r286714 - [AVX-512] Convert the rest of the masked shift by immediate and by single element builtins over to the newly added unmasked builtins and a select.
Craig Topper via cfe-commits
cfe-commits at lists.llvm.org
Fri Nov 11 23:17:00 PST 2016
Author: ctopper
Date: Sat Nov 12 01:16:59 2016
New Revision: 286714
URL: http://llvm.org/viewvc/llvm-project?rev=286714&view=rev
Log:
[AVX-512] Convert the rest of the masked shift by immediate and by single element builtins over to the newly added unmasked builtins and a select.
This should also fix PR30691 since the new builtins are handled like the legacy builtins in the backend.
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/Headers/avx512bwintrin.h
cfe/trunk/lib/Headers/avx512fintrin.h
cfe/trunk/lib/Headers/avx512vlintrin.h
cfe/trunk/lib/Sema/SemaChecking.cpp
cfe/trunk/test/CodeGen/avx512bw-builtins.c
cfe/trunk/test/CodeGen/avx512f-builtins.c
cfe/trunk/test/CodeGen/avx512vl-builtins.c
Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Sat Nov 12 01:16:59 2016
@@ -1359,26 +1359,26 @@ TARGET_BUILTIN(__builtin_ia32_prorvd256_
TARGET_BUILTIN(__builtin_ia32_prorvq128_mask, "V2LLiV2LLiV2LLiV2LLiUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_prorvq256_mask, "V4LLiV4LLiV4LLiV4LLiUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_psllv32hi_mask, "V32sV32sV32sV32sUi","","avx512bw")
-TARGET_BUILTIN(__builtin_ia32_psllw512_mask, "V32sV32sV8sV32sUi","","avx512bw")
-TARGET_BUILTIN(__builtin_ia32_psllwi512_mask, "V32sV32sIiV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psllw512, "V32sV32sV8s","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psllwi512, "V32sV32si","","avx512bw")
TARGET_BUILTIN(__builtin_ia32_psllv16hi_mask, "V16sV16sV16sV16sUs","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psllv8hi_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_pslldi512_mask, "V16iV16iIiV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psllqi512_mask, "V8LLiV8LLiIiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_pslldi512, "V16iV16ii","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psllqi512, "V8LLiV8LLii","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrlv32hi_mask, "V32sV32sV32sV32sUi","","avx512bw")
TARGET_BUILTIN(__builtin_ia32_psrlv16hi_mask, "V16sV16sV16sV16sUs","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psrlv8hi_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_psrldi512_mask, "V16iV16iIiV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psrlqi512_mask, "V8LLiV8LLiIiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psrldi512, "V16iV16ii","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psrlqi512, "V8LLiV8LLii","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrav32hi_mask, "V32sV32sV32sV32sUi","","avx512bw")
TARGET_BUILTIN(__builtin_ia32_psrav16hi_mask, "V16sV16sV16sV16sUs","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psrav8hi_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psravq128_mask, "V2LLiV2LLiV2LLiV2LLiUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_psravq256_mask, "V4LLiV4LLiV4LLiV4LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_psraw512_mask, "V32sV32sV8sV32sUi","","avx512bw")
-TARGET_BUILTIN(__builtin_ia32_psrawi512_mask, "V32sV32sIiV32sUi","","avx512bw")
-TARGET_BUILTIN(__builtin_ia32_psrlw512_mask, "V32sV32sV8sV32sUi","","avx512bw")
-TARGET_BUILTIN(__builtin_ia32_psrlwi512_mask, "V32sV32sIiV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psraw512, "V32sV32sV8s","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psrawi512, "V32sV32si","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psrlw512, "V32sV32sV8s","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psrlwi512, "V32sV32si","","avx512bw")
TARGET_BUILTIN(__builtin_ia32_movdqa32load128_mask, "V4iV4i*V4iUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_movdqa32load256_mask, "V8iV8i*V8iUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_movdqa32load512_mask, "V16iV16iC*V16iUs","","avx512f")
@@ -1539,22 +1539,22 @@ TARGET_BUILTIN(__builtin_ia32_scalefpd51
TARGET_BUILTIN(__builtin_ia32_scalefps512_mask, "V16fV16fV16fV16fUsIi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_scalefsd_round_mask, "V2dV2dV2dV2dUcIi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_scalefss_round_mask, "V4fV4fV4fV4fUcIi","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psradi512_mask, "V16iV16iIiV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psraqi512_mask, "V8LLiV8LLiIiV8LLiUc","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psraq128_mask, "V2LLiV2LLiV2LLiV2LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_psraq256_mask, "V4LLiV4LLiV2LLiV4LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_psraqi128_mask, "V2LLiV2LLiIiV2LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_psraqi256_mask, "V4LLiV4LLiIiV4LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_pslld512_mask, "V16iV16iV4iV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psllq512_mask, "V8LLiV8LLiV2LLiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psradi512, "V16iV16ii","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psraqi512, "V8LLiV8LLii","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psraq128, "V2LLiV2LLiV2LLi","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psraq256, "V4LLiV4LLiV2LLi","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psraqi128, "V2LLiV2LLii","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psraqi256, "V4LLiV4LLii","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pslld512, "V16iV16iV4i","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psllq512, "V8LLiV8LLiV2LLi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psllv16si_mask, "V16iV16iV16iV16iUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psllv8di_mask, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psrad512_mask, "V16iV16iV4iV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psraq512_mask, "V8LLiV8LLiV2LLiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psrad512, "V16iV16iV4i","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psraq512, "V8LLiV8LLiV2LLi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrav16si_mask, "V16iV16iV16iV16iUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrav8di_mask, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psrld512_mask, "V16iV16iV4iV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_psrlq512_mask, "V8LLiV8LLiV2LLiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psrld512, "V16iV16iV4i","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_psrlq512, "V8LLiV8LLiV2LLi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrlv16si_mask, "V16iV16iV16iV16iUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_psrlv8di_mask, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_pternlogd512_mask, "V16iV16iV16iV16iIiUs","","avx512f")
