[clang] 3ef88b3 - Revert rG8ee135dcf8ff060656ad481c3e980fe8763576f5 "[X86] Remove `__builtin_ia32_pmax/min` intrinsics and use generic `__builtin_elementwise_max/min`"
Simon Pilgrim via cfe-commits
cfe-commits at lists.llvm.org
Fri Jan 21 04:35:56 PST 2022
Author: Simon Pilgrim
Date: 2022-01-21T12:34:19Z
New Revision: 3ef88b31843e040c95f23ff2c3c206f1fa399c05
URL: https://github.com/llvm/llvm-project/commit/3ef88b31843e040c95f23ff2c3c206f1fa399c05
DIFF: https://github.com/llvm/llvm-project/commit/3ef88b31843e040c95f23ff2c3c206f1fa399c05.diff
LOG: Revert rG8ee135dcf8ff060656ad481c3e980fe8763576f5 "[X86] Remove `__builtin_ia32_pmax/min` intrinsics and use generic `__builtin_elementwise_max/min`"
Some build bots are referencing the `__builtin_ia32_pmax/min` intrinsics via alternative headers
Added:
Modified:
clang/include/clang/Basic/BuiltinsX86.def
clang/lib/CodeGen/CGBuiltin.cpp
clang/lib/Headers/avx2intrin.h
clang/lib/Headers/avx512bwintrin.h
clang/lib/Headers/avx512fintrin.h
clang/lib/Headers/avx512vlintrin.h
clang/lib/Headers/emmintrin.h
clang/lib/Headers/smmintrin.h
clang/test/CodeGen/builtins-x86.c
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index a8f556724862..9b7c763b0c6c 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -265,6 +265,10 @@ TARGET_BUILTIN(__builtin_ia32_psubusw128, "V8sV8sV8s", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_pmulhw128, "V8sV8sV8s", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_pavgb128, "V16cV16cV16c", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_pavgw128, "V8sV8sV8s", "ncV:128:", "sse2")
+TARGET_BUILTIN(__builtin_ia32_pmaxub128, "V16cV16cV16c", "ncV:128:", "sse2")
+TARGET_BUILTIN(__builtin_ia32_pmaxsw128, "V8sV8sV8s", "ncV:128:", "sse2")
+TARGET_BUILTIN(__builtin_ia32_pminub128, "V16cV16cV16c", "ncV:128:", "sse2")
+TARGET_BUILTIN(__builtin_ia32_pminsw128, "V8sV8sV8s", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_packsswb128, "V16cV8sV8s", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_packssdw128, "V8sV4iV4i", "ncV:128:", "sse2")
TARGET_BUILTIN(__builtin_ia32_packuswb128, "V16cV8sV8s", "ncV:128:", "sse2")
@@ -373,6 +377,14 @@ TARGET_BUILTIN(__builtin_ia32_blendvpd, "V2dV2dV2dV2d", "ncV:128:", "sse4.1")
TARGET_BUILTIN(__builtin_ia32_blendvps, "V4fV4fV4fV4f", "ncV:128:", "sse4.1")
TARGET_BUILTIN(__builtin_ia32_packusdw128, "V8sV4iV4i", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pmaxsb128, "V16cV16cV16c", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pmaxsd128, "V4iV4iV4i", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pmaxud128, "V4iV4iV4i", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pmaxuw128, "V8sV8sV8s", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pminsb128, "V16cV16cV16c", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pminsd128, "V4iV4iV4i", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pminud128, "V4iV4iV4i", "ncV:128:", "sse4.1")
+TARGET_BUILTIN(__builtin_ia32_pminuw128, "V8sV8sV8s", "ncV:128:", "sse4.1")
TARGET_BUILTIN(__builtin_ia32_pmuldq128, "V2OiV4iV4i", "ncV:128:", "sse4.1")
TARGET_BUILTIN(__builtin_ia32_roundps, "V4fV4fIi", "ncV:128:", "sse4.1")
TARGET_BUILTIN(__builtin_ia32_roundss, "V4fV4fV4fIi", "ncV:128:", "sse4.1")
@@ -568,6 +580,18 @@ TARGET_BUILTIN(__builtin_ia32_phsubd256, "V8iV8iV8i", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_phsubsw256, "V16sV16sV16s", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_pmaddubsw256, "V16sV32cV32c", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_pmaddwd256, "V8iV16sV16s", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxub256, "V32cV32cV32c", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxuw256, "V16sV16sV16s", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxud256, "V8iV8iV8i", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxsb256, "V32cV32cV32c", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxsw256, "V16sV16sV16s", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pmaxsd256, "V8iV8iV8i", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminub256, "V32cV32cV32c", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminuw256, "V16sV16sV16s", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminud256, "V8iV8iV8i", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminsb256, "V32cV32cV32c", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminsw256, "V16sV16sV16s", "ncV:256:", "avx2")
