Merge #20603: build: Update crc32c subtree

90c0f267bdedc261d8fdab188e96ca58c206652a Squashed 'src/crc32c/' changes from 224988680f..b5ef9be675 (MarcoFalke)

Pull request description:

  Except for the ARM64 darwin fix this is just code-shuffling in files/functions we don't use

ACKs for top commit:
  jonasschnelli:
    Tested ACK fa7c8d136f6590e54d60c37fb34ebec8da84ebbb - Tested this on an ARM Mac. Linking issue went away (successful depends compilation). Also tested that the ARM64 hardware acceleration code part was used.
  laanwj:
    Code review ACK fa7c8d136f6590e54d60c37fb34ebec8da84ebbb

Tree-SHA512: 1fa156d72c75d22ead2677b165e566978331f795d52a637e478d83d1cf2adddd84eed259d617df6d11270af2e4e57ae6991aec3bc4c0bdf5dec959f44daa14eb

src/Makefile.crc32c.include

@@ -41,7 +41,7 @@ crc32c_libcrc32c_a_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS)
 crc32c_libcrc32c_a_SOURCES =
 crc32c_libcrc32c_a_SOURCES += crc32c/include/crc32c/crc32c.h
 crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_arm64.h
-crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_arm64_linux_check.h
+crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_arm64_check.h
 crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_internal.h
 crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_prefetch.h
 crc32c_libcrc32c_a_SOURCES += crc32c/src/crc32c_read_le.h

src/crc32c/.appveyor.yml

@@ -8,9 +8,9 @@ environment:
   matrix:
     # AppVeyor currently has no custom job name feature.
     # http://help.appveyor.com/discussions/questions/1623-can-i-provide-a-friendly-name-for-jobs
-    - JOB: Visual Studio 2017
+    - JOB: Visual Studio 2019
-      APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
+      APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2019
-      CMAKE_GENERATOR: Visual Studio 15 2017
+      CMAKE_GENERATOR: Visual Studio 16 2019
 
 platform:
   - x86
@@ -24,10 +24,11 @@ build_script:
   - git submodule update --init --recursive
   - mkdir build
   - cd build
-  - if "%platform%"=="x64" set CMAKE_GENERATOR=%CMAKE_GENERATOR% Win64
+  - if "%platform%"=="x86" (set CMAKE_GENERATOR_PLATFORM="Win32")
+      else (set CMAKE_GENERATOR_PLATFORM="%platform%")
   - cmake --version
-  - cmake .. -G "%CMAKE_GENERATOR%" -DCRC32C_USE_GLOG=0
+  - cmake .. -G "%CMAKE_GENERATOR%" -A "%CMAKE_GENERATOR_PLATFORM%"
-      -DCMAKE_CONFIGURATION_TYPES="%CONFIGURATION%"
+      -DCMAKE_CONFIGURATION_TYPES="%CONFIGURATION%" -DCRC32C_USE_GLOG=0
   - cmake --build . --config "%CONFIGURATION%"
   - cd ..
 

src/crc32c/AUTHORS

@@ -7,3 +7,5 @@ Google Inc.
 
 Fangming Fang <Fangming.Fang@arm.com>
 Vadim Skipin <vadim.skipin@gmail.com>
+Rodrigo Tobar <rtobar@icrar.org>
+Harry Mallon <hjmallon@gmail.com>

src/crc32c/CMakeLists.txt

@@ -5,15 +5,21 @@
 cmake_minimum_required(VERSION 3.1)
 project(Crc32c VERSION 1.1.0 LANGUAGES C CXX)
 
+# C standard can be overridden when this is used as a sub-project.
+if(NOT CMAKE_C_STANDARD)
   # This project can use C11, but will gracefully decay down to C89.
   set(CMAKE_C_STANDARD 11)
   set(CMAKE_C_STANDARD_REQUIRED OFF)
   set(CMAKE_C_EXTENSIONS OFF)
+endif(NOT CMAKE_C_STANDARD)
 
+# C++ standard can be overridden when this is used as a sub-project.
+if(NOT CMAKE_CXX_STANDARD)
   # This project requires C++11.
   set(CMAKE_CXX_STANDARD 11)
   set(CMAKE_CXX_STANDARD_REQUIRED ON)
   set(CMAKE_CXX_EXTENSIONS OFF)
+endif(NOT CMAKE_CXX_STANDARD)
 