Modified: cfe/trunk/lib/Headers/avx512bwintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512bwintrin.h?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512bwintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512bwintrin.h Sat Nov 12 01:16:59 2016
@@ -1718,49 +1718,48 @@ _mm512_maskz_sllv_epi16 (__mmask32 __U,
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sll_epi16 (__m512i __A, __m128i __B)
+_mm512_sll_epi16(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) -1);
+ return (__m512i)__builtin_ia32_psllw512((__v32hi) __A, (__v8hi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sll_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
- __m128i __B)
+_mm512_mask_sll_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi) __W,
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_sll_epi16(__A, __B), \
+ (__v32hi)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sll_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sll_epi16(__mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_sll_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
}
-#define _mm512_slli_epi16(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllwi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)-1); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_slli_epi16(__m512i __A, int __B)
+{
+ (__m512i)__builtin_ia32_psllwi512((__v32hi)__A, __B);
+}
-#define _mm512_mask_slli_epi16(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllwi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)(__m512i)(W), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_slli_epi16(__m512i __W, __mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_slli_epi16(__A, __B), \
+ (__v32hi)__W);
+}
-#define _mm512_maskz_slli_epi16(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllwi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_slli_epi16(__mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_slli_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
+}
#define _mm512_bslli_epi128(a, imm) __extension__ ({ \
(__m512i)__builtin_shufflevector( \
@@ -1892,95 +1891,92 @@ _mm512_maskz_srav_epi16 (__mmask32 __U,
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sra_epi16 (__m512i __A, __m128i __B)
+_mm512_sra_epi16(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) -1);
+ return (__m512i)__builtin_ia32_psraw512((__v32hi) __A, (__v8hi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sra_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
- __m128i __B)
+_mm512_mask_sra_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi) __W,
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_sra_epi16(__A, __B), \
+ (__v32hi)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sra_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sra_epi16(__mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_sra_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
}
-#define _mm512_srai_epi16(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrawi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)-1); })
-
-#define _mm512_mask_srai_epi16(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrawi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)(__m512i)(W), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srai_epi16(__m512i __A, int __B)
+{
+ (__m512i)__builtin_ia32_psrawi512((__v32hi)__A, __B);
+}
-#define _mm512_maskz_srai_epi16(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrawi512_mask((__v32hi)(__m512i)(A), (int)(B), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srai_epi16(__m512i __W, __mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srai_epi16(__A, __B), \
+ (__v32hi)__W);
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srai_epi16(__mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srai_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
+}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_srl_epi16 (__m512i __A, __m128i __B)
+_mm512_srl_epi16(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) -1);
+ return (__m512i)__builtin_ia32_psrlw512((__v32hi) __A, (__v8hi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_srl_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
- __m128i __B)
+_mm512_mask_srl_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi) __W,
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srl_epi16(__A, __B), \
+ (__v32hi)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_srl_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
+_mm512_maskz_srl_epi16(__mmask32 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
- (__v8hi) __B,
- (__v32hi)
- _mm512_setzero_hi (),
- (__mmask32) __U);
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srl_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
}
-#define _mm512_srli_epi16(A, imm) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlwi512_mask((__v32hi)(__m512i)(A), (int)(imm), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)-1); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srli_epi16(__m512i __A, int __B)
+{
+ (__m512i)__builtin_ia32_psrlwi512((__v32hi)__A, __B);
+}
-#define _mm512_mask_srli_epi16(W, U, A, imm) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlwi512_mask((__v32hi)(__m512i)(A), (int)(imm), \