+TARGET_BUILTIN(__builtin_ia32_pminsd256, "V8iV8iV8i", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_pmovmskb256, "iV32c", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_pmuldq256, "V4OiV8iV8i", "ncV:256:", "avx2")
TARGET_BUILTIN(__builtin_ia32_pmulhrsw256, "V16sV16sV16s", "ncV:256:", "avx2")
@@ -897,6 +921,14 @@ TARGET_BUILTIN(__builtin_ia32_cvtudq2ps512_mask, "V16fV16iV16fUsIi", "ncV:512:",
TARGET_BUILTIN(__builtin_ia32_cvtpd2ps512_mask, "V8fV8dV8fUcIi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_vcvtps2ph512_mask, "V16sV16fIiV16sUs", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_vcvtph2ps512_mask, "V16fV16sV16fUsIi", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmaxsd512, "V16iV16iV16i", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmaxsq512, "V8OiV8OiV8Oi", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmaxud512, "V16iV16iV16i", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmaxuq512, "V8OiV8OiV8Oi", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pminsd512, "V16iV16iV16i", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pminsq512, "V8OiV8OiV8Oi", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pminud512, "V16iV16iV16i", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pminuq512, "V8OiV8OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_pmuldq512, "V8OiV16iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_pmuludq512, "V8OiV16iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_loaddqusi512_mask, "V16iiC*V16iUs", "nV:512:", "avx512f")
@@ -1015,6 +1047,14 @@ TARGET_BUILTIN(__builtin_ia32_paddusb512, "V64cV64cV64c", "ncV:512:", "avx512bw"
TARGET_BUILTIN(__builtin_ia32_paddusw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pavgb512, "V64cV64cV64c", "ncV:512:", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pavgw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pmaxsb512, "V64cV64cV64c", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pmaxsw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pmaxub512, "V64cV64cV64c", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pmaxuw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pminsb512, "V64cV64cV64c", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pminsw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pminub512, "V64cV64cV64c", "ncV:512:", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pminuw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pshufb512, "V64cV64cV64c", "ncV:512:", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_psubsb512, "V64cV64cV64c", "ncV:512:", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_psubsw512, "V32sV32sV32s", "ncV:512:", "avx512bw")
@@ -1148,6 +1188,14 @@ TARGET_BUILTIN(__builtin_ia32_getexppd128_mask, "V2dV2dV2dUc", "ncV:128:", "avx5
TARGET_BUILTIN(__builtin_ia32_getexppd256_mask, "V4dV4dV4dUc", "ncV:256:", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_getexpps128_mask, "V4fV4fV4fUc", "ncV:128:", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_getexpps256_mask, "V8fV8fV8fUc", "ncV:256:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmaxsq128, "V2OiV2OiV2Oi", "ncV:128:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmaxsq256, "V4OiV4OiV4Oi", "ncV:256:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmaxuq128, "V2OiV2OiV2Oi", "ncV:128:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmaxuq256, "V4OiV4OiV4Oi", "ncV:256:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pminsq128, "V2OiV2OiV2Oi", "ncV:128:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pminsq256, "V4OiV4OiV4Oi", "ncV:256:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pminuq128, "V2OiV2OiV2Oi", "ncV:128:", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pminuq256, "V4OiV4OiV4Oi", "ncV:256:", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_rndscalepd_128_mask, "V2dV2dIiV2dUc", "ncV:128:", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_rndscalepd_256_mask, "V4dV4dIiV4dUc", "ncV:256:", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_rndscaleps_128_mask, "V4fV4fIiV4fUc", "ncV:128:", "avx512vl")
