[libc][reland] Fix builtin definition for memory functions

[michaelrj]

Summary

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.

[michaelrj]

This is a reland of https://reviews.llvm.org/D137868 which was reverted for causing macro redefinition errors. This version prevents that issue, although it still has warnings about unused parameters and unused type aliases due to the empty macros making their functions no-ops.

[gchatelet]

Thx!

[michaelrj]

make the conditions match the ones for including the intrinsic headers.

libc/cmake/modules/LLVMLibCCheckCpuFeatures.cmake

 # ------------------------------------------------------------------------------
 # Cpu features definition and flags
 # ------------------------------------------------------------------------------
 
 # Initialize ALL_CPU_FEATURES as empty list.
 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)
 endif()
 
 # Making sure ALL_CPU_FEATURES is sorted.
 list(SORT ALL_CPU_FEATURES)
 

libc/src/string/CMakeLists.txt

add_implementation(bcmp ${bcmp_name}
     ${ARGN}
   )
 endfunction()
 
 if(${LIBC_TARGET_ARCHITECTURE_IS_X86})
   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()
   add_bcmp(bcmp_opt_host          COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_NATIVE})
   add_bcmp(bcmp)
 endif()
 
 # ------------------------------------------------------------------------------
add_implementation(memcmp ${memcmp_name}
     ${ARGN}
   )
 endfunction()
 
 if(${LIBC_TARGET_ARCHITECTURE_IS_X86})
   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})
   add_memcmp(memcmp_opt_host          COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_NATIVE})
   add_memcmp(memcmp)
 else()
   add_memcmp(memcmp_opt_host          COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_NATIVE})
   add_memcmp(memcmp)

libc/src/string/memory_utils/op_x86.h

 #include "src/__support/architectures.h"
 
 #if defined(LLVM_LIBC_ARCH_X86_64)
 
 #include "src/__support/common.h"
 #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(_MSC_VER) || defined(__SCE__))
 #define _mm512_cmpneq_epi8_mask(A, B) 0
-#define _mm_movemask_epi8(A) 0
+#endif
+#if !defined(__AVX2__) && (defined(_MSC_VER) || defined(__SCE__))
 #define _mm256_movemask_epi8(A) 0
-#endif //  __SSE2__
+#endif
+#if !defined(__SSE2__) && (defined(_MSC_VER) || defined(__SCE__))
+#define _mm_movemask_epi8(A) 0
+#endif
 
 namespace __llvm_libc::x86 {
 
 // A set of constants to check compile time features.
 static inline constexpr bool kSse2 = LLVM_LIBC_IS_DEFINED(__SSE2__);
 static inline constexpr bool kAvx = LLVM_LIBC_IS_DEFINED(__AVX__);
 static inline constexpr bool kAvx2 = LLVM_LIBC_IS_DEFINED(__AVX2__);
 static inline constexpr bool kAvx512F = LLVM_LIBC_IS_DEFINED(__AVX512F__);