[enhancement](crc&bitshuffle) use cpu instruction crc by armv8-a+crc …

…and thirdparty bshuf_untrans_bit_elem_scal func in arm
xutaoustc · Sep 5, 2023 · eae05e4 · eae05e4
1 parent a3f1548
commit eae05e4
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 0 deletions.
diff --git a/be/CMakeLists.txt b/be/CMakeLists.txt
@@ -344,6 +344,11 @@ if (USE_BTHREAD_SCANNER)
     set(CXX_COMMON_FLAGS "${CXX_COMMON_FLAGS} -DUSE_BTHREAD_SCANNER")
 endif()
 
+if ("${CMAKE_BUILD_TARGET_ARCH}" STREQUAL "aarch64")
+    message(STATUS "Add armv8-a+crc compile option for aarch64")
+    set(CXX_COMMON_FLAGS "${CXX_COMMON_FLAGS} -march=armv8-a+crc")
+endif()
+
 if (ENABLE_STACKTRACE)
     add_compile_options(-DENABLE_STACKTRACE)
 endif()

diff --git a/thirdparty/build-thirdparty.sh b/thirdparty/build-thirdparty.sh
@@ -1056,6 +1056,12 @@ build_bitshuffle() {
     cd "${TP_SOURCE_DIR}/${BITSHUFFLE_SOURCE}"
     PREFIX="${TP_INSTALL_DIR}"
 
+    echo "Before change NEON to scal:"
+    grep -n "count = bshuf_untrans_bit_elem_NEON\|count = bshuf_untrans_bit_elem_scal" src/bitshuffle_core.c
+    sed -i 's/count = bshuf_untrans_bit_elem_NEON/count = bshuf_untrans_bit_elem_scal/g' src/bitshuffle_core.c
+    echo "After change NEON to scal:"
+    grep -n "count = bshuf_untrans_bit_elem_NEON\|count = bshuf_untrans_bit_elem_scal" src/bitshuffle_core.c
+
     # This library has significant optimizations when built with AVX2/AVX512. However,
     # we still need to support non-AVX2-capable hardware. So, we build it three times,
     # with the flag AVX2, AVX512 each and once without, and use some linker tricks to