rocksdb/build_tools/build_detect_platform

788 lines
27 KiB
Plaintext
Raw Normal View History

#!/usr/bin/env bash
#
# Detects OS we're compiling on and outputs a file specified by the first
# argument, which in turn gets read while processing Makefile.
#
# The output will set the following variables:
# CC C Compiler path
# CXX C++ Compiler path
# PLATFORM_LDFLAGS Linker flags
# JAVA_LDFLAGS Linker flags for RocksDBJava
# JAVA_STATIC_LDFLAGS Linker flags for RocksDBJava static build
# JAVAC_ARGS Arguments for javac
# PLATFORM_SHARED_EXT Extension for shared libraries
# PLATFORM_SHARED_LDFLAGS Flags for building shared library
# PLATFORM_SHARED_CFLAGS Flags for compiling objects for shared library
# PLATFORM_CCFLAGS C compiler flags
# PLATFORM_CXXFLAGS C++ compiler flags. Will contain:
# PLATFORM_SHARED_VERSIONED Set to 'true' if platform supports versioned
# shared libraries, empty otherwise.
# FIND Command for the find utility
# WATCH Command for the watch utility
#
# The PLATFORM_CCFLAGS and PLATFORM_CXXFLAGS might include the following:
#
# -DROCKSDB_PLATFORM_POSIX if posix-platform based
2014-02-08 03:12:30 +01:00
# -DSNAPPY if the Snappy library is present
# -DLZ4 if the LZ4 library is present
# -DZSTD if the ZSTD library is present
Adding NUMA support to db_bench tests Summary: Changes: - Adding numa_aware flag to db_bench.cc - Using numa.h library to bind memory and cpu of threads to a fixed NUMA node Result: There seems to be no significant change in the micros/op time with numa_aware enabled. I also tried this with other implementations, including a combination of pthread_setaffinity_np, sched_setaffinity and set_mempolicy methods. It'd be great if someone could point out where I'm going wrong and if we can achieve a better micors/op. Test Plan: Ran db_bench tests using following command: ./db_bench --db=/mnt/tmp --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=/mnt/tmp --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --duration=300 --benchmarks=readwhilewriting --use_existing_db=1 --num=157286400 --threads=24 --writes_per_second=10240 --numa_aware=[False/True] The tests were run in private devserver with 24 cores and the db was prepopulated using filluniquerandom test. The tests resulted in 0.145 us/op with numa_aware=False and 0.161 us/op with numa_aware=True. Reviewers: sdong, yhchiang, ljin, igor Reviewed By: ljin, igor Subscribers: igor, leveldb Differential Revision: https://reviews.facebook.net/D19353
2014-07-07 19:53:31 +02:00
# -DNUMA if the NUMA library is present
# -DTBB if the TBB library is present
Provide an allocator for new memory type to be used with RocksDB block cache (#6214) Summary: New memory technologies are being developed by various hardware vendors (Intel DCPMM is one such technology currently available). These new memory types require different libraries for allocation and management (such as PMDK and memkind). The high capacities available make it possible to provision large caches (up to several TBs in size), beyond what is achievable with DRAM. The new allocator provided in this PR uses the memkind library to allocate memory on different media. **Performance** We tested the new allocator using db_bench. - For each test, we vary the size of the block cache (relative to the size of the uncompressed data in the database). - The database is filled sequentially. Throughput is then measured with a readrandom benchmark. - We use a uniform distribution as a worst-case scenario. The plot shows throughput (ops/s) relative to a configuration with no block cache and default allocator. For all tests, p99 latency is below 500 us. ![image](https://user-images.githubusercontent.com/26400080/71108594-42479100-2178-11ea-8231-8a775bbc92db.png) **Changes** - Add MemkindKmemAllocator - Add --use_cache_memkind_kmem_allocator db_bench option (to create an LRU block cache with the new allocator) - Add detection of memkind library with KMEM DAX support - Add test for MemkindKmemAllocator **Minimum Requirements** - kernel 5.3.12 - ndctl v67 - https://github.com/pmem/ndctl - memkind v1.10.0 - https://github.com/memkind/memkind **Memory Configuration** The allocator uses the MEMKIND_DAX_KMEM memory kind. Follow the instructions on[ memkind’s GitHub page](https://github.com/memkind/memkind) to set up NVDIMM memory accordingly. Note on memory allocation with NVDIMM memory exposed as system memory. - The MemkindKmemAllocator will only allocate from NVDIMM memory (using memkind_malloc with MEMKIND_DAX_KMEM kind). - The default allocator is not restricted to RAM by default. Based on NUMA node latency, the kernel should allocate from local RAM preferentially, but it’s a kernel decision. numactl --preferred/--membind can be used to allocate preferentially/exclusively from the local RAM node. **Usage** When creating an LRU cache, pass a MemkindKmemAllocator object as argument. For example (replace capacity with the desired value in bytes): ``` #include "rocksdb/cache.h" #include "memory/memkind_kmem_allocator.h" NewLRUCache( capacity /*size_t*/, 6 /*cache_numshardbits*/, false /*strict_capacity_limit*/, false /*cache_high_pri_pool_ratio*/, std::make_shared<MemkindKmemAllocator>()); ``` Refer to [RocksDB’s block cache documentation](https://github.com/facebook/rocksdb/wiki/Block-Cache) to assign the LRU cache as block cache for a database. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6214 Reviewed By: cheng-chang Differential Revision: D19292435 fbshipit-source-id: 7202f47b769e7722b539c86c2ffd669f64d7b4e1
2020-04-10 05:45:17 +02:00
# -DMEMKIND if the memkind library is present
#
# Using gflags in rocksdb:
# Our project depends on gflags, which requires users to take some extra steps
# before they can compile the whole repository:
# 1. Install gflags. You may download it from here:
# https://gflags.github.io/gflags/ (Mac users can `brew install gflags`)
# 2. Once installed, add the include path for gflags to your CPATH env var and
# the lib path to LIBRARY_PATH. If installed with default settings, the lib
# will be /usr/local/lib and the include path will be /usr/local/include
OUTPUT=$1
if test -z "$OUTPUT"; then
echo "usage: $0 <output-filename>" >&2
exit 1
fi
# we depend on C++11
PLATFORM_CXXFLAGS="-std=c++11"
# we currently depend on POSIX platform
COMMON_FLAGS="-DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX"
# Default to fbcode gcc on internal fb machines
if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
FBCODE_BUILD="true"