- (__v32hi)(__m512i)(W), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srli_epi16(__m512i __W, __mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srli_epi16(__A, __B), \
+ (__v32hi)__W);
+}
-#define _mm512_maskz_srli_epi16(U, A, imm) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlwi512_mask((__v32hi)(__m512i)(A), (int)(imm), \
- (__v32hi)_mm512_setzero_hi(), \
- (__mmask32)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srli_epi16(__mmask32 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, \
+ (__v32hi)_mm512_srli_epi16(__A, __B), \
+ (__v32hi)_mm512_setzero_hi());
+}
#define _mm512_bsrli_epi128(a, imm) __extension__ ({ \
(__m512i)__builtin_shufflevector( \
Modified: cfe/trunk/lib/Headers/avx512fintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512fintrin.h?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512fintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512fintrin.h Sat Nov 12 01:16:59 2016
@@ -5457,67 +5457,91 @@ _mm512_maskz_rolv_epi64 (__mmask8 __U, _
(__v8di)_mm512_setzero_si512(), \
(__mmask8)(U)); })
-#define _mm512_slli_epi32(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_pslldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)-1); })
-
-#define _mm512_mask_slli_epi32(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_pslldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)(__m512i)(W), \
- (__mmask16)(U)); })
-
-#define _mm512_maskz_slli_epi32(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_pslldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)(U)); })
-
-#define _mm512_slli_epi64(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)-1); })
-
-#define _mm512_mask_slli_epi64(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)(__m512i)(W), \
- (__mmask8)(U)); })
-
-#define _mm512_maskz_slli_epi64(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psllqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_slli_epi32(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_pslldi512((__v16si)__A, __B);
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_slli_epi32(__m512i __W, __mmask16 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_slli_epi32(__A, __B), \
+ (__v16si)__W);
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_slli_epi32(__mmask16 __U, __m512i __A, int __B) {
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_slli_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
+}
-#define _mm512_srli_epi32(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)-1); })
-
-#define _mm512_mask_srli_epi32(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)(__m512i)(W), \
- (__mmask16)(U)); })
-
-#define _mm512_maskz_srli_epi32(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrldi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)(U)); })
-
-#define _mm512_srli_epi64(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)-1); })
-
-#define _mm512_mask_srli_epi64(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)(__m512i)(W), \
- (__mmask8)(U)); })
-
-#define _mm512_maskz_srli_epi64(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psrlqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_slli_epi64(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_psllqi512((__v8di)__A, __B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_slli_epi64(__m512i __W, __mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_slli_epi64(__A, __B), \
+ (__v8di)__W);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_slli_epi64(__mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_slli_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srli_epi32(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_psrldi512((__v16si)__A, __B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srli_epi32(__m512i __W, __mmask16 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srli_epi32(__A, __B), \
+ (__v16si)__W);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srli_epi32(__mmask16 __U, __m512i __A, int __B) {
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srli_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srli_epi64(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_psrlqi512((__v8di)__A, __B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srli_epi64(__m512i __W, __mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srli_epi64(__A, __B), \
+ (__v8di)__W);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srli_epi64(__mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srli_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
+}
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_load_epi32 (__m512i __W, __mmask16 __U, void const *__P)
@@ -5992,61 +6016,47 @@ _mm512_mask2_permutex2var_epi32 (__m512i
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sll_epi32 (__m512i __A, __m128i __B)
+_mm512_sll_epi32(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) -1);
+ return (__m512i)__builtin_ia32_pslld512((__v16si) __A, (__v4si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sll_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
+_mm512_mask_sll_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si) __W,
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_sll_epi32(__A, __B), \
+ (__v16si)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sll_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sll_epi32(__mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_sll_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sll_epi64 (__m512i __A, __m128i __B)
+_mm512_sll_epi64(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) -1);
+ return (__m512i)__builtin_ia32_psllq512((__v8di)__A, (__v2di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sll_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
+_mm512_mask_sll_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di) __W,
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_sll_epi64(__A, __B), \
+ (__v8di)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sll_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sll_epi64(__mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_sll_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -6108,61 +6118,47 @@ _mm512_maskz_sllv_epi64 (__mmask8 __U, _
