https://github.com/phoebewang created https://github.com/llvm/llvm-project/pull/104781
E.g.: https://godbolt.org/z/G8zK5svjK Based on Evgenii's work. >From 9a0c15c9b0bb3d1df3902dcfe62d659803cba516 Mon Sep 17 00:00:00 2001 From: "Wang, Phoebe" <phoebe.w...@intel.com> Date: Mon, 19 Aug 2024 22:09:13 +0800 Subject: [PATCH] [X86][AVX10] Fix unexpected error and warning when using intrinsic E.g.: https://godbolt.org/z/G8zK5svjK Based on Evgenii's work. --- clang/lib/Basic/Targets/X86.cpp | 6 +-- clang/lib/Headers/avx2intrin.h | 9 +++++ clang/lib/Headers/avxintrin.h | 9 +++++ clang/lib/Headers/emmintrin.h | 6 +++ clang/lib/Headers/gfniintrin.h | 42 +++++++++++++++----- clang/lib/Headers/mmintrin.h | 6 +++ clang/lib/Headers/pmmintrin.h | 6 +++ clang/lib/Headers/smmintrin.h | 6 +++ clang/lib/Headers/tmmintrin.h | 6 +++ clang/lib/Headers/xmmintrin.h | 9 +++++ clang/test/CodeGen/X86/avx512vlbw-builtins.c | 1 + 11 files changed, 93 insertions(+), 13 deletions(-) diff --git a/clang/lib/Basic/Targets/X86.cpp b/clang/lib/Basic/Targets/X86.cpp index 3fb3587eb59140..a9cbdb7b10dff8 100644 --- a/clang/lib/Basic/Targets/X86.cpp +++ b/clang/lib/Basic/Targets/X86.cpp @@ -186,14 +186,14 @@ bool X86TargetInfo::initFeatureMap( llvm::append_range(UpdatedFeaturesVec, UpdatedAVX10FeaturesVec); // HasEVEX512 is a three-states flag. We need to turn it into [+-]evex512 // according to other features. - if (HasAVX512F) { + if (!HasAVX10_512 && HasAVX512F) { UpdatedFeaturesVec.push_back(HasEVEX512 == FE_FALSE ? "-evex512" : "+evex512"); - if (HasAVX10 && !HasAVX10_512 && HasEVEX512 != FE_FALSE) + if (HasAVX10 && HasEVEX512 != FE_FALSE) Diags.Report(diag::warn_invalid_feature_combination) << LastAVX512 + " " + LastAVX10 + "; will be promoted to avx10.1-512"; } else if (HasAVX10) { - if (HasEVEX512 != FE_NOSET) + if (!HasAVX512F && HasEVEX512 != FE_NOSET) Diags.Report(diag::warn_invalid_feature_combination) << LastAVX10 + (HasEVEX512 == FE_TRUE ? " +evex512" : " -evex512"); UpdatedFeaturesVec.push_back(HasAVX10_512 ? "+evex512" : "-evex512"); diff --git a/clang/lib/Headers/avx2intrin.h b/clang/lib/Headers/avx2intrin.h index 096cae01b57d01..dc9fc073143236 100644 --- a/clang/lib/Headers/avx2intrin.h +++ b/clang/lib/Headers/avx2intrin.h @@ -15,12 +15,21 @@ #define __AVX2INTRIN_H /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS256 \ __attribute__((__always_inline__, __nodebug__, \ __target__("avx2,no-evex512"), __min_vector_width__(256))) #define __DEFAULT_FN_ATTRS128 \ __attribute__((__always_inline__, __nodebug__, \ __target__("avx2,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS256 \ + __attribute__((__always_inline__, __nodebug__, __target__("avx2"), \ + __min_vector_width__(256))) +#define __DEFAULT_FN_ATTRS128 \ + __attribute__((__always_inline__, __nodebug__, __target__("avx2"), \ + __min_vector_width__(128))) +#endif /* SSE4 Multiple Packed Sums of Absolute Difference. */ /// Computes sixteen sum of absolute difference (SAD) operations on sets of diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h index 4983f331137001..73707c623065e7 100644 --- a/clang/lib/Headers/avxintrin.h +++ b/clang/lib/Headers/avxintrin.h @@ -50,12 +50,21 @@ typedef __bf16 __m256bh __attribute__((__vector_size__(32), __aligned__(32))); #endif /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, __target__("avx,no-evex512"), \ __min_vector_width__(256))) #define __DEFAULT_FN_ATTRS128 \ __attribute__((__always_inline__, __nodebug__, __target__("avx,no-evex512"), \ __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("avx"), \ + __min_vector_width__(256))) +#define __DEFAULT_FN_ATTRS128 \ + __attribute__((__always_inline__, __nodebug__, __target__("avx"), \ + __min_vector_width__(128))) +#endif /* Arithmetic */ /// Adds two 256-bit vectors of [4 x double]. diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h index a3176570a468f7..e607ccd19d42b1 100644 --- a/clang/lib/Headers/emmintrin.h +++ b/clang/lib/Headers/emmintrin.h @@ -49,9 +49,15 @@ typedef __bf16 __m128bh __attribute__((__vector_size__(16), __aligned__(16))); #endif /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, \ __target__("sse2,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("sse2"), \ + __min_vector_width__(128))) +#endif #define __trunc64(x) \ (__m64) __builtin_shufflevector((__v2di)(x), __extension__(__v2di){}, 0) diff --git a/clang/lib/Headers/gfniintrin.h b/clang/lib/Headers/gfniintrin.h index 73b04a824aba8e..9a5743d4b673bc 100644 --- a/clang/lib/Headers/gfniintrin.h +++ b/clang/lib/Headers/gfniintrin.h @@ -14,6 +14,7 @@ #ifndef __GFNIINTRIN_H #define __GFNIINTRIN_H +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) /* Default attributes for simple form (no masking). */ #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, \ @@ -25,26 +26,47 @@ __target__("avx,gfni,no-evex512"), \ __min_vector_width__(256))) -/* Default attributes for ZMM unmasked forms. */ -#define __DEFAULT_FN_ATTRS_Z \ +/* Default attributes for VLX masked forms. */ +#define __DEFAULT_FN_ATTRS_VL128 \ __attribute__((__always_inline__, __nodebug__, \ - __target__("avx512f,evex512,gfni"), \ - __min_vector_width__(512))) -/* Default attributes for ZMM masked forms. */ -#define __DEFAULT_FN_ATTRS_Z_MASK \ + __target__("avx512bw,avx512vl,gfni,no-evex512"), \ + __min_vector_width__(128))) +#define __DEFAULT_FN_ATTRS_VL256 \ __attribute__((__always_inline__, __nodebug__, \ - __target__("avx512bw,evex512,gfni"), \ - __min_vector_width__(512))) + __target__("avx512bw,avx512vl,gfni,no-evex512"), \ + __min_vector_width__(256))) +#else +/* Default attributes for simple form (no masking). */ +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("gfni"), \ + __min_vector_width__(128))) + +/* Default attributes for YMM unmasked form. */ +#define __DEFAULT_FN_ATTRS_Y \ + __attribute__((__always_inline__, __nodebug__, __target__("avx,gfni"), \ + __min_vector_width__(256))) /* Default attributes for VLX masked forms. */ #define __DEFAULT_FN_ATTRS_VL128 \ __attribute__((__always_inline__, __nodebug__, \ - __target__("avx512bw,avx512vl,gfni,no-evex512"), \ + __target__("avx512bw,avx512vl,gfni"), \ __min_vector_width__(128))) #define __DEFAULT_FN_ATTRS_VL256 \ __attribute__((__always_inline__, __nodebug__, \ - __target__("avx512bw,avx512vl,gfni,no-evex512"), \ + __target__("avx512bw,avx512vl,gfni"), \ __min_vector_width__(256))) +#endif + +/* Default attributes for ZMM unmasked forms. */ +#define __DEFAULT_FN_ATTRS_Z \ + __attribute__((__always_inline__, __nodebug__, \ + __target__("avx512f,evex512,gfni"), \ + __min_vector_width__(512))) +/* Default attributes for ZMM masked forms. */ +#define __DEFAULT_FN_ATTRS_Z_MASK \ + __attribute__((__always_inline__, __nodebug__, \ + __target__("avx512bw,evex512,gfni"), \ + __min_vector_width__(512))) #define _mm_gf2p8affineinv_epi64_epi8(A, B, I) \ ((__m128i)__builtin_ia32_vgf2p8affineinvqb_v16qi((__v16qi)(__m128i)(A), \ diff --git a/clang/lib/Headers/mmintrin.h b/clang/lib/Headers/mmintrin.h index 9d1e135be63be6..0347c5ccf8254a 100644 --- a/clang/lib/Headers/mmintrin.h +++ b/clang/lib/Headers/mmintrin.h @@ -39,9 +39,15 @@ typedef short __v8hi __attribute__((__vector_size__(16))); typedef char __v16qi __attribute__((__vector_size__(16))); /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS_SSE2 \ __attribute__((__always_inline__, __nodebug__, \ __target__("sse2,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS_SSE2 \ + __attribute__((__always_inline__, __nodebug__, __target__("sse2"), \ + __min_vector_width__(128))) +#endif #define __trunc64(x) \ (__m64) __builtin_shufflevector((__v2di)(x), __extension__(__v2di){}, 0) diff --git a/clang/lib/Headers/pmmintrin.h b/clang/lib/Headers/pmmintrin.h index 91cee1edda3067..9ad76579668b35 100644 --- a/clang/lib/Headers/pmmintrin.h +++ b/clang/lib/Headers/pmmintrin.h @@ -17,9 +17,15 @@ #include <emmintrin.h> /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, \ __target__("sse3,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("sse3"), \ + __min_vector_width__(128))) +#endif /// Loads data from an unaligned memory location to elements in a 128-bit /// vector. diff --git a/clang/lib/Headers/smmintrin.h b/clang/lib/Headers/smmintrin.h index b3fec474e35a1e..bc6fe4c801d7ea 100644 --- a/clang/lib/Headers/smmintrin.h +++ b/clang/lib/Headers/smmintrin.h @@ -17,9 +17,15 @@ #include <tmmintrin.h> /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, \ __target__("sse4.1,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("sse4.1"), \ + __min_vector_width__(128))) +#endif /* SSE4 Rounding macros. */ #define _MM_FROUND_TO_NEAREST_INT 0x00 diff --git a/clang/lib/Headers/tmmintrin.h b/clang/lib/Headers/tmmintrin.h index bd832ce8dddfdf..371cc82e3dc9da 100644 --- a/clang/lib/Headers/tmmintrin.h +++ b/clang/lib/Headers/tmmintrin.h @@ -17,9 +17,15 @@ #include <pmmintrin.h> /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, \ __target__("ssse3,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("ssse3"), \ + __min_vector_width__(128))) +#endif #define __trunc64(x) \ (__m64) __builtin_shufflevector((__v2di)(x), __extension__(__v2di){}, 0) diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h index 5b9e90e8f061c1..158cd1a98e451a 100644 --- a/clang/lib/Headers/xmmintrin.h +++ b/clang/lib/Headers/xmmintrin.h @@ -32,12 +32,21 @@ typedef unsigned int __v4su __attribute__((__vector_size__(16))); #endif /* Define the default attributes for the functions in this file. */ +#if defined(__EVEX512__) && !defined(__AVX10_1_512__) #define __DEFAULT_FN_ATTRS \ __attribute__((__always_inline__, __nodebug__, __target__("sse,no-evex512"), \ __min_vector_width__(128))) #define __DEFAULT_FN_ATTRS_SSE2 \ __attribute__((__always_inline__, __nodebug__, \ __target__("sse2,no-evex512"), __min_vector_width__(128))) +#else +#define __DEFAULT_FN_ATTRS \ + __attribute__((__always_inline__, __nodebug__, __target__("sse"), \ + __min_vector_width__(128))) +#define __DEFAULT_FN_ATTRS_SSE2 \ + __attribute__((__always_inline__, __nodebug__, __target__("sse2"), \ + __min_vector_width__(128))) +#endif #define __trunc64(x) \ (__m64) __builtin_shufflevector((__v2di)(x), __extension__(__v2di){}, 0) diff --git a/clang/test/CodeGen/X86/avx512vlbw-builtins.c b/clang/test/CodeGen/X86/avx512vlbw-builtins.c index e2ce348d0e077f..4ec499caabf04e 100644 --- a/clang/test/CodeGen/X86/avx512vlbw-builtins.c +++ b/clang/test/CodeGen/X86/avx512vlbw-builtins.c @@ -1,5 +1,6 @@ // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512bw -target-feature +avx512vl -emit-llvm -o - -Wall -Werror -Wsign-conversion | FileCheck %s // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512bw -target-feature +avx512vl -fno-signed-char -emit-llvm -o - -Wall -Werror -Wsign-conversion | FileCheck %s +// RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx10.1-512 -emit-llvm -o - -Wall -Werror -Wsign-conversion | FileCheck %s #include <immintrin.h> _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