 # https://github.com/izenecloud/cmake/blob/master/SetCompilerWarningAll.cmake
 if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
@@ -269,7 +275,7 @@ target_sources(crc32c
   PRIVATE
     "${PROJECT_BINARY_DIR}/include/crc32c/crc32c_config.h"
     "src/crc32c_arm64.h"
-    "src/crc32c_arm64_linux_check.h"
+    "src/crc32c_arm64_check.h"
     "src/crc32c_internal.h"
     "src/crc32c_portable.cc"
     "src/crc32c_prefetch.h"
@@ -405,19 +411,24 @@ if(CRC32C_INSTALL)
   )
 
   include(CMakePackageConfigHelpers)
+  configure_package_config_file(
+    "${PROJECT_NAME}Config.cmake.in"
+    "${PROJECT_BINARY_DIR}/${PROJECT_NAME}Config.cmake"
+    INSTALL_DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/${PROJECT_NAME}"
+  )
   write_basic_package_version_file(
-      "${PROJECT_BINARY_DIR}/Crc32cConfigVersion.cmake"
+    "${PROJECT_BINARY_DIR}/${PROJECT_NAME}ConfigVersion.cmake"
     COMPATIBILITY SameMajorVersion
   )
   install(
     EXPORT Crc32cTargets
     NAMESPACE Crc32c::
-    DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/Crc32c"
+    DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/${PROJECT_NAME}"
   )
   install(
     FILES
-      "Crc32cConfig.cmake"
+      "${PROJECT_BINARY_DIR}/${PROJECT_NAME}Config.cmake"
-      "${PROJECT_BINARY_DIR}/Crc32cConfigVersion.cmake"
+      "${PROJECT_BINARY_DIR}/${PROJECT_NAME}ConfigVersion.cmake"
-    DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/Crc32c"
+    DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/${PROJECT_NAME}"
   )
 endif(CRC32C_INSTALL)

src/crc32c/Crc32cConfig.cmake.in

@@ -2,4 +2,8 @@
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file. See the AUTHORS file for names of contributors.
 
+@PACKAGE_INIT@
+
 include("${CMAKE_CURRENT_LIST_DIR}/Crc32cTargets.cmake")
+
+check_required_components(Crc32c)

src/crc32c/src/crc32c.cc

@@ -8,7 +8,7 @@
 #include <cstdint>
 
 #include "./crc32c_arm64.h"
-#include "./crc32c_arm64_linux_check.h"
+#include "./crc32c_arm64_check.h"
 #include "./crc32c_internal.h"
 #include "./crc32c_sse42.h"
 #include "./crc32c_sse42_check.h"
@@ -20,8 +20,8 @@ uint32_t Extend(uint32_t crc, const uint8_t* data, size_t count) {
   static bool can_use_sse42 = CanUseSse42();
   if (can_use_sse42) return ExtendSse42(crc, data, count);
 #elif HAVE_ARM64_CRC32C
-  static bool can_use_arm_linux = CanUseArm64Linux();
+  static bool can_use_arm64_crc32 = CanUseArm64Crc32();
-  if (can_use_arm_linux) return ExtendArm64(crc, data, count);
+  if (can_use_arm64_crc32) return ExtendArm64(crc, data, count);
 #endif  // HAVE_SSE42 && (defined(_M_X64) || defined(__x86_64__))
 
   return ExtendPortable(crc, data, count);

src/crc32c/src/crc32c_arm64.cc

@@ -64,7 +64,7 @@
 
 namespace crc32c {
 
-uint32_t ExtendArm64(uint32_t crc, const uint8_t *buf, size_t size) {
+uint32_t ExtendArm64(uint32_t crc, const uint8_t *data, size_t size) {
   int64_t length = size;
   uint32_t crc0, crc1, crc2, crc3;
   uint64_t t0, t1, t2;
@@ -74,7 +74,6 @@ uint32_t ExtendArm64(uint32_t crc, const uint8_t *buf, size_t size) {
   const poly64_t k0 = 0x8d96551c, k1 = 0xbd6f81f8, k2 = 0xdcb17aa4;
 
   crc = crc ^ kCRC32Xor;
-  const uint8_t *p = reinterpret_cast<const uint8_t *>(buf);
 
   while (length >= KBYTES) {
     crc0 = crc;
@@ -83,14 +82,14 @@ uint32_t ExtendArm64(uint32_t crc, const uint8_t *buf, size_t size) {
     crc3 = 0;
 