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sra_epi32 (__m512i __A, __m128i __B)
+_mm512_sra_epi32(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) -1);
+ return (__m512i)__builtin_ia32_psrad512((__v16si) __A, (__v4si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sra_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
+_mm512_mask_sra_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si) __W,
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_sra_epi32(__A, __B), \
+ (__v16si)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sra_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sra_epi32(__mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_sra_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_sra_epi64 (__m512i __A, __m128i __B)
+_mm512_sra_epi64(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) -1);
+ return (__m512i)__builtin_ia32_psraq512((__v8di)__A, (__v2di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_sra_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
+_mm512_mask_sra_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di) __W,
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_sra_epi64(__A, __B), \
+ (__v8di)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_sra_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
+_mm512_maskz_sra_epi64(__mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_sra_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -6224,61 +6220,47 @@ _mm512_maskz_srav_epi64 (__mmask8 __U, _
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_srl_epi32 (__m512i __A, __m128i __B)
+_mm512_srl_epi32(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) -1);
+ return (__m512i)__builtin_ia32_psrld512((__v16si) __A, (__v4si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_srl_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
+_mm512_mask_srl_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si) __W,
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srl_epi32(__A, __B), \
+ (__v16si)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_srl_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
+_mm512_maskz_srl_epi32(__mmask16 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
- (__v4si) __B,
- (__v16si)
- _mm512_setzero_si512 (),
- (__mmask16) __U);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srl_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_srl_epi64 (__m512i __A, __m128i __B)
+_mm512_srl_epi64(__m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) -1);
+ return (__m512i)__builtin_ia32_psrlq512((__v8di)__A, (__v2di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_srl_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
+_mm512_mask_srl_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di) __W,
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srl_epi64(__A, __B), \
+ (__v8di)__W);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_srl_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
+_mm512_maskz_srl_epi64(__mmask8 __U, __m512i __A, __m128i __B)
{
- return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
- (__v2di) __B,
- (__v8di)
- _mm512_setzero_si512 (),
- (__mmask8) __U);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srl_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -7110,35 +7092,48 @@ _mm_maskz_scalef_ss (__mmask8 __U, __m12
(__mmask8)(U), \
_MM_FROUND_CUR_DIRECTION); })
-#define _mm512_srai_epi32(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psradi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)-1); })
-
-#define _mm512_mask_srai_epi32(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psradi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)(__m512i)(W), \
- (__mmask16)(U)); })
-
-#define _mm512_maskz_srai_epi32(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psradi512_mask((__v16si)(__m512i)(A), (int)(B), \
- (__v16si)_mm512_setzero_si512(), \
- (__mmask16)(U)); })
-
-#define _mm512_srai_epi64(A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psraqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)-1); })
-
-#define _mm512_mask_srai_epi64(W, U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psraqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)(__m512i)(W), \
- (__mmask8)(U)); })
-
-#define _mm512_maskz_srai_epi64(U, A, B) __extension__ ({ \
- (__m512i)__builtin_ia32_psraqi512_mask((__v8di)(__m512i)(A), (int)(B), \
- (__v8di)_mm512_setzero_si512(), \
- (__mmask8)(U)); })
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srai_epi32(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_psradi512((__v16si)__A, __B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srai_epi32(__m512i __W, __mmask16 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srai_epi32(__A, __B), \
+ (__v16si)__W);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srai_epi32(__mmask16 __U, __m512i __A, int __B) {
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__U, \
+ (__v16si)_mm512_srai_epi32(__A, __B), \
+ (__v16si)_mm512_setzero_si512());
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srai_epi64(__m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_psraqi512((__v8di)__A, __B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srai_epi64(__m512i __W, __mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srai_epi64(__A, __B), \
+ (__v8di)__W);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srai_epi64(__mmask8 __U, __m512i __A, int __B)
+{
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__U, \
+ (__v8di)_mm512_srai_epi64(__A, __B), \
+ (__v8di)_mm512_setzero_si512());
+}
#define _mm512_shuffle_f32x4(A, B, imm) __extension__ ({ \
(__m512)__builtin_ia32_shuf_f32x4_mask((__v16sf)(__m512)(A), \
Modified: cfe/trunk/lib/Headers/avx512vlintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512vlintrin.h?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512vlintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512vlintrin.h Sat Nov 12 01:16:59 2016