# If we're compiling with TSAN we need pic build
PIC_BUILD=$COMPILE_WITH_TSAN
if [ -n "$ROCKSDB_FBCODE_BUILD_WITH_481" ]; then
# we need this to build with MySQL. Don't use for other purposes.
source "$PWD/build_tools/fbcode_config4.8.1.sh"
elif [ -n "$ROCKSDB_FBCODE_BUILD_WITH_5xx" ]; then
source "$PWD/build_tools/fbcode_config.sh"
else
source "$PWD/build_tools/fbcode_config_platform007.sh"
fi
fi
# Delete existing output, if it exists
rm -f "$OUTPUT"
touch "$OUTPUT"
if test -z "$CC"; then
if [ -x "$(command -v cc)" ]; then
CC=cc
elif [ -x "$(command -v clang)" ]; then
CC=clang
else
CC=cc
fi
fi
if test -z "$CXX"; then
if [ -x "$(command -v g++)" ]; then
CXX=g++
elif [ -x "$(command -v clang++)" ]; then
CXX=clang++
else
CXX=g++
fi
fi
# Detect OS
if test -z "$TARGET_OS"; then
TARGET_OS=`uname -s`
fi
if test -z "$TARGET_ARCHITECTURE"; then
TARGET_ARCHITECTURE=`uname -m`
fi
if test -z "$CLANG_SCAN_BUILD"; then
CLANG_SCAN_BUILD=scan-build
fi
build: do not relink every single binary just for a timestamp Summary: Prior to this change, "make check" would always waste a lot of time relinking 60+ binaries. With this change, it does that only when the generated file, util/build_version.cc, changes, and that happens only when the date changes or when the current git SHA changes. This change makes some other improvements: before, there was no rule to build a deleted util/build_version.cc. If it was somehow removed, any attempt to link a program would fail. There is no longer any need for the separate file, build_tools/build_detect_version. Its functionality is now in the Makefile. * Makefile (DEPFILES): Don't filter-out util/build_version.cc. No need, and besides, removing that dependency was wrong. (date, git_sha, gen_build_version): New helper variables. (util/build_version.cc): New rule, to create this file and update it only if it would contain new information. * build_tools/build_detect_platform: Remove file. * db/db_impl.cc: Now, print only date (not the time). * util/build_version.h (rocksdb_build_compile_time): Remove declaration. No longer used. Test Plan: - Run "make check" twice, and note that the second time no linking is performed. - Remove util/build_version.cc and ensure that any "make" command regenerates it before doing anything else. - Run this: strings librocksdb.a|grep _build_. That prints output including the following: rocksdb_build_git_date:2015-02-19 rocksdb_build_git_sha:2.8.fb-1792-g3cb6cc0 Reviewers: ljin, sdong, igor Reviewed By: igor Subscribers: dhruba Differential Revision: https://reviews.facebook.net/D33591
2015-02-19 22:11:10 +01:00
if test -z "$CLANG_ANALYZER"; then
CLANG_ANALYZER=$(command -v clang++ 2> /dev/null)
fi
if test -z "$FIND"; then
FIND=find
fi
if test -z "$WATCH"; then
WATCH=watch
fi
COMMON_FLAGS="$COMMON_FLAGS ${CFLAGS}"
CROSS_COMPILE=
PLATFORM_CCFLAGS=
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS"
PLATFORM_SHARED_EXT="so"
PLATFORM_SHARED_LDFLAGS="-Wl,--no-as-needed -shared -Wl,-soname -Wl,"
PLATFORM_SHARED_CFLAGS="-fPIC"
PLATFORM_SHARED_VERSIONED=true
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# generic port files (working on all platform by #ifdef) go directly in /port
GENERIC_PORT_FILES=`cd "$ROCKSDB_ROOT"; find port -name '*.cc' | tr "\n" " "`
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# On GCC, we pick libc's memcmp over GCC's memcmp via -fno-builtin-memcmp
case "$TARGET_OS" in
Darwin)
PLATFORM=OS_MACOSX
COMMON_FLAGS="$COMMON_FLAGS -DOS_MACOSX"
PLATFORM_SHARED_EXT=dylib
PLATFORM_SHARED_LDFLAGS="-dynamiclib -install_name "
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/darwin/darwin_specific.cc
;;
IOS)
PLATFORM=IOS
COMMON_FLAGS="$COMMON_FLAGS -DOS_MACOSX -DIOS_CROSS_COMPILE -DROCKSDB_LITE"
PLATFORM_SHARED_EXT=dylib
PLATFORM_SHARED_LDFLAGS="-dynamiclib -install_name "
CROSS_COMPILE=true
PLATFORM_SHARED_VERSIONED=
;;
Linux)
PLATFORM=OS_LINUX
COMMON_FLAGS="$COMMON_FLAGS -DOS_LINUX"
if [ -z "$USE_CLANG" ]; then
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
else
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
fi
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
if test $ROCKSDB_USE_IO_URING; then
# check for liburing
$CXX $CFLAGS -x c++ - -luring -o /dev/null 2>/dev/null <<EOF
#include <liburing.h>
int main() {
struct io_uring ring;
io_uring_queue_init(1, &ring, 0);
return 0;
}
EOF
if [ "$?" = 0 ]; then
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -luring"
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_IOURING_PRESENT"
fi
fi
if test -z "$USE_FOLLY_DISTRIBUTED_MUTEX"; then
USE_FOLLY_DISTRIBUTED_MUTEX=1
fi
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/linux/linux_specific.cc
;;
SunOS)
PLATFORM=OS_SOLARIS
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_SOLARIS -m64"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt -static-libstdc++ -static-libgcc -m64"
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/sunos/sunos_specific.cc
;;
AIX)
PLATFORM=OS_AIX
CC=gcc
COMMON_FLAGS="$COMMON_FLAGS -maix64 -pthread -fno-builtin-memcmp -D_REENTRANT -DOS_AIX -D__STDC_FORMAT_MACROS"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -pthread -lpthread -lrt -maix64 -static-libstdc++ -static-libgcc"
# PORT_FILES=port/aix/aix_specific.cc
;;
FreeBSD)
PLATFORM=OS_FREEBSD
CXX=clang++
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_FREEBSD"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/freebsd/freebsd_specific.cc
;;
GNU/kFreeBSD)
PLATFORM=OS_GNU_KFREEBSD
COMMON_FLAGS="$COMMON_FLAGS -DOS_GNU_KFREEBSD"
if [ -z "$USE_CLANG" ]; then
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
else
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
fi
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
# PORT_FILES=port/gnu_kfreebsd/gnu_kfreebsd_specific.cc
;;
NetBSD)
PLATFORM=OS_NETBSD
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_NETBSD"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lgcc_s"
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/netbsd/netbsd_specific.cc
;;
OpenBSD)
PLATFORM=OS_OPENBSD
CXX=clang++
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_OPENBSD"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -pthread"
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/openbsd/openbsd_specific.cc
FIND=gfind
WATCH=gnuwatch
;;
DragonFly)
PLATFORM=OS_DRAGONFLYBSD
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_DRAGONFLYBSD"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/dragonfly/dragonfly_specific.cc
;;
Cygwin)
PLATFORM=CYGWIN
PLATFORM_SHARED_CFLAGS=""
PLATFORM_CXXFLAGS="-std=gnu++11"
COMMON_FLAGS="$COMMON_FLAGS -DCYGWIN"
if [ -z "$USE_CLANG" ]; then
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
else
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -latomic"
fi
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
# PORT_FILES=port/linux/linux_specific.cc
;;
OS_ANDROID_CROSSCOMPILE)
PLATFORM=OS_ANDROID
COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_ANDROID -DROCKSDB_PLATFORM_POSIX"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS " # All pthread features are in the Android C library
[RocksDB] Add stacktrace signal handler Summary: This diff provides the ability to print out a stacktrace when the process receives certain signals. Currently, we enable this for the following signals (program error related): SIGILL SIGSEGV SIGBUS SIGABRT Application simply #include "util/stack_trace.h" and call leveldb::InstallStackTraceHandler() during initialization, if signal handler is needed. It's not done automatically when openning db, because it's the application(process)'s responsibility to install signal handler and some applications might already have their own (like fbcode). Sample output: Received signal 11 (Segmentation fault) #0 0x408ff0 ./signal_test() [0x408ff0] /home/haobo/rocksdb/util/signal_test.cc:4 #1 0x40827d ./signal_test() [0x40827d] /home/haobo/rocksdb/util/signal_test.cc:24 #2 0x7f8bb183172e /usr/local/fbcode/gcc-4.7.1-glibc-2.14.1/lib/libc.so.6(__libc_start_main+0x10e) [0x7f8bb183172e] ??:0 #3 0x408ebc ./signal_test() [0x408ebc] /home/engshare/third-party/src/glibc/glibc-2.14.1/glibc-2.14.1/csu/../sysdeps/x86_64/elf/start.S:113 Segmentation fault (core dumped) For each frame, we print the raw pointer, the symbol provided by backtrace_symbols (still not good enough), and the source file/line. Note that address translation is done by directly shell out to addr2line. ??:0 means addr2line fails to do the translation. Hacky, but I think it's good for now. Test Plan: signal_test.cc Reviewers: dhruba, MarkCallaghan Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D10173
2013-04-11 19:54:35 +02:00
# PORT_FILES=port/android/android.cc
CROSS_COMPILE=true
;;
*)
echo "Unknown platform!" >&2
exit 1
esac
PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS ${CXXFLAGS}"
JAVA_LDFLAGS="$PLATFORM_LDFLAGS"
JAVA_STATIC_LDFLAGS="$PLATFORM_LDFLAGS"
JAVAC_ARGS="-source 7"
