Add crc32-vpmsum to contrib

.gitmodules

@@ -290,3 +290,6 @@
 [submodule "contrib/morton-nd"]
 	path = contrib/morton-nd
 	url = https://github.com/morton-nd/morton-nd
+[submodule "contrib/crc32-vpmsum"]
+	path = contrib/crc32-vpmsum
+	url = https://github.com/antonblanchard/crc32-vpmsum.git

contrib/CMakeLists.txt

@@ -51,6 +51,7 @@ endfunction()
 
 add_contrib (miniselect-cmake miniselect)
 add_contrib (pdqsort-cmake pdqsort)
+add_contrib (crc32-vpmsum-cmake crc32-vpmsum)
 add_contrib (sparsehash-c11-cmake sparsehash-c11)
 add_contrib (abseil-cpp-cmake abseil-cpp)
 add_contrib (magic-enum-cmake magic_enum)

contrib/crc32-vpmsum-cmake/CMakeLists.txt

@@ -0,0 +1,12 @@
+if (NOT ARCH_PPC64LE)
+    message(STATUS "crc32-vpmsum library is only supported on ppc64le")
+    return()
+endif()
+
+SET(LIBRARY_DIR "${ClickHouse_SOURCE_DIR}/contrib/crc32-vpmsum")
+
+add_library(_crc32-vpmsum 
+	"${LIBRARY_DIR}/vec_crc32.c"
+	)
+target_include_directories(_crc32-vpmsum SYSTEM BEFORE PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}")
+add_library(ch_contrib::crc32-vpmsum ALIAS _crc32-vpmsum)

contrib/crc32-vpmsum-cmake/README.md

@@ -0,0 +1,8 @@
+# To Generate crc32_constants.h 
+
+- Run make file in `../crc32-vpmsum` diretory using folling options and CRC polynomial. These options should use the same polynomial and order used by intel intrinisic functions
+```bash
+make crc32_constants.h CRC="0x182f63b78" OPTIONS="-x"
+```
+- move the generated `crc32_constants.h` into this directory
+- To understand more about this go here: https://masterchef2209.wordpress.com/2020/06/17/guide-to-intel-sse4-2-crc-intrinisics-implementation-for-simde/

contrib/crc32-vpmsum-cmake/vec_crc32.h

@@ -0,0 +1,29 @@
+#ifndef VEC_CRC32
+#define VEC_CRC32
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+unsigned int crc32_vpmsum(unsigned int crc, const unsigned char *p, unsigned long len);
+
+static inline uint32_t crc32_ppc(uint64_t crc, unsigned char const *buffer, size_t len)
+{
+	unsigned char *emptybuffer;
+    if (!buffer) {
+        emptybuffer = (unsigned char *)malloc(len);
+        bzero(emptybuffer, len);
+        crc = crc32_vpmsum(crc, emptybuffer, len);
+        free(emptybuffer);
+    } else {
+        crc = crc32_vpmsum(crc, buffer, (unsigned long)len);
+    }
+	return crc;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/CMakeLists.txt

@@ -377,6 +377,7 @@ target_link_libraries(clickhouse_common_io
         PUBLIC
             boost::program_options
             boost::system
+            ch_contrib::crc32-vpmsum
             ch_contrib::cityhash
             ch_contrib::zlib
             pcg_random

src/Common/HashTable/Hash.h

@@ -48,6 +48,10 @@ inline DB::UInt64 intHash64(DB::UInt64 x)
 #include <arm_acle.h>
 #endif
 
+#if (defined(__PPC64__) || defined(__powerpc64__)) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#include "vec_crc32.h"
+#endif
+
 /// NOTE: Intel intrinsic can be confusing.
 /// - https://code.google.com/archive/p/sse-intrinsics/wikis/PmovIntrinsicBug.wiki
 /// - https://stackoverflow.com/questions/15752770/mm-crc32-u64-poorly-defined
@@ -57,6 +61,8 @@ inline DB::UInt64 intHashCRC32(DB::UInt64 x)
     return _mm_crc32_u64(-1ULL, x);
 #elif defined(__aarch64__) && defined(__ARM_FEATURE_CRC32)
     return __crc32cd(-1U, x);
+#elif (defined(__PPC64__) || defined(__powerpc64__)) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    return crc32_ppc(-1U, reinterpret_cast<const unsigned char *>(&x), sizeof(x));
 #else
     /// On other platforms we do not have CRC32. NOTE This can be confusing.
     /// NOTE: consider using intHash32()
@@ -69,6 +75,8 @@ inline DB::UInt64 intHashCRC32(DB::UInt64 x, DB::UInt64 updated_value)
     return _mm_crc32_u64(updated_value, x);
 #elif defined(__aarch64__) && defined(__ARM_FEATURE_CRC32)
     return __crc32cd(static_cast<UInt32>(updated_value), x);
+#elif (defined(__PPC64__) || defined(__powerpc64__)) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    return crc32_ppc(updated_value, reinterpret_cast<const unsigned char *>(&x), sizeof(x));
 #else
     /// On other platforms we do not have CRC32. NOTE This can be confusing.
     return intHash64(x) ^ updated_value;