[libc-commits] [PATCH] D138163: [libc][reland] Fix builtin definition for memory functions

[Michael Jones via Phabricator via libc-commits]

michaelrj created this revision.
michaelrj added reviewers: gchatelet, sivachandra.
Herald added subscribers: libc-commits, ecnelises, tschuett, pengfei.
Herald added projects: libc-project, All.
michaelrj requested review of this revision.

The memory functions are highly performance sensitive and use builtins
where possible, but also need to define those functions names when they
don't exist to avoid compilation errors. Previously all those
redefinitions were behind the SSE2 flag for x86, which caused errors on
CPUs that supported SSE2 but not AVX512. This patch splits the various
CPU extensions out to avoid errors on such CPUs.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D138163

Files:
  libc/cmake/modules/LLVMLibCCheckCpuFeatures.cmake
  libc/src/string/CMakeLists.txt
  libc/src/string/memory_utils/op_x86.h


Index: libc/src/string/memory_utils/op_x86.h
===================================================================
--- libc/src/string/memory_utils/op_x86.h
+++ libc/src/string/memory_utils/op_x86.h
@@ -20,15 +20,22 @@
 #include "src/string/memory_utils/op_builtin.h"
 #include "src/string/memory_utils/op_generic.h"
 
-#ifdef __SSE2__
+#if defined(__AVX512BW__) || defined(__AVX512F__) || defined(__AVX2__) ||      \
+    defined(__SSE2__)
 #include <immintrin.h>
-#else
+#endif
+
 // Define fake functions to prevent the compiler from failing on undefined
-// functions in case SSE2 is not present.
+// functions in case the CPU extension is not present.
+#if !defined(__AVX512BW__) && !defined(_mm512_cmpneq_epi8_mask)
 #define _mm512_cmpneq_epi8_mask(A, B) 0
-#define _mm_movemask_epi8(A) 0
+#endif
+#if !defined(__AVX2__) && !defined(_mm256_movemask_epi8)
 #define _mm256_movemask_epi8(A) 0
-#endif //  __SSE2__
+#endif
+#if !defined(__SSE2__) && !defined(_mm_movemask_epi8)
+#define _mm_movemask_epi8(A) 0
+#endif
 
 namespace __llvm_libc::x86 {
 
Index: libc/src/string/CMakeLists.txt
===================================================================
--- libc/src/string/CMakeLists.txt
+++ libc/src/string/CMakeLists.txt
@@ -355,7 +355,7 @@
   add_bcmp(bcmp_x86_64_opt_sse2   COMPILE_OPTIONS -march=k8             REQUIRE SSE2)
   add_bcmp(bcmp_x86_64_opt_sse4   COMPILE_OPTIONS -march=nehalem        REQUIRE SSE4_2)
   add_bcmp(bcmp_x86_64_opt_avx2   COMPILE_OPTIONS -march=haswell        REQUIRE AVX2)
-  add_bcmp(bcmp_x86_64_opt_avx512 COMPILE_OPTIONS -march=skylake-avx512 REQUIRE AVX512F)
+  add_bcmp(bcmp_x86_64_opt_avx512 COMPILE_OPTIONS -march=skylake-avx512 REQUIRE AVX512BW)
   add_bcmp(bcmp_opt_host          COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_NATIVE})
   add_bcmp(bcmp)
 else()
@@ -409,7 +409,7 @@
   add_memcmp(memcmp_x86_64_opt_sse2   COMPILE_OPTIONS -march=k8             REQUIRE SSE2)
   add_memcmp(memcmp_x86_64_opt_sse4   COMPILE_OPTIONS -march=nehalem        REQUIRE SSE4_2)
   add_memcmp(memcmp_x86_64_opt_avx2   COMPILE_OPTIONS -march=haswell        REQUIRE AVX2)
-  add_memcmp(memcmp_x86_64_opt_avx512 COMPILE_OPTIONS -march=skylake-avx512 REQUIRE AVX512F)
+  add_memcmp(memcmp_x86_64_opt_avx512 COMPILE_OPTIONS -march=skylake-avx512 REQUIRE AVX512BW)
   add_memcmp(memcmp_opt_host          COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_NATIVE})
   add_memcmp(memcmp)
 elseif(${LIBC_TARGET_ARCHITECTURE_IS_AARCH64})
Index: libc/cmake/modules/LLVMLibCCheckCpuFeatures.cmake
===================================================================
--- libc/cmake/modules/LLVMLibCCheckCpuFeatures.cmake
+++ libc/cmake/modules/LLVMLibCCheckCpuFeatures.cmake
@@ -6,7 +6,7 @@
 set(ALL_CPU_FEATURES "")
 
 if(${LIBC_TARGET_ARCHITECTURE_IS_X86})
-  set(ALL_CPU_FEATURES SSE2 SSE4_2 AVX2 AVX512F FMA)
+  set(ALL_CPU_FEATURES SSE2 SSE4_2 AVX2 AVX512F AVX512BW FMA)
   set(LIBC_COMPILE_OPTIONS_NATIVE -march=native)
 elseif(${LIBC_TARGET_ARCHITECTURE_IS_AARCH64})
   set(LIBC_COMPILE_OPTIONS_NATIVE -mcpu=native)


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D138163.475956.patch
Type: text/x-patch
Size: 3023 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/libc-commits/attachments/20221117/668aeb94/attachment.bin>