if [ "$CROSS_COMPILE" = "true" -o "$FBCODE_BUILD" = "true" ]; then
# Cross-compiling; do not try any compilation tests.
# Also don't need any compilation tests if compiling on fbcode
if [ "$FBCODE_BUILD" = "true" ]; then
# Enable backtrace on fbcode since the necessary libraries are present
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
fi
2014-01-06 20:53:19 +01:00
true
else
if ! test $ROCKSDB_DISABLE_FALLOCATE; then
# Test whether fallocate is available
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <fcntl.h>
#include <linux/falloc.h>
int main() {
int fd = open("/dev/null", 0);
fallocate(fd, FALLOC_FL_KEEP_SIZE, 0, 1024);
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_FALLOCATE_PRESENT"
fi
fi
if ! test $ROCKSDB_DISABLE_SNAPPY; then
# Test whether Snappy library is installed
# http://code.google.com/p/snappy/
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <snappy.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DSNAPPY"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lsnappy"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lsnappy"
fi
fi
if ! test $ROCKSDB_DISABLE_GFLAGS; then
# Test whether gflags library is installed
# http://gflags.github.io/gflags/
# check if the namespace is gflags
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
#include <gflags/gflags.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=1"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
else
# check if namespace is google
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null << EOF
#include <gflags/gflags.h>
using namespace google;
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DGFLAGS=google"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lgflags"
fi
fi
fi
if ! test $ROCKSDB_DISABLE_ZLIB; then
# Test whether zlib library is installed
$CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <zlib.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DZLIB"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lz"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lz"
fi
fi
if ! test $ROCKSDB_DISABLE_BZIP; then
# Test whether bzip library is installed
$CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <bzlib.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DBZIP2"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbz2"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lbz2"
fi
fi
if ! test $ROCKSDB_DISABLE_LZ4; then
# Test whether lz4 library is installed
$CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <lz4.h>
#include <lz4hc.h>
int main() {}
2014-02-08 03:12:30 +01:00
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DLZ4"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -llz4"
JAVA_LDFLAGS="$JAVA_LDFLAGS -llz4"
fi
2014-02-08 03:12:30 +01:00
fi
if ! test $ROCKSDB_DISABLE_ZSTD; then
# Test whether zstd library is installed
$CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <zstd.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DZSTD"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lzstd"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lzstd"
fi
fi
if ! test $ROCKSDB_DISABLE_NUMA; then
# Test whether numa is available
$CXX $CFLAGS -x c++ - -o /dev/null -lnuma 2>/dev/null <<EOF
#include <numa.h>
#include <numaif.h>
int main() {}
Adding NUMA support to db_bench tests Summary: Changes: - Adding numa_aware flag to db_bench.cc - Using numa.h library to bind memory and cpu of threads to a fixed NUMA node Result: There seems to be no significant change in the micros/op time with numa_aware enabled. I also tried this with other implementations, including a combination of pthread_setaffinity_np, sched_setaffinity and set_mempolicy methods. It'd be great if someone could point out where I'm going wrong and if we can achieve a better micors/op. Test Plan: Ran db_bench tests using following command: ./db_bench --db=/mnt/tmp --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=/mnt/tmp --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --duration=300 --benchmarks=readwhilewriting --use_existing_db=1 --num=157286400 --threads=24 --writes_per_second=10240 --numa_aware=[False/True] The tests were run in private devserver with 24 cores and the db was prepopulated using filluniquerandom test. The tests resulted in 0.145 us/op with numa_aware=False and 0.161 us/op with numa_aware=True. Reviewers: sdong, yhchiang, ljin, igor Reviewed By: ljin, igor Subscribers: igor, leveldb Differential Revision: https://reviews.facebook.net/D19353
2014-07-07 19:53:31 +02:00
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DNUMA"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lnuma"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lnuma"
fi
Adding NUMA support to db_bench tests Summary: Changes: - Adding numa_aware flag to db_bench.cc - Using numa.h library to bind memory and cpu of threads to a fixed NUMA node Result: There seems to be no significant change in the micros/op time with numa_aware enabled. I also tried this with other implementations, including a combination of pthread_setaffinity_np, sched_setaffinity and set_mempolicy methods. It'd be great if someone could point out where I'm going wrong and if we can achieve a better micors/op. Test Plan: Ran db_bench tests using following command: ./db_bench --db=/mnt/tmp --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --block_size=4096 --cache_size=17179869184 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=/mnt/tmp --sync=0 --disable_data_sync=1 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_grandparent_overlap_factor=10 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --perf_level=0 --duration=300 --benchmarks=readwhilewriting --use_existing_db=1 --num=157286400 --threads=24 --writes_per_second=10240 --numa_aware=[False/True] The tests were run in private devserver with 24 cores and the db was prepopulated using filluniquerandom test. The tests resulted in 0.145 us/op with numa_aware=False and 0.161 us/op with numa_aware=True. Reviewers: sdong, yhchiang, ljin, igor Reviewed By: ljin, igor Subscribers: igor, leveldb Differential Revision: https://reviews.facebook.net/D19353
2014-07-07 19:53:31 +02:00
fi
if ! test $ROCKSDB_DISABLE_TBB; then
# Test whether tbb is available
$CXX $CFLAGS $LDFLAGS -x c++ - -o /dev/null -ltbb 2>/dev/null <<EOF
#include <tbb/tbb.h>
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DTBB"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltbb"
JAVA_LDFLAGS="$JAVA_LDFLAGS -ltbb"
fi
fi
if ! test $ROCKSDB_DISABLE_JEMALLOC; then
# Test whether jemalloc is available
if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null -ljemalloc \
2>/dev/null; then
# This will enable some preprocessor identifiers in the Makefile
JEMALLOC=1
# JEMALLOC can be enabled either using the flag (like here) or by
# providing direct link to the jemalloc library
WITH_JEMALLOC_FLAG=1
# check for JEMALLOC installed with HomeBrew
if [ "$PLATFORM" == "OS_MACOSX" ]; then
if hash brew 2>/dev/null && brew ls --versions jemalloc > /dev/null; then
JEMALLOC_VER=$(brew ls --versions jemalloc | tail -n 1 | cut -f 2 -d ' ')
JEMALLOC_INCLUDE="-I/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/include"
JEMALLOC_LIB="/usr/local/Cellar/jemalloc/${JEMALLOC_VER}/lib/libjemalloc_pic.a"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS $JEMALLOC_LIB"
JAVA_STATIC_LDFLAGS="$JAVA_STATIC_LDFLAGS $JEMALLOC_LIB"
fi
fi
fi
fi
if ! test $JEMALLOC && ! test $ROCKSDB_DISABLE_TCMALLOC; then
# jemalloc is not available. Let's try tcmalloc
if echo 'int main() {}' | $CXX $CFLAGS -x c++ - -o /dev/null \
-ltcmalloc 2>/dev/null; then
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -ltcmalloc"
JAVA_LDFLAGS="$JAVA_LDFLAGS -ltcmalloc"
fi
fi
if ! test $ROCKSDB_DISABLE_MALLOC_USABLE_SIZE; then
# Test whether malloc_usable_size is available
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <malloc.h>
int main() {
size_t res = malloc_usable_size(0);
(void)res;
return 0;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_MALLOC_USABLE_SIZE"
fi
fi