     // Process 1024 bytes in parallel.
-    CRC32C1024BYTES(p);
+    CRC32C1024BYTES(data);
 
     // Merge the 4 partial CRC32C values.
     t2 = (uint64_t)vmull_p64(crc2, k2);
     t1 = (uint64_t)vmull_p64(crc1, k1);
     t0 = (uint64_t)vmull_p64(crc0, k0);
-    crc = __crc32cd(crc3, *(uint64_t *)p);
+    crc = __crc32cd(crc3, *(uint64_t *)data);
-    p += sizeof(uint64_t);
+    data += sizeof(uint64_t);
     crc ^= __crc32cd(0, t2);
     crc ^= __crc32cd(0, t1);
     crc ^= __crc32cd(0, t0);
@@ -99,23 +98,23 @@ uint32_t ExtendArm64(uint32_t crc, const uint8_t *buf, size_t size) {
   }
 
   while (length >= 8) {
-    crc = __crc32cd(crc, *(uint64_t *)p);
+    crc = __crc32cd(crc, *(uint64_t *)data);
-    p += 8;
+    data += 8;
     length -= 8;
   }
 
   if (length & 4) {
-    crc = __crc32cw(crc, *(uint32_t *)p);
+    crc = __crc32cw(crc, *(uint32_t *)data);
-    p += 4;
+    data += 4;
   }
 
   if (length & 2) {
-    crc = __crc32ch(crc, *(uint16_t *)p);
+    crc = __crc32ch(crc, *(uint16_t *)data);
-    p += 2;
+    data += 2;
   }
 
   if (length & 1) {
-    crc = __crc32cb(crc, *p);
+    crc = __crc32cb(crc, *data);
   }
 
   return crc ^ kCRC32Xor;

src/crc32c/src/crc32c_arm64.h

@@ -2,10 +2,10 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
-// Linux-specific code checking the availability for ARM CRC32C instructions.
+// ARM-specific code
 
-#ifndef CRC32C_CRC32C_ARM_LINUX_H_
+#ifndef CRC32C_CRC32C_ARM_H_
-#define CRC32C_CRC32C_ARM_LINUX_H_
+#define CRC32C_CRC32C_ARM_H_
 
 #include <cstddef>
 #include <cstdint>
@@ -24,4 +24,4 @@ uint32_t ExtendArm64(uint32_t crc, const uint8_t* data, size_t count);
 
 #endif  // HAVE_ARM64_CRC32C
 
-#endif  // CRC32C_CRC32C_ARM_LINUX_H_
+#endif  // CRC32C_CRC32C_ARM_H_

src/crc32c/src/crc32c_arm64_check.h

@@ -2,12 +2,10 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
-// ARM Linux-specific code checking for the availability of CRC32C instructions.
+// ARM-specific code checking for the availability of CRC32C instructions.
 
-#ifndef CRC32C_CRC32C_ARM_LINUX_CHECK_H_
+#ifndef CRC32C_CRC32C_ARM_CHECK_H_
-#define CRC32C_CRC32C_ARM_LINUX_CHECK_H_
+#define CRC32C_CRC32C_ARM_CHECK_H_
-
-// X86-specific code checking for the availability of SSE4.2 instructions.
 
 #include <cstddef>
 #include <cstdint>
@@ -18,6 +16,7 @@
 
 #if HAVE_ARM64_CRC32C
 
+#ifdef __linux__
 #if HAVE_STRONG_GETAUXVAL
 #include <sys/auxv.h>
 #elif HAVE_WEAK_GETAUXVAL
@@ -27,17 +26,28 @@ extern "C" unsigned long getauxval(unsigned long type) __attribute__((weak));
 