@@ -6845,94 +6845,92 @@ _mm256_maskz_srai_epi32(__mmask8 __U, __
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_sra_epi64 (__m128i __A, __m128i __B)
+_mm_sra_epi64(__m128i __A, __m128i __B)
{
- return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
- (__v2di) __B,
- (__v2di)
- _mm_setzero_di (),
- (__mmask8) -1);
+ return (__m128i)__builtin_ia32_psraq128((__v2di)__A, (__v2di)__B);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_sra_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
- __m128i __B)
+_mm_mask_sra_epi64(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
{
- return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
- (__v2di) __B,
- (__v2di) __W,
- (__mmask8) __U);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, \
+ (__v2di)_mm_sra_epi64(__A, __B), \
+ (__v2di)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_sra_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+_mm_maskz_sra_epi64(__mmask8 __U, __m128i __A, __m128i __B)
{
- return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
- (__v2di) __B,
- (__v2di)
- _mm_setzero_di (),
- (__mmask8) __U);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, \
+ (__v2di)_mm_sra_epi64(__A, __B), \
+ (__v2di)_mm_setzero_di());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_sra_epi64 (__m256i __A, __m128i __B)
+_mm256_sra_epi64(__m256i __A, __m128i __B)
{
- return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
- (__v2di) __B,
- (__v4di)
- _mm256_setzero_si256 (),
- (__mmask8) -1);
+ return (__m256i)__builtin_ia32_psraq256((__v4di) __A, (__v2di) __B);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_sra_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
- __m128i __B)
+_mm256_mask_sra_epi64(__m256i __W, __mmask8 __U, __m256i __A, __m128i __B)
{
- return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
- (__v2di) __B,
- (__v4di) __W,
- (__mmask8) __U);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__v4di)_mm256_sra_epi64(__A, __B), \
+ (__v4di)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_sra_epi64 (__mmask8 __U, __m256i __A, __m128i __B)
+_mm256_maskz_sra_epi64(__mmask8 __U, __m256i __A, __m128i __B)
{
- return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
- (__v2di) __B,
- (__v4di)
- _mm256_setzero_si256 (),
- (__mmask8) __U);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__v4di)_mm256_sra_epi64(__A, __B), \
+ (__v4di)_mm256_setzero_si256());
}
-#define _mm_srai_epi64(A, imm) __extension__ ({ \
- (__m128i)__builtin_ia32_psraqi128_mask((__v2di)(__m128i)(A), (int)(imm), \
- (__v2di)_mm_setzero_di(), \
- (__mmask8)-1); })
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_srai_epi64(__m128i __A, int __imm)
+{
+ return (__m128i)__builtin_ia32_psraqi128((__v2di)__A, __imm);
+}
-#define _mm_mask_srai_epi64(W, U, A, imm) __extension__ ({ \
- (__m128i)__builtin_ia32_psraqi128_mask((__v2di)(__m128i)(A), (int)(imm), \
- (__v2di)(__m128i)(W), \
- (__mmask8)(U)); })
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_srai_epi64(__m128i __W, __mmask8 __U, __m128i __A, int __imm)
+{
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, \
+ (__v2di)_mm_srai_epi64(__A, __imm), \
+ (__v2di)__W);
+}
-#define _mm_maskz_srai_epi64(U, A, imm) __extension__ ({ \
- (__m128i)__builtin_ia32_psraqi128_mask((__v2di)(__m128i)(A), (int)(imm), \
- (__v2di)_mm_setzero_si128(), \
- (__mmask8)(U)); })
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_srai_epi64(__mmask8 __U, __m128i __A, int __imm)
+{
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, \
+ (__v2di)_mm_srai_epi64(__A, __imm), \
+ (__v2di)_mm_setzero_di());
+}
-#define _mm256_srai_epi64(A, imm) __extension__ ({ \
- (__m256i)__builtin_ia32_psraqi256_mask((__v4di)(__m256i)(A), (int)(imm), \
- (__v4di)_mm256_setzero_si256(), \
- (__mmask8)-1); })
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_srai_epi64(__m256i __A, int __imm)
+{
+ return (__m256i)__builtin_ia32_psraqi256((__v4di)__A, __imm);
+}
-#define _mm256_mask_srai_epi64(W, U, A, imm) __extension__ ({ \
- (__m256i)__builtin_ia32_psraqi256_mask((__v4di)(__m256i)(A), (int)(imm), \
- (__v4di)(__m256i)(W), \
- (__mmask8)(U)); })
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_srai_epi64(__m256i __W, __mmask8 __U, __m256i __A, int __imm)
+{
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__v4di)_mm256_srai_epi64(__A, __imm), \
+ (__v4di)__W);
+}
-#define _mm256_maskz_srai_epi64(U, A, imm) __extension__ ({ \
- (__m256i)__builtin_ia32_psraqi256_mask((__v4di)(__m256i)(A), (int)(imm), \
- (__v4di)_mm256_setzero_si256(), \
- (__mmask8)(U)); })
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_srai_epi64(__mmask8 __U, __m256i __A, int __imm)
+{
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, \
+ (__v4di)_mm256_srai_epi64(__A, __imm), \
+ (__v4di)_mm256_setzero_si256());
+}
#define _mm_ternarylogic_epi32(A, B, C, imm) __extension__ ({ \
(__m128i)__builtin_ia32_pternlogd128_mask((__v4si)(__m128i)(A), \
Modified: cfe/trunk/lib/Sema/SemaChecking.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaChecking.cpp?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/lib/Sema/SemaChecking.cpp (original)
+++ cfe/trunk/lib/Sema/SemaChecking.cpp Sat Nov 12 01:16:59 2016
@@ -2107,17 +2107,6 @@ bool Sema::CheckX86BuiltinFunctionCall(u
case X86::BI__builtin_ia32_prord256_mask:
case X86::BI__builtin_ia32_prorq128_mask:
case X86::BI__builtin_ia32_prorq256_mask:
- case X86::BI__builtin_ia32_psllwi512_mask:
- case X86::BI__builtin_ia32_psrldi512_mask:
- case X86::BI__builtin_ia32_psrlqi512_mask:
- case X86::BI__builtin_ia32_psrawi512_mask:
- case X86::BI__builtin_ia32_psrlwi512_mask:
- case X86::BI__builtin_ia32_psradi512_mask:
- case X86::BI__builtin_ia32_psraqi128_mask:
- case X86::BI__builtin_ia32_psraqi256_mask:
- case X86::BI__builtin_ia32_psraqi512_mask:
- case X86::BI__builtin_ia32_pslldi512_mask:
- case X86::BI__builtin_ia32_psllqi512_mask:
case X86::BI__builtin_ia32_fpclasspd128_mask:
case X86::BI__builtin_ia32_fpclasspd256_mask:
case X86::BI__builtin_ia32_fpclassps128_mask:
Modified: cfe/trunk/test/CodeGen/avx512bw-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512bw-builtins.c?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512bw-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512bw-builtins.c Sat Nov 12 01:16:59 2016
@@ -1225,37 +1225,41 @@ __m512i test_mm512_maskz_sllv_epi16(__mm
__m512i test_mm512_sll_epi16(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sll_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.w.512
+ // CHECK: @llvm.x86.avx512.psll.w.512
return _mm512_sll_epi16(__A, __B);
}
__m512i test_mm512_mask_sll_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sll_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.w.512
+ // CHECK: @llvm.x86.avx512.psll.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_sll_epi16(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sll_epi16(__mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sll_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.w.512
+ // CHECK: @llvm.x86.avx512.psll.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_sll_epi16(__U, __A, __B);
}
__m512i test_mm512_slli_epi16(__m512i __A) {
// CHECK-LABEL: @test_mm512_slli_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.wi.512
+ // CHECK: @llvm.x86.avx512.pslli.w.512
return _mm512_slli_epi16(__A, 5);
}
__m512i test_mm512_mask_slli_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_slli_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.wi.512
+ // CHECK: @llvm.x86.avx512.pslli.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_slli_epi16(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_slli_epi16(__mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_slli_epi16
- // CHECK: @llvm.x86.avx512.mask.psll.wi.512
+ // CHECK: @llvm.x86.avx512.pslli.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_slli_epi16(__U, __A, 5);
}
@@ -1303,73 +1307,81 @@ __m512i test_mm512_maskz_srav_epi16(__mm
__m512i test_mm512_sra_epi16(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sra_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.w.512
+ // CHECK: @llvm.x86.avx512.psra.w.512
return _mm512_sra_epi16(__A, __B);
}
__m512i test_mm512_mask_sra_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sra_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.w.512
+ // CHECK: @llvm.x86.avx512.psra.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_sra_epi16(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sra_epi16(__mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sra_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.w.512
+ // CHECK: @llvm.x86.avx512.psra.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_sra_epi16(__U, __A, __B);
}
__m512i test_mm512_srai_epi16(__m512i __A) {
// CHECK-LABEL: @test_mm512_srai_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.wi.512
+ // CHECK: @llvm.x86.avx512.psrai.w.512
return _mm512_srai_epi16(__A, 5);
}
__m512i test_mm512_mask_srai_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srai_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.wi.512
+ // CHECK: @llvm.x86.avx512.psrai.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_srai_epi16(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srai_epi16(__mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srai_epi16
- // CHECK: @llvm.x86.avx512.mask.psra.wi.512
+ // CHECK: @llvm.x86.avx512.psrai.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_srai_epi16(__U, __A, 5);
}
__m512i test_mm512_srl_epi16(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_srl_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ // CHECK: @llvm.x86.avx512.psrl.w.512
return _mm512_srl_epi16(__A, __B);
}
__m512i test_mm512_mask_srl_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_srl_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ // CHECK: @llvm.x86.avx512.psrl.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_srl_epi16(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_srl_epi16(__mmask32 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_srl_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ // CHECK: @llvm.x86.avx512.psrl.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_srl_epi16(__U, __A, __B);
}
__m512i test_mm512_srli_epi16(__m512i __A) {
// CHECK-LABEL: @test_mm512_srli_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ // CHECK: @llvm.x86.avx512.psrli.w.512
return _mm512_srli_epi16(__A, 5);
}
__m512i test_mm512_mask_srli_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srli_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ // CHECK: @llvm.x86.avx512.psrli.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_mask_srli_epi16(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srli_epi16(__mmask32 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srli_epi16
- // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ // CHECK: @llvm.x86.avx512.psrli.w.512
+ // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
return _mm512_maskz_srli_epi16(__U, __A, 5);
}
Modified: cfe/trunk/test/CodeGen/avx512f-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512f-builtins.c?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512f-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512f-builtins.c Sat Nov 12 01:16:59 2016
@@ -2808,73 +2808,80 @@ __m512i test_mm512_maskz_rorv_epi64(__mm
__m512i test_mm512_slli_epi32(__m512i __A) {
// CHECK-LABEL: @test_mm512_slli_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.di.512
+ // CHECK: @llvm.x86.avx512.pslli.d.512
return _mm512_slli_epi32(__A, 5);
}
__m512i test_mm512_mask_slli_epi32(__m512i __W, __mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_slli_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.di.512
+ // CHECK: @llvm.x86.avx512.pslli.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_slli_epi32(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_slli_epi32(__mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_slli_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.di.512