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 4c68b20067b9..49f054ec1a98 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -14285,6 +14285,58 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
return Builder.CreateCall(F, Ops[0]);
}
}
+ case X86::BI__builtin_ia32_pmaxsb128:
+ case X86::BI__builtin_ia32_pmaxsw128:
+ case X86::BI__builtin_ia32_pmaxsd128:
+ case X86::BI__builtin_ia32_pmaxsq128:
+ case X86::BI__builtin_ia32_pmaxsb256:
+ case X86::BI__builtin_ia32_pmaxsw256:
+ case X86::BI__builtin_ia32_pmaxsd256:
+ case X86::BI__builtin_ia32_pmaxsq256:
+ case X86::BI__builtin_ia32_pmaxsb512:
+ case X86::BI__builtin_ia32_pmaxsw512:
+ case X86::BI__builtin_ia32_pmaxsd512:
+ case X86::BI__builtin_ia32_pmaxsq512:
+ return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::smax);
+ case X86::BI__builtin_ia32_pmaxub128:
+ case X86::BI__builtin_ia32_pmaxuw128:
+ case X86::BI__builtin_ia32_pmaxud128:
+ case X86::BI__builtin_ia32_pmaxuq128:
+ case X86::BI__builtin_ia32_pmaxub256:
+ case X86::BI__builtin_ia32_pmaxuw256:
+ case X86::BI__builtin_ia32_pmaxud256:
+ case X86::BI__builtin_ia32_pmaxuq256:
+ case X86::BI__builtin_ia32_pmaxub512:
+ case X86::BI__builtin_ia32_pmaxuw512:
+ case X86::BI__builtin_ia32_pmaxud512:
+ case X86::BI__builtin_ia32_pmaxuq512:
+ return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::umax);
+ case X86::BI__builtin_ia32_pminsb128:
+ case X86::BI__builtin_ia32_pminsw128:
+ case X86::BI__builtin_ia32_pminsd128:
+ case X86::BI__builtin_ia32_pminsq128:
+ case X86::BI__builtin_ia32_pminsb256:
+ case X86::BI__builtin_ia32_pminsw256:
+ case X86::BI__builtin_ia32_pminsd256:
+ case X86::BI__builtin_ia32_pminsq256:
+ case X86::BI__builtin_ia32_pminsb512:
+ case X86::BI__builtin_ia32_pminsw512:
+ case X86::BI__builtin_ia32_pminsd512:
+ case X86::BI__builtin_ia32_pminsq512:
+ return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::smin);
+ case X86::BI__builtin_ia32_pminub128:
+ case X86::BI__builtin_ia32_pminuw128:
+ case X86::BI__builtin_ia32_pminud128:
+ case X86::BI__builtin_ia32_pminuq128:
+ case X86::BI__builtin_ia32_pminub256:
+ case X86::BI__builtin_ia32_pminuw256:
+ case X86::BI__builtin_ia32_pminud256:
+ case X86::BI__builtin_ia32_pminuq256:
+ case X86::BI__builtin_ia32_pminub512:
+ case X86::BI__builtin_ia32_pminuw512:
+ case X86::BI__builtin_ia32_pminud512:
+ case X86::BI__builtin_ia32_pminuq512:
+ return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::umin);
case X86::BI__builtin_ia32_pmuludq128:
case X86::BI__builtin_ia32_pmuludq256:
diff --git a/clang/lib/Headers/avx2intrin.h b/clang/lib/Headers/avx2intrin.h
index e33514a60ff3..c9ad74ce3fa4 100644
--- a/clang/lib/Headers/avx2intrin.h
+++ b/clang/lib/Headers/avx2intrin.h
@@ -253,73 +253,73 @@ _mm256_madd_epi16(__m256i __a, __m256i __b)
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epi8(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v32qs)__a, (__v32qs)__b);
+ return (__m256i)__builtin_ia32_pmaxsb256((__v32qi)__a, (__v32qi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epi16(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v16hi)__a, (__v16hi)__b);
+ return (__m256i)__builtin_ia32_pmaxsw256((__v16hi)__a, (__v16hi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epi32(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v8si)__a, (__v8si)__b);
+ return (__m256i)__builtin_ia32_pmaxsd256((__v8si)__a, (__v8si)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epu8(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v32qu)__a, (__v32qu)__b);