Provide an allocator for new memory type to be used with RocksDB block cache (#6214) Summary: New memory technologies are being developed by various hardware vendors (Intel DCPMM is one such technology currently available). These new memory types require different libraries for allocation and management (such as PMDK and memkind). The high capacities available make it possible to provision large caches (up to several TBs in size), beyond what is achievable with DRAM. The new allocator provided in this PR uses the memkind library to allocate memory on different media. **Performance** We tested the new allocator using db_bench. - For each test, we vary the size of the block cache (relative to the size of the uncompressed data in the database). - The database is filled sequentially. Throughput is then measured with a readrandom benchmark. - We use a uniform distribution as a worst-case scenario. The plot shows throughput (ops/s) relative to a configuration with no block cache and default allocator. For all tests, p99 latency is below 500 us. ![image](https://user-images.githubusercontent.com/26400080/71108594-42479100-2178-11ea-8231-8a775bbc92db.png) **Changes** - Add MemkindKmemAllocator - Add --use_cache_memkind_kmem_allocator db_bench option (to create an LRU block cache with the new allocator) - Add detection of memkind library with KMEM DAX support - Add test for MemkindKmemAllocator **Minimum Requirements** - kernel 5.3.12 - ndctl v67 - https://github.com/pmem/ndctl - memkind v1.10.0 - https://github.com/memkind/memkind **Memory Configuration** The allocator uses the MEMKIND_DAX_KMEM memory kind. Follow the instructions on[ memkind’s GitHub page](https://github.com/memkind/memkind) to set up NVDIMM memory accordingly. Note on memory allocation with NVDIMM memory exposed as system memory. - The MemkindKmemAllocator will only allocate from NVDIMM memory (using memkind_malloc with MEMKIND_DAX_KMEM kind). - The default allocator is not restricted to RAM by default. Based on NUMA node latency, the kernel should allocate from local RAM preferentially, but it’s a kernel decision. numactl --preferred/--membind can be used to allocate preferentially/exclusively from the local RAM node. **Usage** When creating an LRU cache, pass a MemkindKmemAllocator object as argument. For example (replace capacity with the desired value in bytes): ``` #include "rocksdb/cache.h" #include "memory/memkind_kmem_allocator.h" NewLRUCache( capacity /*size_t*/, 6 /*cache_numshardbits*/, false /*strict_capacity_limit*/, false /*cache_high_pri_pool_ratio*/, std::make_shared<MemkindKmemAllocator>()); ``` Refer to [RocksDB’s block cache documentation](https://github.com/facebook/rocksdb/wiki/Block-Cache) to assign the LRU cache as block cache for a database. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6214 Reviewed By: cheng-chang Differential Revision: D19292435 fbshipit-source-id: 7202f47b769e7722b539c86c2ffd669f64d7b4e1
2020-04-10 05:45:17 +02:00
if ! test $ROCKSDB_DISABLE_MEMKIND; then
# Test whether memkind library is installed
$CXX $CFLAGS $COMMON_FLAGS -lmemkind -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <memkind.h>
int main() {
memkind_malloc(MEMKIND_DAX_KMEM, 1024);
return 0;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DMEMKIND"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lmemkind"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lmemkind"
fi
fi
if ! test $ROCKSDB_DISABLE_PTHREAD_MUTEX_ADAPTIVE_NP; then
# Test whether PTHREAD_MUTEX_ADAPTIVE_NP mutex type is available
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <pthread.h>
int main() {
int x = PTHREAD_MUTEX_ADAPTIVE_NP;
(void)x;
return 0;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_PTHREAD_ADAPTIVE_MUTEX"
fi
fi
if ! test $ROCKSDB_DISABLE_BACKTRACE; then
# Test whether backtrace is available
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <execinfo.h>
int main() {
void* frames[1];
backtrace_symbols(frames, backtrace(frames, 1));
return 0;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
else
# Test whether execinfo library is installed
$CXX $CFLAGS -lexecinfo -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <execinfo.h>
int main() {
void* frames[1];
backtrace_symbols(frames, backtrace(frames, 1));
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_BACKTRACE"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lexecinfo"
JAVA_LDFLAGS="$JAVA_LDFLAGS -lexecinfo"
fi
fi
fi
if ! test $ROCKSDB_DISABLE_PG; then
# Test if -pg is supported
$CXX $CFLAGS -pg -x c++ - -o /dev/null 2>/dev/null <<EOF
int main() {
return 0;
}
EOF
if [ "$?" = 0 ]; then
PROFILING_FLAGS=-pg
fi
fi
if ! test $ROCKSDB_DISABLE_SYNC_FILE_RANGE; then
# Test whether sync_file_range is supported for compatibility with an old glibc
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <fcntl.h>
int main() {
int fd = open("/dev/null", 0);
sync_file_range(fd, 0, 1024, SYNC_FILE_RANGE_WRITE);
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_RANGESYNC_PRESENT"
fi
fi
if ! test $ROCKSDB_DISABLE_SCHED_GETCPU; then
# Test whether sched_getcpu is supported
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <sched.h>
int main() {
int cpuid = sched_getcpu();
(void)cpuid;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SCHED_GETCPU_PRESENT"
fi
fi
if ! test $ROCKSDB_DISABLE_AUXV_GETAUXVAL; then
# Test whether getauxval is supported
$CXX $CFLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <sys/auxv.h>
int main() {
uint64_t auxv = getauxval(AT_HWCAP);
(void)auxv;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_AUXV_GETAUXVAL_PRESENT"
fi
fi
if ! test $ROCKSDB_DISABLE_ALIGNED_NEW; then
# Test whether c++17 aligned-new is supported
$CXX $PLATFORM_CXXFLAGS -faligned-new -x c++ - -o /dev/null 2>/dev/null <<EOF
struct alignas(1024) t {int a;};
int main() {}
EOF
if [ "$?" = 0 ]; then
PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS -faligned-new -DHAVE_ALIGNED_NEW"
fi
fi
fi
# TODO(tec): Fix -Wshorten-64-to-32 errors on FreeBSD and enable the warning.
# -Wshorten-64-to-32 breaks compilation on FreeBSD i386
if ! [ "$TARGET_OS" = FreeBSD -a "$TARGET_ARCHITECTURE" = i386 ]; then
# Test whether -Wshorten-64-to-32 is available
$CXX $CFLAGS -x c++ - -o /dev/null -Wshorten-64-to-32 2>/dev/null <<EOF
int main() {}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -Wshorten-64-to-32"
fi
fi
# shall we use HDFS?
if test "$USE_HDFS"; then
if test -z "$JAVA_HOME"; then
echo "JAVA_HOME has to be set for HDFS usage." >&2
exit 1
fi
HDFS_CCFLAGS="$HDFS_CCFLAGS -I$JAVA_HOME/include -I$JAVA_HOME/include/linux -DUSE_HDFS -I$HADOOP_HOME/include"
HDFS_LDFLAGS="$HDFS_LDFLAGS -lhdfs -L$JAVA_HOME/jre/lib/amd64 -L$HADOOP_HOME/lib/native"
HDFS_LDFLAGS="$HDFS_LDFLAGS -L$JAVA_HOME/jre/lib/amd64/server -L$GLIBC_RUNTIME_PATH/lib"
HDFS_LDFLAGS="$HDFS_LDFLAGS -ldl -lverify -ljava -ljvm"
COMMON_FLAGS="$COMMON_FLAGS $HDFS_CCFLAGS"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS $HDFS_LDFLAGS"
JAVA_LDFLAGS="$JAVA_LDFLAGS $HDFS_LDFLAGS"
fi
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
if test "0$PORTABLE" -eq 0; then
if test -n "`echo $TARGET_ARCHITECTURE | grep ^ppc64`"; then
# Tune for this POWER processor, treating '+' models as base models
POWER=`LD_SHOW_AUXV=1 /bin/true | grep AT_PLATFORM | grep -E -o power[0-9]+`
COMMON_FLAGS="$COMMON_FLAGS -mcpu=$POWER -mtune=$POWER "
elif test -n "`echo $TARGET_ARCHITECTURE | grep ^s390x`"; then
COMMON_FLAGS="$COMMON_FLAGS -march=z10 "
elif test -n "`echo $TARGET_ARCHITECTURE | grep -e^arm -e^aarch64`"; then
# TODO: Handle this with approprite options.
COMMON_FLAGS="$COMMON_FLAGS"
elif test -n "`echo $TARGET_ARCHITECTURE | grep ^aarch64`"; then
COMMON_FLAGS="$COMMON_FLAGS"
elif [ "$TARGET_OS" == "IOS" ]; then
COMMON_FLAGS="$COMMON_FLAGS"
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
elif [ "$TARGET_OS" == "AIX" ] || [ "$TARGET_OS" == "SunOS" ]; then
# TODO: Not sure why we don't use -march=native on these OSes
if test "$USE_SSE"; then
TRY_SSE_ETC="1"
fi
else
COMMON_FLAGS="$COMMON_FLAGS -march=native "
fi
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
else
# PORTABLE=1
if test "$USE_SSE"; then
TRY_SSE_ETC="1"
fi
if [[ "${PLATFORM}" == "OS_MACOSX" ]]; then
# For portability compile for macOS 10.12 (2016) or newer
COMMON_FLAGS="$COMMON_FLAGS -mmacosx-version-min=10.12"
PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -mmacosx-version-min=10.12"
PLATFORM_SHARED_LDFLAGS="$PLATFORM_SHARED_LDFLAGS -mmacosx-version-min=10.12"