+ // CHECK: @llvm.x86.avx512.pslli.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_slli_epi32(__U, __A, 5);
}
__m512i test_mm512_slli_epi64(__m512i __A) {
// CHECK-LABEL: @test_mm512_slli_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.qi.512
+ // CHECK: @llvm.x86.avx512.pslli.q.512
return _mm512_slli_epi64(__A, 5);
}
__m512i test_mm512_mask_slli_epi64(__m512i __W, __mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_slli_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.qi.512
+ // CHECK: @llvm.x86.avx512.pslli.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_slli_epi64(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_slli_epi64(__mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_slli_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.qi.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_slli_epi64(__U, __A, 5);
}
__m512i test_mm512_srli_epi32(__m512i __A) {
// CHECK-LABEL: @test_mm512_srli_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.di.512
+ // CHECK: @llvm.x86.avx512.psrli.d.512
return _mm512_srli_epi32(__A, 5);
}
__m512i test_mm512_mask_srli_epi32(__m512i __W, __mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srli_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.di.512
+ // CHECK: @llvm.x86.avx512.psrli.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_srli_epi32(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srli_epi32(__mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srli_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.di.512
+ // CHECK: @llvm.x86.avx512.psrli.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_srli_epi32(__U, __A, 5);
}
__m512i test_mm512_srli_epi64(__m512i __A) {
// CHECK-LABEL: @test_mm512_srli_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.qi.512
+ // CHECK: @llvm.x86.avx512.psrli.q.512
return _mm512_srli_epi64(__A, 5);
}
__m512i test_mm512_mask_srli_epi64(__m512i __W, __mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srli_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.qi.512
+ // CHECK: @llvm.x86.avx512.psrli.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_srli_epi64(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srli_epi64(__mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srli_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.qi.512
+ // CHECK: @llvm.x86.avx512.psrli.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_srli_epi64(__U, __A, 5);
}
@@ -4138,73 +4145,81 @@ __m128 test_mm_maskz_scalef_round_ss(__m
__m512i test_mm512_srai_epi32(__m512i __A) {
// CHECK-LABEL: @test_mm512_srai_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.di.512
+ // CHECK: @llvm.x86.avx512.psrai.d.512
return _mm512_srai_epi32(__A, 5);
}
__m512i test_mm512_mask_srai_epi32(__m512i __W, __mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srai_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.di.512
+ // CHECK: @llvm.x86.avx512.psrai.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_srai_epi32(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srai_epi32(__mmask16 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srai_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.di.512
+ // CHECK: @llvm.x86.avx512.psrai.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_srai_epi32(__U, __A, 5);
}
__m512i test_mm512_srai_epi64(__m512i __A) {
// CHECK-LABEL: @test_mm512_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.512
+ // CHECK: @llvm.x86.avx512.psrai.q.512
return _mm512_srai_epi64(__A, 5);
}
__m512i test_mm512_mask_srai_epi64(__m512i __W, __mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_mask_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.512
+ // CHECK: @llvm.x86.avx512.psrai.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_srai_epi64(__W, __U, __A, 5);
}
__m512i test_mm512_maskz_srai_epi64(__mmask8 __U, __m512i __A) {
// CHECK-LABEL: @test_mm512_maskz_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.512
+ // CHECK: @llvm.x86.avx512.psrai.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_srai_epi64(__U, __A, 5);
}
__m512i test_mm512_sll_epi32(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sll_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.d
+ // CHECK: @llvm.x86.avx512.psll.d.512
return _mm512_sll_epi32(__A, __B);
}
__m512i test_mm512_mask_sll_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sll_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.d
+ // CHECK: @llvm.x86.avx512.psll.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_sll_epi32(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sll_epi32(__mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sll_epi32
- // CHECK: @llvm.x86.avx512.mask.psll.d
+ // CHECK: @llvm.x86.avx512.psll.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_sll_epi32(__U, __A, __B);
}
__m512i test_mm512_sll_epi64(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sll_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.q
+ // CHECK: @llvm.x86.avx512.psll.q.512
return _mm512_sll_epi64(__A, __B);
}
__m512i test_mm512_mask_sll_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sll_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.q
+ // CHECK: @llvm.x86.avx512.psll.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_sll_epi64(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sll_epi64(__mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sll_epi64
- // CHECK: @llvm.x86.avx512.mask.psll.q
+ // CHECK: @llvm.x86.avx512.psll.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_sll_epi64(__U, __A, __B);
}
@@ -4246,37 +4261,41 @@ __m512i test_mm512_maskz_sllv_epi64(__mm
__m512i test_mm512_sra_epi32(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sra_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.d
+ // CHECK: @llvm.x86.avx512.psra.d.512
return _mm512_sra_epi32(__A, __B);
}
__m512i test_mm512_mask_sra_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sra_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.d
+ // CHECK: @llvm.x86.avx512.psra.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_sra_epi32(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sra_epi32(__mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sra_epi32