+ return (__m256i)__builtin_ia32_pmaxub256((__v32qi)__a, (__v32qi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epu16(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v16hu)__a, (__v16hu)__b);
+ return (__m256i)__builtin_ia32_pmaxuw256((__v16hi)__a, (__v16hi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epu32(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_max((__v8su)__a, (__v8su)__b);
+ return (__m256i)__builtin_ia32_pmaxud256((__v8si)__a, (__v8si)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epi8(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v32qs)__a, (__v32qs)__b);
+ return (__m256i)__builtin_ia32_pminsb256((__v32qi)__a, (__v32qi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epi16(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v16hi)__a, (__v16hi)__b);
+ return (__m256i)__builtin_ia32_pminsw256((__v16hi)__a, (__v16hi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epi32(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v8si)__a, (__v8si)__b);
+ return (__m256i)__builtin_ia32_pminsd256((__v8si)__a, (__v8si)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epu8(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v32qu)__a, (__v32qu)__b);
+ return (__m256i)__builtin_ia32_pminub256((__v32qi)__a, (__v32qi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epu16(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v16hu)__a, (__v16hu)__b);
+ return (__m256i)__builtin_ia32_pminuw256 ((__v16hi)__a, (__v16hi)__b);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epu32(__m256i __a, __m256i __b)
{
- return (__m256i)__builtin_elementwise_min((__v8su)__a, (__v8su)__b);
+ return (__m256i)__builtin_ia32_pminud256((__v8si)__a, (__v8si)__b);
}
static __inline__ int __DEFAULT_FN_ATTRS256
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 522ef100bab1..53319eb23011 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -751,7 +751,7 @@ _mm512_maskz_avg_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epi8 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v64qs) __A, (__v64qs) __B);
+ return (__m512i)__builtin_ia32_pmaxsb512((__v64qi) __A, (__v64qi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -773,7 +773,7 @@ _mm512_mask_max_epi8 (__m512i __W, __mmask64 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epi16 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v32hi) __A, (__v32hi) __B);
+ return (__m512i)__builtin_ia32_pmaxsw512((__v32hi) __A, (__v32hi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -796,7 +796,7 @@ _mm512_mask_max_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epu8 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v64qu)__A, (__v64qu)__B);
+ return (__m512i)__builtin_ia32_pmaxub512((__v64qi)__A, (__v64qi)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -818,7 +818,7 @@ _mm512_mask_max_epu8 (__m512i __W, __mmask64 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epu16 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v32hu)__A, (__v32hu)__B);
+ return (__m512i)__builtin_ia32_pmaxuw512((__v32hi)__A, (__v32hi)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -840,7 +840,7 @@ _mm512_mask_max_epu16 (__m512i __W, __mmask32 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epi8 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v64qs) __A, (__v64qs) __B);
+ return (__m512i)__builtin_ia32_pminsb512((__v64qi) __A, (__v64qi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -862,7 +862,7 @@ _mm512_mask_min_epi8 (__m512i __W, __mmask64 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epi16 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v32hi) __A, (__v32hi) __B);
+ return (__m512i)__builtin_ia32_pminsw512((__v32hi) __A, (__v32hi) __B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -884,7 +884,7 @@ _mm512_mask_min_epi16 (__m512i __W, __mmask32 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epu8 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v64qu)__A, (__v64qu)__B);