fi
fi
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
if test "$TRY_SSE_ETC"; then
# The USE_SSE flag now means "attempt to compile with widely-available
# Intel architecture extensions utilized by specific optimizations in the
# source code." It's a qualifier on PORTABLE=1 that means "mostly portable."
# It doesn't even really check that your current CPU is compatible.
#
# SSE4.2 available since nehalem, ca. 2008-2010
TRY_SSE42="-msse4.2"
# PCLMUL available since westmere, ca. 2010-2011
TRY_PCLMUL="-mpclmul"
New Bloom filter implementation for full and partitioned filters (#6007) Summary: Adds an improved, replacement Bloom filter implementation (FastLocalBloom) for full and partitioned filters in the block-based table. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single filter. Speed The improved speed, at least on recent x86_64, comes from * Using fastrange instead of modulo (%) * Using our new hash function (XXH3 preview, added in a previous commit), which is much faster for large keys and only *slightly* slower on keys around 12 bytes if hashing the same size many thousands of times in a row. * Optimizing the Bloom filter queries with AVX2 SIMD operations. (Added AVX2 to the USE_SSE=1 build.) Careful design was required to support (a) SIMD-optimized queries, (b) compatible non-SIMD code that's simple and efficient, (c) flexible choice of number of probes, and (d) essentially maximized accuracy for a cache-local Bloom filter. Probes are made eight at a time, so any number of probes up to 8 is the same speed, then up to 16, etc. * Prefetching cache lines when building the filter. Although this optimization could be applied to the old structure as well, it seems to balance out the small added cost of accumulating 64 bit hashes for adding to the filter rather than 32 bit hashes. Here's nominal speed data from filter_bench (200MB in filters, about 10k keys each, 10 bits filter data / key, 6 probes, avg key size 24 bytes, includes hashing time) on Skylake DE (relatively low clock speed): $ ./filter_bench -quick -impl=2 -net_includes_hashing # New Bloom filter Build avg ns/key: 47.7135 Mixed inside/outside queries... Single filter net ns/op: 26.2825 Random filter net ns/op: 150.459 Average FP rate %: 0.954651 $ ./filter_bench -quick -impl=0 -net_includes_hashing # Old Bloom filter Build avg ns/key: 47.2245 Mixed inside/outside queries... Single filter net ns/op: 63.2978 Random filter net ns/op: 188.038 Average FP rate %: 1.13823 Similar build time but dramatically faster query times on hot data (63 ns to 26 ns), and somewhat faster on stale data (188 ns to 150 ns). Performance differences on batched and skewed query loads are between these extremes as expected. The only other interesting thing about speed is "inside" (query key was added to filter) vs. "outside" (query key was not added to filter) query times. The non-SIMD implementations are substantially slower when most queries are "outside" vs. "inside". This goes against what one might expect or would have observed years ago, as "outside" queries only need about two probes on average, due to short-circuiting, while "inside" always have num_probes (say 6). The problem is probably the nastily unpredictable branch. The SIMD implementation has few branches (very predictable) and has pretty consistent running time regardless of query outcome. Accuracy The generally improved accuracy (re: Issue https://github.com/facebook/rocksdb/issues/5857) comes from a better design for probing indices within a cache line (re: Issue https://github.com/facebook/rocksdb/issues/4120) and improved accuracy for millions of keys in a single filter from using a 64-bit hash function (XXH3p). Design details in code comments. Accuracy data (generalizes, except old impl gets worse with millions of keys): Memory bits per key: FP rate percent old impl -> FP rate percent new impl 6: 5.70953 -> 5.69888 8: 2.45766 -> 2.29709 10: 1.13977 -> 0.959254 12: 0.662498 -> 0.411593 16: 0.353023 -> 0.0873754 24: 0.261552 -> 0.0060971 50: 0.225453 -> ~0.00003 (less than 1 in a million queries are FP) Fixes https://github.com/facebook/rocksdb/issues/5857 Fixes https://github.com/facebook/rocksdb/issues/4120 Unlike the old implementation, this implementation has a fixed cache line size (64 bytes). At 10 bits per key, the accuracy of this new implementation is very close to the old implementation with 128-byte cache line size. If there's sufficient demand, this implementation could be generalized. Compatibility Although old releases would see the new structure as corrupt filter data and read the table as if there's no filter, we've decided only to enable the new Bloom filter with new format_version=5. This provides a smooth path for automatic adoption over time, with an option for early opt-in. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6007 Test Plan: filter_bench has been used thoroughly to validate speed, accuracy, and correctness. Unit tests have been carefully updated to exercise new and old implementations, as well as the logic to select an implementation based on context (format_version). Differential Revision: D18294749 Pulled By: pdillinger fbshipit-source-id: d44c9db3696e4d0a17caaec47075b7755c262c5f
2019-11-14 01:31:26 +01:00
# AVX2 available since haswell, ca. 2013-2015
TRY_AVX2="-mavx2"
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
fi
$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS $TRY_SSE42 -x c++ - -o /dev/null 2>/dev/null <<EOF
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
#include <cstdint>
#include <nmmintrin.h>
int main() {
volatile uint32_t x = _mm_crc32_u32(0, 0);
(void)x;
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
}
EOF
if [ "$?" = 0 ]; then
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
COMMON_FLAGS="$COMMON_FLAGS $TRY_SSE42 -DHAVE_SSE42"
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
elif test "$USE_SSE"; then
echo "warning: USE_SSE specified but compiler could not use SSE intrinsics, disabling" >&2
Port 3 way SSE4.2 crc32c implementation from Folly Summary: **# Summary** RocksDB uses SSE crc32 intrinsics to calculate the crc32 values but it does it in single way fashion (not pipelined on single CPU core). Intel's whitepaper () published an algorithm that uses 3-way pipelining for the crc32 intrinsics, then use pclmulqdq intrinsic to combine the values. Because pclmulqdq has overhead on its own, this algorithm will show perf gains on buffers larger than 216 bytes, which makes RocksDB a perfect user, since most of the buffers RocksDB call crc32c on is over 4KB. Initial db_bench show tremendous CPU gain. This change uses the 3-way SSE algorithm by default. The old SSE algorithm is now behind a compiler tag NO_THREEWAY_CRC32C. If user compiles the code with NO_THREEWAY_CRC32C=1 then the old SSE Crc32c algorithm would be used. If the server does not have SSE4.2 at the run time the slow way (Non SSE) will be used. **# Performance Test Results** We ran the FillRandom and ReadRandom benchmarks in db_bench. ReadRandom is the point of interest here since it calculates the CRC32 for the in-mem buffers. We did 3 runs for each algorithm. Before this change the CRC32 value computation takes about 11.5% of total CPU cost, and with the new 3-way algorithm it reduced to around 4.5%. The overall throughput also improved from 25.53MB/s to 27.63MB/s. 1) ReadRandom in db_bench overall metrics PER RUN Algorithm | run | micros/op | ops/sec |Throughput (MB/s) 3-way | 1 | 4.143 | 241387 | 26.7 3-way | 2 | 3.775 | 264872 | 29.3 3-way | 3 | 4.116 | 242929 | 26.9 FastCrc32c|1 | 4.037 | 247727 | 27.4 FastCrc32c|2 | 4.648 | 215166 | 23.8 FastCrc32c|3 | 4.352 | 229799 | 25.4 AVG Algorithm | Average of micros/op | Average of ops/sec | Average of Throughput (MB/s) 3-way | 4.01 | 249,729 | 27.63 FastCrc32c | 4.35 | 230,897 | 25.53 2) Crc32c computation CPU cost (inclusive samples percentage) PER RUN Implementation | run |  TotalSamples | Crc32c percentage 3-way   | 1    |  4,572,250,000 | 4.37% 3-way   | 2    |  3,779,250,000 | 4.62% 3-way   | 3    |  4,129,500,000 | 4.48% FastCrc32c     | 1    |  4,663,500,000 | 11.24% FastCrc32c     | 2    |  4,047,500,000 | 12.34% FastCrc32c     | 3    |  4,366,750,000 | 11.68% **# Test Plan** make -j64 corruption_test && ./corruption_test By default it uses 3-way SSE algorithm NO_THREEWAY_CRC32C=1 make -j64 corruption_test && ./corruption_test make clean && DEBUG_LEVEL=0 make -j64 db_bench make clean && DEBUG_LEVEL=0 NO_THREEWAY_CRC32C=1 make -j64 db_bench Closes https://github.com/facebook/rocksdb/pull/3173 Differential Revision: D6330882 Pulled By: yingsu00 fbshipit-source-id: 8ec3d89719533b63b536a736663ca6f0dd4482e9
