512i __B) {
return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) _mm512_setzero_hi(),
(__mmask32) __U);
}
#define _mm512_cmp_epi8_mask(a, b, p) __extension__ ({ \ #define _mm512_cmp_epi8_mask(a, b, p) __extension__ ({ \
(__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \ (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
(__v64qi)(__m512i)(b), \ (__v64qi)(__m512i)(b), \
(p), (__mmask64)-1); }) (p), (__mmask64)-1); })
#define _mm512_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \ #define _mm512_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \
(__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \ (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
(__v64qi)(__m512i)(b), \ (__v64qi)(__m512i)(b), \
Show All 36 Lines

cfe/trunk/test/CodeGen/avx512bw-builtins.c

Show First 20 Lines • Show All 904 Lines • ▼ Show 20 Lines__m512i test_mm512_mask_madd_epi16(__m512i __W, __mmask16 __U, __m512i __A, __m512i __B) {
// CHECK: @llvm.x86.avx512.mask.pmaddw.d.512 // CHECK: @llvm.x86.avx512.mask.pmaddw.d.512
return _mm512_mask_madd_epi16(__W,__U,__A,__B); return _mm512_mask_madd_epi16(__W,__U,__A,__B);
} }
__m512i test_mm512_maskz_madd_epi16(__mmask16 __U, __m512i __A, __m512i __B) { __m512i test_mm512_maskz_madd_epi16(__mmask16 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_madd_epi16 // CHECK-LABEL: @test_mm512_maskz_madd_epi16
// CHECK: @llvm.x86.avx512.mask.pmaddw.d.512 // CHECK: @llvm.x86.avx512.mask.pmaddw.d.512
return _mm512_maskz_madd_epi16(__U,__A,__B); return _mm512_maskz_madd_epi16(__U,__A,__B);
} }
__m256i test_mm512_cvtsepi16_epi8(__m512i __A) {
// CHECK-LABEL: @test_mm512_cvtsepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovs.wb.512
return _mm512_cvtsepi16_epi8(__A);
}
__m256i test_mm512_mask_cvtsepi16_epi8(__m256i __O, __mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_cvtsepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovs.wb.512
return _mm512_mask_cvtsepi16_epi8(__O, __M, __A);
}
__m256i test_mm512_maskz_cvtsepi16_epi8(__mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_cvtsepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovs.wb.512
return _mm512_maskz_cvtsepi16_epi8(__M, __A);
}
__m256i test_mm512_cvtusepi16_epi8(__m512i __A) {
// CHECK-LABEL: @test_mm512_cvtusepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovus.wb.512
return _mm512_cvtusepi16_epi8(__A);
}
__m256i test_mm512_mask_cvtusepi16_epi8(__m256i __O, __mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_cvtusepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovus.wb.512
return _mm512_mask_cvtusepi16_epi8(__O, __M, __A);
}
__m256i test_mm512_maskz_cvtusepi16_epi8(__mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_cvtusepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmovus.wb.512
return _mm512_maskz_cvtusepi16_epi8(__M, __A);
}
__m256i test_mm512_cvtepi16_epi8(__m512i __A) {
// CHECK-LABEL: @test_mm512_cvtepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmov.wb.512
return _mm512_cvtepi16_epi8(__A);
}
__m256i test_mm512_mask_cvtepi16_epi8(__m256i __O, __mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_cvtepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmov.wb.512
return _mm512_mask_cvtepi16_epi8(__O, __M, __A);
}
__m256i test_mm512_maskz_cvtepi16_epi8(__mmask32 __M, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_cvtepi16_epi8
// CHECK: @llvm.x86.avx512.mask.pmov.wb.512
return _mm512_maskz_cvtepi16_epi8(__M, __A);
}
__m512i test_mm512_unpackhi_epi8(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_unpackhi_epi8
// CHECK: @llvm.x86.avx512.mask.punpckhb.w.512
return _mm512_unpackhi_epi8(__A, __B);
}
__m512i test_mm512_mask_unpackhi_epi8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_unpackhi_epi8
// CHECK: @llvm.x86.avx512.mask.punpckhb.w.512
return _mm512_mask_unpackhi_epi8(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_unpackhi_epi8(__mmask64 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_unpackhi_epi8
// CHECK: @llvm.x86.avx512.mask.punpckhb.w.512
return _mm512_maskz_unpackhi_epi8(__U, __A, __B);
}
__m512i test_mm512_unpackhi_epi16(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_unpackhi_epi16
// CHECK: @llvm.x86.avx512.mask.punpckhw.d.512
return _mm512_unpackhi_epi16(__A, __B);
}
__m512i test_mm512_mask_unpackhi_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_unpackhi_epi16
// CHECK: @llvm.x86.avx512.mask.punpckhw.d.512
return _mm512_mask_unpackhi_epi16(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_unpackhi_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_unpackhi_epi16
// CHECK: @llvm.x86.avx512.mask.punpckhw.d.512
return _mm512_maskz_unpackhi_epi16(__U, __A, __B);
}
__m512i test_mm512_unpacklo_epi8(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_unpacklo_epi8
// CHECK: @llvm.x86.avx512.mask.punpcklb.w.512
return _mm512_unpacklo_epi8(__A, __B);
}
__m512i test_mm512_mask_unpacklo_epi8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_unpacklo_epi8
// CHECK: @llvm.x86.avx512.mask.punpcklb.w.512
return _mm512_mask_unpacklo_epi8(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_unpacklo_epi8(__mmask64 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_unpacklo_epi8
// CHECK: @llvm.x86.avx512.mask.punpcklb.w.512
return _mm512_maskz_unpacklo_epi8(__U, __A, __B);
}
__m512i test_mm512_unpacklo_epi16(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_unpacklo_epi16
// CHECK: @llvm.x86.avx512.mask.punpcklw.d.512
return _mm512_unpacklo_epi16(__A, __B);
}
__m512i test_mm512_mask_unpacklo_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_unpacklo_epi16
// CHECK: @llvm.x86.avx512.mask.punpcklw.d.512
return _mm512_mask_unpacklo_epi16(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_unpacklo_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_unpacklo_epi16
// CHECK: @llvm.x86.avx512.mask.punpcklw.d.512
return _mm512_maskz_unpacklo_epi16(__U, __A, __B);
}