+ return (__m512i)__builtin_ia32_pminub512((__v64qi)__A, (__v64qi)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -906,7 +906,7 @@ _mm512_mask_min_epu8 (__m512i __W, __mmask64 __M, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epu16 (__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v32hu)__A, (__v32hu)__B);
+ return (__m512i)__builtin_ia32_pminuw512((__v32hi)__A, (__v32hi)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 8695aeb94de2..9b02a7cffc64 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -1090,7 +1090,7 @@ static __inline __m512i
__DEFAULT_FN_ATTRS512
_mm512_max_epi32(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v16si)__A, (__v16si)__B);
+ return (__m512i)__builtin_ia32_pmaxsd512((__v16si)__A, (__v16si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1112,7 +1112,7 @@ _mm512_maskz_max_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epu32(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v16su)__A, (__v16su)__B);
+ return (__m512i)__builtin_ia32_pmaxud512((__v16si)__A, (__v16si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1134,7 +1134,7 @@ _mm512_maskz_max_epu32 (__mmask16 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epi64(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v8di)__A, (__v8di)__B);
+ return (__m512i)__builtin_ia32_pmaxsq512((__v8di)__A, (__v8di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1156,7 +1156,7 @@ _mm512_maskz_max_epi64 (__mmask8 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_max_epu64(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_max((__v8du)__A, (__v8du)__B);
+ return (__m512i)__builtin_ia32_pmaxuq512((__v8di)__A, (__v8di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1325,7 +1325,7 @@ static __inline __m512i
__DEFAULT_FN_ATTRS512
_mm512_min_epi32(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v16si)__A, (__v16si)__B);
+ return (__m512i)__builtin_ia32_pminsd512((__v16si)__A, (__v16si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1347,7 +1347,7 @@ _mm512_maskz_min_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epu32(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v16su)__A, (__v16su)__B);
+ return (__m512i)__builtin_ia32_pminud512((__v16si)__A, (__v16si)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1369,7 +1369,7 @@ _mm512_maskz_min_epu32 (__mmask16 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epi64(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v8di)__A, (__v8di)__B);
+ return (__m512i)__builtin_ia32_pminsq512((__v8di)__A, (__v8di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
@@ -1391,7 +1391,7 @@ _mm512_maskz_min_epi64 (__mmask8 __M, __m512i __A, __m512i __B)
static __inline __m512i __DEFAULT_FN_ATTRS512
_mm512_min_epu64(__m512i __A, __m512i __B)
{
- return (__m512i)__builtin_elementwise_min((__v8du)__A, (__v8du)__B);
+ return (__m512i)__builtin_ia32_pminuq512((__v8di)__A, (__v8di)__B);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS512
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 178c9dbc0e6e..eddb99902e3d 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -3054,7 +3054,7 @@ _mm256_mask_max_epi32(__m256i __W, __mmask8 __M, __m256i __A, __m256i __B) {
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_max_epi64 (__m128i __A, __m128i __B) {
- return (__m128i)__builtin_elementwise_max((__v2di)__A, (__v2di)__B);
+ return (__m128i)__builtin_ia32_pmaxsq128((__v2di)__A, (__v2di)__B);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3073,7 +3073,7 @@ _mm_mask_max_epi64 (__m128i __W, __mmask8 __M, __m128i __A, __m128i __B) {
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epi64 (__m256i __A, __m256i __B) {
- return (__m256i)__builtin_elementwise_max((__v4di)__A, (__v4di)__B);
+ return (__m256i)__builtin_ia32_pmaxsq256((__v4di)__A, (__v4di)__B);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -3120,7 +3120,7 @@ _mm256_mask_max_epu32(__m256i __W, __mmask8 __M, __m256i __A, __m256i __B) {
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_max_epu64 (__m128i __A, __m128i __B) {