2017-12-20 03:20:50 +01:00
fi
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS $TRY_PCLMUL -x c++ - -o /dev/null 2>/dev/null <<EOF
Port 3 way SSE4.2 crc32c implementation from Folly Summary: **# Summary** RocksDB uses SSE crc32 intrinsics to calculate the crc32 values but it does it in single way fashion (not pipelined on single CPU core). Intel's whitepaper () published an algorithm that uses 3-way pipelining for the crc32 intrinsics, then use pclmulqdq intrinsic to combine the values. Because pclmulqdq has overhead on its own, this algorithm will show perf gains on buffers larger than 216 bytes, which makes RocksDB a perfect user, since most of the buffers RocksDB call crc32c on is over 4KB. Initial db_bench show tremendous CPU gain. This change uses the 3-way SSE algorithm by default. The old SSE algorithm is now behind a compiler tag NO_THREEWAY_CRC32C. If user compiles the code with NO_THREEWAY_CRC32C=1 then the old SSE Crc32c algorithm would be used. If the server does not have SSE4.2 at the run time the slow way (Non SSE) will be used. **# Performance Test Results** We ran the FillRandom and ReadRandom benchmarks in db_bench. ReadRandom is the point of interest here since it calculates the CRC32 for the in-mem buffers. We did 3 runs for each algorithm. Before this change the CRC32 value computation takes about 11.5% of total CPU cost, and with the new 3-way algorithm it reduced to around 4.5%. The overall throughput also improved from 25.53MB/s to 27.63MB/s. 1) ReadRandom in db_bench overall metrics PER RUN Algorithm | run | micros/op | ops/sec |Throughput (MB/s) 3-way | 1 | 4.143 | 241387 | 26.7 3-way | 2 | 3.775 | 264872 | 29.3 3-way | 3 | 4.116 | 242929 | 26.9 FastCrc32c|1 | 4.037 | 247727 | 27.4 FastCrc32c|2 | 4.648 | 215166 | 23.8 FastCrc32c|3 | 4.352 | 229799 | 25.4 AVG Algorithm | Average of micros/op | Average of ops/sec | Average of Throughput (MB/s) 3-way | 4.01 | 249,729 | 27.63 FastCrc32c | 4.35 | 230,897 | 25.53 2) Crc32c computation CPU cost (inclusive samples percentage) PER RUN Implementation | run |  TotalSamples | Crc32c percentage 3-way   | 1    |  4,572,250,000 | 4.37% 3-way   | 2    |  3,779,250,000 | 4.62% 3-way   | 3    |  4,129,500,000 | 4.48% FastCrc32c     | 1    |  4,663,500,000 | 11.24% FastCrc32c     | 2    |  4,047,500,000 | 12.34% FastCrc32c     | 3    |  4,366,750,000 | 11.68% **# Test Plan** make -j64 corruption_test && ./corruption_test By default it uses 3-way SSE algorithm NO_THREEWAY_CRC32C=1 make -j64 corruption_test && ./corruption_test make clean && DEBUG_LEVEL=0 make -j64 db_bench make clean && DEBUG_LEVEL=0 NO_THREEWAY_CRC32C=1 make -j64 db_bench Closes https://github.com/facebook/rocksdb/pull/3173 Differential Revision: D6330882 Pulled By: yingsu00 fbshipit-source-id: 8ec3d89719533b63b536a736663ca6f0dd4482e9
2017-12-20 03:20:50 +01:00
#include <cstdint>
#include <wmmintrin.h>
int main() {
const auto a = _mm_set_epi64x(0, 0);
const auto b = _mm_set_epi64x(0, 0);
const auto c = _mm_clmulepi64_si128(a, b, 0x00);
auto d = _mm_cvtsi128_si64(c);
(void)d;
Port 3 way SSE4.2 crc32c implementation from Folly Summary: **# Summary** RocksDB uses SSE crc32 intrinsics to calculate the crc32 values but it does it in single way fashion (not pipelined on single CPU core). Intel's whitepaper () published an algorithm that uses 3-way pipelining for the crc32 intrinsics, then use pclmulqdq intrinsic to combine the values. Because pclmulqdq has overhead on its own, this algorithm will show perf gains on buffers larger than 216 bytes, which makes RocksDB a perfect user, since most of the buffers RocksDB call crc32c on is over 4KB. Initial db_bench show tremendous CPU gain. This change uses the 3-way SSE algorithm by default. The old SSE algorithm is now behind a compiler tag NO_THREEWAY_CRC32C. If user compiles the code with NO_THREEWAY_CRC32C=1 then the old SSE Crc32c algorithm would be used. If the server does not have SSE4.2 at the run time the slow way (Non SSE) will be used. **# Performance Test Results** We ran the FillRandom and ReadRandom benchmarks in db_bench. ReadRandom is the point of interest here since it calculates the CRC32 for the in-mem buffers. We did 3 runs for each algorithm. Before this change the CRC32 value computation takes about 11.5% of total CPU cost, and with the new 3-way algorithm it reduced to around 4.5%. The overall throughput also improved from 25.53MB/s to 27.63MB/s. 1) ReadRandom in db_bench overall metrics PER RUN Algorithm | run | micros/op | ops/sec |Throughput (MB/s) 3-way | 1 | 4.143 | 241387 | 26.7 3-way | 2 | 3.775 | 264872 | 29.3 3-way | 3 | 4.116 | 242929 | 26.9 FastCrc32c|1 | 4.037 | 247727 | 27.4 FastCrc32c|2 | 4.648 | 215166 | 23.8 FastCrc32c|3 | 4.352 | 229799 | 25.4 AVG Algorithm | Average of micros/op | Average of ops/sec | Average of Throughput (MB/s) 3-way | 4.01 | 249,729 | 27.63 FastCrc32c | 4.35 | 230,897 | 25.53 2) Crc32c computation CPU cost (inclusive samples percentage) PER RUN Implementation | run |  TotalSamples | Crc32c percentage 3-way   | 1    |  4,572,250,000 | 4.37% 3-way   | 2    |  3,779,250,000 | 4.62% 3-way   | 3    |  4,129,500,000 | 4.48% FastCrc32c     | 1    |  4,663,500,000 | 11.24% FastCrc32c     | 2    |  4,047,500,000 | 12.34% FastCrc32c     | 3    |  4,366,750,000 | 11.68% **# Test Plan** make -j64 corruption_test && ./corruption_test By default it uses 3-way SSE algorithm NO_THREEWAY_CRC32C=1 make -j64 corruption_test && ./corruption_test make clean && DEBUG_LEVEL=0 make -j64 db_bench make clean && DEBUG_LEVEL=0 NO_THREEWAY_CRC32C=1 make -j64 db_bench Closes https://github.com/facebook/rocksdb/pull/3173 Differential Revision: D6330882 Pulled By: yingsu00 fbshipit-source-id: 8ec3d89719533b63b536a736663ca6f0dd4482e9
2017-12-20 03:20:50 +01:00
}
EOF
if [ "$?" = 0 ]; then
Clean up + fix build scripts re: USE_SSE= and PORTABLE= (#5800) Summary: In preparing to utilize a new Intel instruction extension, I noticed problems with the existing build script in regard to the existing utilized extensions, either with USE_SSE or PORTABLE flags. * PORTABLE=0 was interpreted the same as PORTABLE=1. Now empty and 0 mean the same. (I guess you were not supposed to set PORTABLE= if you wanted non-portable--except that...) * The Facebook build script extensions would set PORTABLE=1 even if it's already set in a make var or environment. Now it does not override a non-empty setting, so use PORTABLE=0 for fully optimized build, overriding Facebook environment default. * Put in an explanation of the USE_SSE flag where it's used by build_detect_platform, and cleaned up some confusing/redundant associated logic. * If USE_SSE was set and expected intrinsics were not available, build_detect_platform would exit early but build would proceed with broken, incomplete configuration. Now warning is gracefully recovered. * If USE_SSE was set and expected intrinsics were not available, build would still try to use flags like -msse4.2 etc. which could lead to unexpected compilation failure or binary incompatibility. Now those flags are not used if the warning is issued. This should not break or change existing, valid build scripts. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5800 Test Plan: manual case testing Differential Revision: D17369543 Pulled By: pdillinger fbshipit-source-id: 4ee244911680ae71144d272c40aceea548e3ce88
2019-09-13 20:04:52 +02:00
COMMON_FLAGS="$COMMON_FLAGS $TRY_PCLMUL -DHAVE_PCLMUL"