 #define AT_HWCAP 16
 #endif  // HAVE_STRONG_GETAUXVAL || HAVE_WEAK_GETAUXVAL
+#endif  // defined (__linux__)
+
+#ifdef __APPLE__
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif  // defined (__APPLE__)
 
 namespace crc32c {
 
-inline bool CanUseArm64Linux() {
+inline bool CanUseArm64Crc32() {
-#if HAVE_STRONG_GETAUXVAL || HAVE_WEAK_GETAUXVAL
+#if defined (__linux__) && (HAVE_STRONG_GETAUXVAL || HAVE_WEAK_GETAUXVAL)
   // From 'arch/arm64/include/uapi/asm/hwcap.h' in Linux kernel source code.
   constexpr unsigned long kHWCAP_PMULL = 1 << 4;
   constexpr unsigned long kHWCAP_CRC32 = 1 << 7;
   unsigned long hwcap = (&getauxval != nullptr) ? getauxval(AT_HWCAP) : 0;
   return (hwcap & (kHWCAP_PMULL | kHWCAP_CRC32)) ==
          (kHWCAP_PMULL | kHWCAP_CRC32);
+#elif defined(__APPLE__)
+  int val = 0;
+  size_t len = sizeof(val);
+  return sysctlbyname("hw.optional.armv8_crc32", &val, &len, nullptr, 0) == 0
+             && val != 0;
 #else
   return false;
 #endif  // HAVE_STRONG_GETAUXVAL || HAVE_WEAK_GETAUXVAL
@@ -47,4 +57,4 @@ inline bool CanUseArm64Linux() {
 
 #endif  // HAVE_ARM64_CRC32C
 
-#endif  // CRC32C_CRC32C_ARM_LINUX_CHECK_H_
+#endif  // CRC32C_CRC32C_ARM_CHECK_H_

src/crc32c/src/crc32c_benchmark.cc

@@ -16,7 +16,7 @@
 #endif  // CRC32C_TESTS_BUILT_WITH_GLOG
 
 #include "./crc32c_arm64.h"
-#include "./crc32c_arm64_linux_check.h"
+#include "./crc32c_arm64_check.h"
 #include "./crc32c_internal.h"
 #include "./crc32c_sse42.h"
 #include "./crc32c_sse42_check.h"
@@ -58,8 +58,8 @@ BENCHMARK_REGISTER_F(CRC32CBenchmark, Portable)
 
 #if HAVE_ARM64_CRC32C
 
-BENCHMARK_DEFINE_F(CRC32CBenchmark, ArmLinux)(benchmark::State& state) {
+BENCHMARK_DEFINE_F(CRC32CBenchmark, ArmCRC32C)(benchmark::State& state) {
-  if (!crc32c::CanUseArm64Linux()) {
+  if (!crc32c::CanUseArm64Crc32()) {
     state.SkipWithError("ARM CRC32C instructions not available or not enabled");
     return;
   }
@@ -69,7 +69,7 @@ BENCHMARK_DEFINE_F(CRC32CBenchmark, ArmLinux)(benchmark::State& state) {
     crc = crc32c::ExtendArm64(crc, block_buffer_, block_size_);
   state.SetBytesProcessed(state.iterations() * block_size_);
 }
-BENCHMARK_REGISTER_F(CRC32CBenchmark, ArmLinux)
+BENCHMARK_REGISTER_F(CRC32CBenchmark, ArmCRC32C)
     ->RangeMultiplier(16)
     ->Range(256, 16777216);  // Block size.
 

src/crc32c/src/crc32c_read_le.h

@@ -32,14 +32,14 @@ inline uint32_t ReadUint32LE(const uint8_t* buffer) {
 // Reads a little-endian 64-bit integer from a 64-bit-aligned buffer.
 inline uint64_t ReadUint64LE(const uint8_t* buffer) {
 #if BYTE_ORDER_BIG_ENDIAN
-  return ((static_cast<uint32_t>(static_cast<uint8_t>(buffer[0]))) |
+  return ((static_cast<uint64_t>(static_cast<uint8_t>(buffer[0]))) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[1])) << 8) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[1])) << 8) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[2])) << 16) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[2])) << 16) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[3])) << 24) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[3])) << 24) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[4])) << 32) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[4])) << 32) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[5])) << 40) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[5])) << 40) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[6])) << 48) |
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[6])) << 48) |
-          (static_cast<uint32_t>(static_cast<uint8_t>(buffer[7])) << 56));
+          (static_cast<uint64_t>(static_cast<uint8_t>(buffer[7])) << 56));
 #else   // !BYTE_ORDER_BIG_ENDIAN
   uint64_t result;
   // This should be optimized to a single instruction.