- return (__m128i)__builtin_elementwise_max((__v2du)__A, (__v2du)__B);
+ return (__m128i)__builtin_ia32_pmaxuq128((__v2di)__A, (__v2di)__B);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3139,7 +3139,7 @@ _mm_mask_max_epu64 (__m128i __W, __mmask8 __M, __m128i __A, __m128i __B) {
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_max_epu64 (__m256i __A, __m256i __B) {
- return (__m256i)__builtin_elementwise_max((__v4du)__A, (__v4du)__B);
+ return (__m256i)__builtin_ia32_pmaxuq256((__v4di)__A, (__v4di)__B);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -3186,7 +3186,7 @@ _mm256_mask_min_epi32(__m256i __W, __mmask8 __M, __m256i __A, __m256i __B) {
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_min_epi64 (__m128i __A, __m128i __B) {
- return (__m128i)__builtin_elementwise_min((__v2di)__A, (__v2di)__B);
+ return (__m128i)__builtin_ia32_pminsq128((__v2di)__A, (__v2di)__B);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3205,7 +3205,7 @@ _mm_maskz_min_epi64 (__mmask8 __M, __m128i __A, __m128i __B) {
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epi64 (__m256i __A, __m256i __B) {
- return (__m256i)__builtin_elementwise_min((__v4di)__A, (__v4di)__B);
+ return (__m256i)__builtin_ia32_pminsq256((__v4di)__A, (__v4di)__B);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -3252,7 +3252,7 @@ _mm256_mask_min_epu32(__m256i __W, __mmask8 __M, __m256i __A, __m256i __B) {
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_min_epu64 (__m128i __A, __m128i __B) {
- return (__m128i)__builtin_elementwise_min((__v2du)__A, (__v2du)__B);
+ return (__m128i)__builtin_ia32_pminuq128((__v2di)__A, (__v2di)__B);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3271,7 +3271,7 @@ _mm_maskz_min_epu64 (__mmask8 __M, __m128i __A, __m128i __B) {
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_min_epu64 (__m256i __A, __m256i __B) {
- return (__m256i)__builtin_elementwise_min((__v4du)__A, (__v4du)__B);
+ return (__m256i)__builtin_ia32_pminuq256((__v4di)__A, (__v4di)__B);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index 4618b808efc4..6e9c3032c21f 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -2375,7 +2375,7 @@ _mm_madd_epi16(__m128i __a, __m128i __b)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epi16(__m128i __a, __m128i __b)
{
- return (__m128i)__builtin_elementwise_max((__v8hi)__a, (__v8hi)__b);
+ return (__m128i)__builtin_ia32_pmaxsw128((__v8hi)__a, (__v8hi)__b);
}
/// Compares corresponding elements of two 128-bit unsigned [16 x i8]
@@ -2395,7 +2395,7 @@ _mm_max_epi16(__m128i __a, __m128i __b)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epu8(__m128i __a, __m128i __b)
{
- return (__m128i)__builtin_elementwise_max((__v16qu)__a, (__v16qu)__b);
+ return (__m128i)__builtin_ia32_pmaxub128((__v16qi)__a, (__v16qi)__b);
}
/// Compares corresponding elements of two 128-bit signed [8 x i16]
@@ -2415,7 +2415,7 @@ _mm_max_epu8(__m128i __a, __m128i __b)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epi16(__m128i __a, __m128i __b)
{
- return (__m128i)__builtin_elementwise_min((__v8hi)__a, (__v8hi)__b);
+ return (__m128i)__builtin_ia32_pminsw128((__v8hi)__a, (__v8hi)__b);
}
/// Compares corresponding elements of two 128-bit unsigned [16 x i8]
@@ -2435,7 +2435,7 @@ _mm_min_epi16(__m128i __a, __m128i __b)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epu8(__m128i __a, __m128i __b)
{
- return (__m128i)__builtin_elementwise_min((__v16qu)__a, (__v16qu)__b);
+ return (__m128i)__builtin_ia32_pminub128((__v16qi)__a, (__v16qi)__b);
}
/// Multiplies the corresponding elements of two signed [8 x i16]
diff --git a/clang/lib/Headers/smmintrin.h b/clang/lib/Headers/smmintrin.h
index 0df59c5fcc59..710e55aaa120 100644
--- a/clang/lib/Headers/smmintrin.h
+++ b/clang/lib/Headers/smmintrin.h
@@ -668,7 +668,7 @@ _mm_stream_load_si128 (__m128i const *__V)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epi8 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_min((__v16qs) __V1, (__v16qs) __V2);