- // CHECK: @llvm.x86.avx512.mask.psra.d
+ // CHECK: @llvm.x86.avx512.psra.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_sra_epi32(__U, __A, __B);
}
__m512i test_mm512_sra_epi64(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q
+ // CHECK: @llvm.x86.avx512.psra.q.512
return _mm512_sra_epi64(__A, __B);
}
__m512i test_mm512_mask_sra_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q
+ // CHECK: @llvm.x86.avx512.psra.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_sra_epi64(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_sra_epi64(__mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q
+ // CHECK: @llvm.x86.avx512.psra.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_sra_epi64(__U, __A, __B);
}
@@ -4318,37 +4337,41 @@ __m512i test_mm512_maskz_srav_epi64(__mm
__m512i test_mm512_srl_epi32(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_srl_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.d
+ // CHECK: @llvm.x86.avx512.psrl.d.512
return _mm512_srl_epi32(__A, __B);
}
__m512i test_mm512_mask_srl_epi32(__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_srl_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.d
+ // CHECK: @llvm.x86.avx512.psrl.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_srl_epi32(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_srl_epi32(__mmask16 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_srl_epi32
- // CHECK: @llvm.x86.avx512.mask.psrl.d
+ // CHECK: @llvm.x86.avx512.psrl.d.512
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_srl_epi32(__U, __A, __B);
}
__m512i test_mm512_srl_epi64(__m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_srl_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.q
+ // CHECK: @llvm.x86.avx512.psrl.q.512
return _mm512_srl_epi64(__A, __B);
}
__m512i test_mm512_mask_srl_epi64(__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_mask_srl_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.q
+ // CHECK: @llvm.x86.avx512.psrl.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_mask_srl_epi64(__W, __U, __A, __B);
}
__m512i test_mm512_maskz_srl_epi64(__mmask8 __U, __m512i __A, __m128i __B) {
// CHECK-LABEL: @test_mm512_maskz_srl_epi64
- // CHECK: @llvm.x86.avx512.mask.psrl.q
+ // CHECK: @llvm.x86.avx512.psrl.q.512
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
return _mm512_maskz_srl_epi64(__U, __A, __B);
}
Modified: cfe/trunk/test/CodeGen/avx512vl-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512vl-builtins.c?rev=286714&r1=286713&r2=286714&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512vl-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512vl-builtins.c Sat Nov 12 01:16:59 2016
@@ -5358,73 +5358,81 @@ __m256i test_mm256_maskz_srai_epi32(__mm
__m128i test_mm_sra_epi64(__m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.128
+ // CHECK: @llvm.x86.avx512.psra.q.128
return _mm_sra_epi64(__A, __B);
}
__m128i test_mm_mask_sra_epi64(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_mask_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.128
+ // CHECK: @llvm.x86.avx512.psra.q.128
+ // CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_mask_sra_epi64(__W, __U, __A, __B);
}
__m128i test_mm_maskz_sra_epi64(__mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_maskz_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.128
+ // CHECK: @llvm.x86.avx512.psra.q.128
+ // CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_maskz_sra_epi64(__U, __A, __B);
}
__m256i test_mm256_sra_epi64(__m256i __A, __m128i __B) {
// CHECK-LABEL: @test_mm256_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.256
+ // CHECK: @llvm.x86.avx512.psra.q.256
return _mm256_sra_epi64(__A, __B);
}
__m256i test_mm256_mask_sra_epi64(__m256i __W, __mmask8 __U, __m256i __A, __m128i __B) {
// CHECK-LABEL: @test_mm256_mask_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.256
+ // CHECK: @llvm.x86.avx512.psra.q.256
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_mask_sra_epi64(__W, __U, __A, __B);
}
__m256i test_mm256_maskz_sra_epi64(__mmask8 __U, __m256i __A, __m128i __B) {
// CHECK-LABEL: @test_mm256_maskz_sra_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.q.256
+ // CHECK: @llvm.x86.avx512.psra.q.256
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_maskz_sra_epi64(__U, __A, __B);
}
__m128i test_mm_srai_epi64(__m128i __A) {
// CHECK-LABEL: @test_mm_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.128
+ // CHECK: @llvm.x86.avx512.psrai.q.128
return _mm_srai_epi64(__A, 5);
}
__m128i test_mm_mask_srai_epi64(__m128i __W, __mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_mask_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.128
+ // CHECK: @llvm.x86.avx512.psrai.q.128
+ // CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_mask_srai_epi64(__W, __U, __A, 5);
}
__m128i test_mm_maskz_srai_epi64(__mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_maskz_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.128
+ // CHECK: @llvm.x86.avx512.psrai.q.128
+ // CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_maskz_srai_epi64(__U, __A, 5);
}
__m256i test_mm256_srai_epi64(__m256i __A) {
// CHECK-LABEL: @test_mm256_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.256
+ // CHECK: @llvm.x86.avx512.psrai.q.256
return _mm256_srai_epi64(__A, 5);
}
__m256i test_mm256_mask_srai_epi64(__m256i __W, __mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_mask_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.256
+ // CHECK: @llvm.x86.avx512.psrai.q.256
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_mask_srai_epi64(__W, __U, __A, 5);
}
__m256i test_mm256_maskz_srai_epi64(__mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_maskz_srai_epi64
- // CHECK: @llvm.x86.avx512.mask.psra.qi.256
+ // CHECK: @llvm.x86.avx512.psrai.q.256
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_maskz_srai_epi64(__U, __A, 5);
}
More information about the cfe-commits
mailing list