Port 3 way SSE4.2 crc32c implementation from Folly Summary: **# Summary** RocksDB uses SSE crc32 intrinsics to calculate the crc32 values but it does it in single way fashion (not pipelined on single CPU core). Intel's whitepaper () published an algorithm that uses 3-way pipelining for the crc32 intrinsics, then use pclmulqdq intrinsic to combine the values. Because pclmulqdq has overhead on its own, this algorithm will show perf gains on buffers larger than 216 bytes, which makes RocksDB a perfect user, since most of the buffers RocksDB call crc32c on is over 4KB. Initial db_bench show tremendous CPU gain. This change uses the 3-way SSE algorithm by default. The old SSE algorithm is now behind a compiler tag NO_THREEWAY_CRC32C. If user compiles the code with NO_THREEWAY_CRC32C=1 then the old SSE Crc32c algorithm would be used. If the server does not have SSE4.2 at the run time the slow way (Non SSE) will be used. **# Performance Test Results** We ran the FillRandom and ReadRandom benchmarks in db_bench. ReadRandom is the point of interest here since it calculates the CRC32 for the in-mem buffers. We did 3 runs for each algorithm. Before this change the CRC32 value computation takes about 11.5% of total CPU cost, and with the new 3-way algorithm it reduced to around 4.5%. The overall throughput also improved from 25.53MB/s to 27.63MB/s. 1) ReadRandom in db_bench overall metrics PER RUN Algorithm | run | micros/op | ops/sec |Throughput (MB/s) 3-way | 1 | 4.143 | 241387 | 26.7 3-way | 2 | 3.775 | 264872 | 29.3 3-way | 3 | 4.116 | 242929 | 26.9 FastCrc32c|1 | 4.037 | 247727 | 27.4 FastCrc32c|2 | 4.648 | 215166 | 23.8 FastCrc32c|3 | 4.352 | 229799 | 25.4 AVG Algorithm | Average of micros/op | Average of ops/sec | Average of Throughput (MB/s) 3-way | 4.01 | 249,729 | 27.63 FastCrc32c | 4.35 | 230,897 | 25.53 2) Crc32c computation CPU cost (inclusive samples percentage) PER RUN Implementation | run |  TotalSamples | Crc32c percentage 3-way   | 1    |  4,572,250,000 | 4.37% 3-way   | 2    |  3,779,250,000 | 4.62% 3-way   | 3    |  4,129,500,000 | 4.48% FastCrc32c     | 1    |  4,663,500,000 | 11.24% FastCrc32c     | 2    |  4,047,500,000 | 12.34% FastCrc32c     | 3    |  4,366,750,000 | 11.68% **# Test Plan** make -j64 corruption_test && ./corruption_test By default it uses 3-way SSE algorithm NO_THREEWAY_CRC32C=1 make -j64 corruption_test && ./corruption_test make clean && DEBUG_LEVEL=0 make -j64 db_bench make clean && DEBUG_LEVEL=0 NO_THREEWAY_CRC32C=1 make -j64 db_bench Closes https://github.com/facebook/rocksdb/pull/3173 Differential Revision: D6330882 Pulled By: yingsu00 fbshipit-source-id: 8ec3d89719533b63b536a736663ca6f0dd4482e9
2017-12-20 03:20:50 +01:00
elif test "$USE_SSE"; then
echo "warning: USE_SSE specified but compiler could not use PCLMUL intrinsics, disabling" >&2
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
fi
New Bloom filter implementation for full and partitioned filters (#6007) Summary: Adds an improved, replacement Bloom filter implementation (FastLocalBloom) for full and partitioned filters in the block-based table. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single filter. Speed The improved speed, at least on recent x86_64, comes from * Using fastrange instead of modulo (%) * Using our new hash function (XXH3 preview, added in a previous commit), which is much faster for large keys and only *slightly* slower on keys around 12 bytes if hashing the same size many thousands of times in a row. * Optimizing the Bloom filter queries with AVX2 SIMD operations. (Added AVX2 to the USE_SSE=1 build.) Careful design was required to support (a) SIMD-optimized queries, (b) compatible non-SIMD code that's simple and efficient, (c) flexible choice of number of probes, and (d) essentially maximized accuracy for a cache-local Bloom filter. Probes are made eight at a time, so any number of probes up to 8 is the same speed, then up to 16, etc. * Prefetching cache lines when building the filter. Although this optimization could be applied to the old structure as well, it seems to balance out the small added cost of accumulating 64 bit hashes for adding to the filter rather than 32 bit hashes. Here's nominal speed data from filter_bench (200MB in filters, about 10k keys each, 10 bits filter data / key, 6 probes, avg key size 24 bytes, includes hashing time) on Skylake DE (relatively low clock speed): $ ./filter_bench -quick -impl=2 -net_includes_hashing # New Bloom filter Build avg ns/key: 47.7135 Mixed inside/outside queries... Single filter net ns/op: 26.2825 Random filter net ns/op: 150.459 Average FP rate %: 0.954651 $ ./filter_bench -quick -impl=0 -net_includes_hashing # Old Bloom filter Build avg ns/key: 47.2245 Mixed inside/outside queries... Single filter net ns/op: 63.2978 Random filter net ns/op: 188.038 Average FP rate %: 1.13823 Similar build time but dramatically faster query times on hot data (63 ns to 26 ns), and somewhat faster on stale data (188 ns to 150 ns). Performance differences on batched and skewed query loads are between these extremes as expected. The only other interesting thing about speed is "inside" (query key was added to filter) vs. "outside" (query key was not added to filter) query times. The non-SIMD implementations are substantially slower when most queries are "outside" vs. "inside". This goes against what one might expect or would have observed years ago, as "outside" queries only need about two probes on average, due to short-circuiting, while "inside" always have num_probes (say 6). The problem is probably the nastily unpredictable branch. The SIMD implementation has few branches (very predictable) and has pretty consistent running time regardless of query outcome. Accuracy The generally improved accuracy (re: Issue https://github.com/facebook/rocksdb/issues/5857) comes from a better design for probing indices within a cache line (re: Issue https://github.com/facebook/rocksdb/issues/4120) and improved accuracy for millions of keys in a single filter from using a 64-bit hash function (XXH3p). Design details in code comments. Accuracy data (generalizes, except old impl gets worse with millions of keys): Memory bits per key: FP rate percent old impl -> FP rate percent new impl 6: 5.70953 -> 5.69888 8: 2.45766 -> 2.29709 10: 1.13977 -> 0.959254 12: 0.662498 -> 0.411593 16: 0.353023 -> 0.0873754 24: 0.261552 -> 0.0060971 50: 0.225453 -> ~0.00003 (less than 1 in a million queries are FP) Fixes https://github.com/facebook/rocksdb/issues/5857 Fixes https://github.com/facebook/rocksdb/issues/4120 Unlike the old implementation, this implementation has a fixed cache line size (64 bytes). At 10 bits per key, the accuracy of this new implementation is very close to the old implementation with 128-byte cache line size. If there's sufficient demand, this implementation could be generalized. Compatibility Although old releases would see the new structure as corrupt filter data and read the table as if there's no filter, we've decided only to enable the new Bloom filter with new format_version=5. This provides a smooth path for automatic adoption over time, with an option for early opt-in. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6007 Test Plan: filter_bench has been used thoroughly to validate speed, accuracy, and correctness. Unit tests have been carefully updated to exercise new and old implementations, as well as the logic to select an implementation based on context (format_version). Differential Revision: D18294749 Pulled By: pdillinger fbshipit-source-id: d44c9db3696e4d0a17caaec47075b7755c262c5f
2019-11-14 01:31:26 +01:00
$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS $TRY_AVX2 -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <cstdint>
#include <immintrin.h>
int main() {
const auto a = _mm256_setr_epi32(0, 1, 2, 3, 4, 7, 6, 5);
const auto b = _mm256_permutevar8x32_epi32(a, a);
(void)b;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS $TRY_AVX2 -DHAVE_AVX2"
elif test "$USE_SSE"; then
echo "warning: USE_SSE specified but compiler could not use AVX2 intrinsics, disabling" >&2
fi
$CXX $PLATFORM_CXXFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#include <cstdint>
int main() {
uint64_t a = 0xffffFFFFffffFFFF;
__uint128_t b = __uint128_t(a) * a;
a = static_cast<uint64_t>(b >> 64);
(void)a;
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DHAVE_UINT128_EXTENSION"