+ return (__m128i) __builtin_ia32_pminsb128 ((__v16qi) __V1, (__v16qi) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -687,7 +687,7 @@ _mm_min_epi8 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epi8 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_max((__v16qs) __V1, (__v16qs) __V2);
+ return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi) __V1, (__v16qi) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -706,7 +706,7 @@ _mm_max_epi8 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epu16 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_min((__v8hu) __V1, (__v8hu) __V2);
+ return (__m128i) __builtin_ia32_pminuw128 ((__v8hi) __V1, (__v8hi) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -725,7 +725,7 @@ _mm_min_epu16 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epu16 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_max((__v8hu) __V1, (__v8hu) __V2);
+ return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi) __V1, (__v8hi) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -744,7 +744,7 @@ _mm_max_epu16 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epi32 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_min((__v4si) __V1, (__v4si) __V2);
+ return (__m128i) __builtin_ia32_pminsd128 ((__v4si) __V1, (__v4si) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -763,7 +763,7 @@ _mm_min_epi32 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epi32 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_max((__v4si) __V1, (__v4si) __V2);
+ return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si) __V1, (__v4si) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -782,7 +782,7 @@ _mm_max_epi32 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_min_epu32 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_min((__v4su) __V1, (__v4su) __V2);
+ return (__m128i) __builtin_ia32_pminud128((__v4si) __V1, (__v4si) __V2);
}
/// Compares the corresponding elements of two 128-bit vectors of
@@ -801,7 +801,7 @@ _mm_min_epu32 (__m128i __V1, __m128i __V2)
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_max_epu32 (__m128i __V1, __m128i __V2)
{
- return (__m128i) __builtin_elementwise_max((__v4su) __V1, (__v4su) __V2);
+ return (__m128i) __builtin_ia32_pmaxud128((__v4si) __V1, (__v4si) __V2);
}
/* SSE4 Insertion and Extraction from XMM Register Instructions. */
diff --git a/clang/test/CodeGen/builtins-x86.c b/clang/test/CodeGen/builtins-x86.c
index 9eb5f2f5d149..bfcd30072fc1 100644
--- a/clang/test/CodeGen/builtins-x86.c
+++ b/clang/test/CodeGen/builtins-x86.c
@@ -221,6 +221,10 @@ void f0() {
tmp_V16c = __builtin_ia32_psubusb128(tmp_V16c, tmp_V16c);
tmp_V8s = __builtin_ia32_psubusw128(tmp_V8s, tmp_V8s);
tmp_V8s = __builtin_ia32_pmulhw128(tmp_V8s, tmp_V8s);
+ tmp_V16c = __builtin_ia32_pmaxub128(tmp_V16c, tmp_V16c);
+ tmp_V8s = __builtin_ia32_pmaxsw128(tmp_V8s, tmp_V8s);
+ tmp_V16c = __builtin_ia32_pminub128(tmp_V16c, tmp_V16c);
+ tmp_V8s = __builtin_ia32_pminsw128(tmp_V8s, tmp_V8s);
tmp_V16c = __builtin_ia32_packsswb128(tmp_V8s, tmp_V8s);
tmp_V8s = __builtin_ia32_packssdw128(tmp_V4i, tmp_V4i);
tmp_V16c = __builtin_ia32_packuswb128(tmp_V8s, tmp_V8s);
@@ -451,6 +455,14 @@ void f0() {
tmp_V2d = __builtin_ia32_blendvpd(tmp_V2d, tmp_V2d, tmp_V2d);
tmp_V4f = __builtin_ia32_blendvps(tmp_V4f, tmp_V4f, tmp_V4f);
tmp_V8s = __builtin_ia32_packusdw128(tmp_V4i, tmp_V4i);
+ tmp_V16c = __builtin_ia32_pmaxsb128(tmp_V16c, tmp_V16c);
+ tmp_V4i = __builtin_ia32_pmaxsd128(tmp_V4i, tmp_V4i);
+ tmp_V4i = __builtin_ia32_pmaxud128(tmp_V4i, tmp_V4i);
+ tmp_V8s = __builtin_ia32_pmaxuw128(tmp_V8s, tmp_V8s);
+ tmp_V16c = __builtin_ia32_pminsb128(tmp_V16c, tmp_V16c);
+ tmp_V4i = __builtin_ia32_pminsd128(tmp_V4i, tmp_V4i);
+ tmp_V4i = __builtin_ia32_pminud128(tmp_V4i, tmp_V4i);
+ tmp_V8s = __builtin_ia32_pminuw128(tmp_V8s, tmp_V8s);
tmp_V2LLi = __builtin_ia32_pmuldq128(tmp_V4i, tmp_V4i);
tmp_V4f = __builtin_ia32_roundps(tmp_V4f, imm_i_0_16);
tmp_V4f = __builtin_ia32_roundss(tmp_V4f, tmp_V4f, imm_i_0_16);
More information about the cfe-commits
mailing list