fi
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
# iOS doesn't support thread-local storage, but this check would erroneously
# succeed because the cross-compiler flags are added by the Makefile, not this
# script.
if [ "$PLATFORM" != IOS ]; then
$CXX $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null <<EOF
#if defined(_MSC_VER) && !defined(__thread)
#define __thread __declspec(thread)
#endif
int main() {
static __thread int tls;
(void)tls;
cross-platform compatibility improvements Summary: We've had a couple CockroachDB users fail to build RocksDB on exotic platforms, so I figured I'd try my hand at solving these issues upstream. The problems stem from a) `USE_SSE=1` being too aggressive about turning on SSE4.2, even on toolchains that don't support SSE4.2 and b) RocksDB attempting to detect support for thread-local storage based on OS, even though it can vary by compiler on the same OS. See the individual commit messages for details. Regarding SSE support, this PR should change virtually nothing for non-CMake based builds. `make`, `PORTABLE=1 make`, `USE_SSE=1 make`, and `PORTABLE=1 USE_SSE=1 make` function exactly as before, except that SSE support will be automatically disabled when a simple SSE4.2-using test program fails to compile, as it does on OpenBSD. (OpenBSD's ports GCC supports SSE4.2, but its binutils do not, so `__SSE_4_2__` is defined but an SSE4.2-using program will fail to assemble.) A warning is emitted in this case. The CMake build is modified to support the same set of options, except that `USE_SSE` is spelled `FORCE_SSE42` because `USE_SSE` is rather useless now that we can automatically detect SSE support, and I figure changing options in the CMake build is less disruptive than changing the non-CMake build. I've tested these changes on all the platforms I can get my hands on (macOS, Windows MSVC, Windows MinGW, and OpenBSD) and it all works splendidly. Let me know if there's anything you object to—I obviously don't mean to break any of your build pipelines in the process of fixing ours downstream. Closes https://github.com/facebook/rocksdb/pull/2199 Differential Revision: D5054042 Pulled By: yiwu-arbug fbshipit-source-id: 938e1fc665c049c02ae15698e1409155b8e72171
2017-05-15 23:42:32 +02:00
}
EOF
if [ "$?" = 0 ]; then
COMMON_FLAGS="$COMMON_FLAGS -DROCKSDB_SUPPORT_THREAD_LOCAL"
fi
fi
if [ "$FBCODE_BUILD" != "true" -a "$PLATFORM" = OS_LINUX ]; then
$CXX $COMMON_FLAGS $PLATFORM_SHARED_CFLAGS -x c++ -c - -o test_dl.o 2>/dev/null <<EOF
void dummy_func() {}
EOF
if [ "$?" = 0 ]; then
$CXX $COMMON_FLAGS $PLATFORM_SHARED_LDFLAGS test_dl.o -o /dev/null 2>/dev/null
if [ "$?" = 0 ]; then
EXEC_LDFLAGS+="-ldl"
rm -f test_dl.o
fi
fi
fi
PLATFORM_CCFLAGS="$PLATFORM_CCFLAGS $COMMON_FLAGS"
PLATFORM_CXXFLAGS="$PLATFORM_CXXFLAGS $COMMON_FLAGS"
VALGRIND_VER="$VALGRIND_VER"
ROCKSDB_MAJOR=`build_tools/version.sh major`
ROCKSDB_MINOR=`build_tools/version.sh minor`
ROCKSDB_PATCH=`build_tools/version.sh patch`
echo "CC=$CC" >> "$OUTPUT"
echo "CXX=$CXX" >> "$OUTPUT"
echo "PLATFORM=$PLATFORM" >> "$OUTPUT"
echo "PLATFORM_LDFLAGS=$PLATFORM_LDFLAGS" >> "$OUTPUT"
echo "JAVA_LDFLAGS=$JAVA_LDFLAGS" >> "$OUTPUT"
echo "JAVA_STATIC_LDFLAGS=$JAVA_STATIC_LDFLAGS" >> "$OUTPUT"
echo "JAVAC_ARGS=$JAVAC_ARGS" >> "$OUTPUT"
echo "VALGRIND_VER=$VALGRIND_VER" >> "$OUTPUT"
echo "PLATFORM_CCFLAGS=$PLATFORM_CCFLAGS" >> "$OUTPUT"
echo "PLATFORM_CXXFLAGS=$PLATFORM_CXXFLAGS" >> "$OUTPUT"
echo "PLATFORM_SHARED_CFLAGS=$PLATFORM_SHARED_CFLAGS" >> "$OUTPUT"
echo "PLATFORM_SHARED_EXT=$PLATFORM_SHARED_EXT" >> "$OUTPUT"
echo "PLATFORM_SHARED_LDFLAGS=$PLATFORM_SHARED_LDFLAGS" >> "$OUTPUT"
echo "PLATFORM_SHARED_VERSIONED=$PLATFORM_SHARED_VERSIONED" >> "$OUTPUT"
echo "EXEC_LDFLAGS=$EXEC_LDFLAGS" >> "$OUTPUT"
echo "JEMALLOC_INCLUDE=$JEMALLOC_INCLUDE" >> "$OUTPUT"
echo "JEMALLOC_LIB=$JEMALLOC_LIB" >> "$OUTPUT"
echo "ROCKSDB_MAJOR=$ROCKSDB_MAJOR" >> "$OUTPUT"
echo "ROCKSDB_MINOR=$ROCKSDB_MINOR" >> "$OUTPUT"
echo "ROCKSDB_PATCH=$ROCKSDB_PATCH" >> "$OUTPUT"
echo "CLANG_SCAN_BUILD=$CLANG_SCAN_BUILD" >> "$OUTPUT"
echo "CLANG_ANALYZER=$CLANG_ANALYZER" >> "$OUTPUT"
echo "PROFILING_FLAGS=$PROFILING_FLAGS" >> "$OUTPUT"
echo "FIND=$FIND" >> "$OUTPUT"
echo "WATCH=$WATCH" >> "$OUTPUT"
# This will enable some related identifiers for the preprocessor
if test -n "$JEMALLOC"; then
echo "JEMALLOC=1" >> "$OUTPUT"
fi
# Indicates that jemalloc should be enabled using -ljemalloc flag
# The alternative is to porvide a direct link to the library via JEMALLOC_LIB
# and JEMALLOC_INCLUDE
if test -n "$WITH_JEMALLOC_FLAG"; then
echo "WITH_JEMALLOC_FLAG=$WITH_JEMALLOC_FLAG" >> "$OUTPUT"
fi
echo "LUA_PATH=$LUA_PATH" >> "$OUTPUT"
if test -n "$USE_FOLLY_DISTRIBUTED_MUTEX"; then
echo "USE_FOLLY_DISTRIBUTED_MUTEX=$USE_FOLLY_DISTRIBUTED_MUTEX" >> "$OUTPUT"
fi