The gcc-12 build is failing due to FbgemmSparseDenseInt8Avx2. : ‘mask_int32_v’ may be used uninitialized [-Werror=maybe-uninitialized]
jayagami opened this issue · 4 comments
jayagami commented
Failed with gcc12, passed with gcc11
Error:
[ 9%] Building CXX object CMakeFiles/fbgemm_avx2.dir/src/FbgemmI8DepthwisePerChannelQuantAvx2.cc.o
[ 9%] Building CXX object CMakeFiles/fbgemm_avx2.dir/src/FbgemmSparseDenseAvx2.cc.o
[ 9%] Building CXX object CMakeFiles/fbgemm_avx2.dir/src/FbgemmSparseDenseInt8Avx2.cc.o
In file included from /usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include/immintrin.h:47,
from /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:13:
In function ‘void _mm256_maskstore_epi32(int*, __m256i, __m256i)’,
inlined from ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = true; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::TENSOR]’ at /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:193:35:
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include/avx2intrin.h:1155:32: error: ‘mask_int32_v’ may be used uninitialized [-Werror=maybe-uninitialized] 1155 | __builtin_ia32_maskstored256 ((__v8si *)__X, (__v8si)__M, (__v8si)__Y);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc: In function ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = true; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::TENSOR]’:
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:168:19: note: ‘mask_int32_v’ was declared here
168 | __m256i mask_int32_v;
| ^~~~~~~~~~~~
In function ‘void _mm256_maskstore_epi32(int*, __m256i, __m256i)’, inlined from ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = true; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::OUT_CHANNEL]’ at /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:193:35: /usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include/avx2intrin.h:1155:32: error: ‘mask_int32_v’ may be used uninitialized [-Werror=maybe-uninitialized] 1155 | __builtin_ia32_maskstored256 ((__v8si *)__X, (__v8si)__M, (__v8si)__Y);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc: In function ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = true; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::OUT_CHANNEL]’:
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:168:19: note: ‘mask_int32_v’ was declared here
168 | __m256i mask_int32_v;
| ^~~~~~~~~~~~
In function ‘void _mm256_maskstore_epi32(int*, __m256i, __m256i)’,
inlined from ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = false; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::TENSOR]’ at /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:193:35:
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include/avx2intrin.h:1155:32: error: ‘mask_int32_v’ may be used uninitialized [-Werror=maybe-uninitialized] 1155 | __builtin_ia32_maskstored256 ((__v8si *)__X, (__v8si)__M, (__v8si)__Y);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc: In function ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = false; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::TENSOR]’:
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:168:19: note: ‘mask_int32_v’ was declared here
168 | __m256i mask_int32_v;
| ^~~~~~~~~~~~
In function ‘void _mm256_maskstore_epi32(int*, __m256i, __m256i)’,
inlined from ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = false; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::OUT_CHANNEL]’ at /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:193:35:
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include/avx2intrin.h:1155:32: error: ‘mask_int32_v’ may be used uninitialized [-Werror=maybe-uninitialized] 1155 | __builtin_ia32_maskstored256 ((__v8si *)__X, (__v8si)__M, (__v8si)__Y); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ /home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc: In function ‘void fbgemm::internal::SparseDenseInt8MMAvx2(int, const std::unique_ptr<fbgemm::BCSRMatrix<> >&, const uint8_t*, int, int32_t*, uint8_t*, int, fbgemm::trRequantizationParams_t&, bool, int, int) [with bool FUSE_RELU = false; fbgemm::QuantizationGranularity Q_GRAN = fbgemm::QuantizationGranularity::OUT_CHANNEL]’:
/home/git/FBGEMM/src/FbgemmSparseDenseInt8Avx2.cc:168:19: note: ‘mask_int32_v’ was declared here
168 | __m256i mask_int32_v;
| ^~~~~~~~~~~~
cc1plus: all warnings being treated as errors
make[2]: *** [CMakeFiles/fbgemm_avx2.dir/build.make:174: CMakeFiles/fbgemm_avx2.dir/src/FbgemmSparseDenseInt8Avx2.cc.o] Error 1 make[1]: *** [CMakeFiles/Makefile2:368: CMakeFiles/fbgemm_avx2.dir/all] Error 2
make: *** [Makefile:146: all] Error 2
Env:
Linux ya 6.2.8-1-MANJARO #1 SMP PREEMPT_DYNAMIC Wed Mar 22 21:14:50 UTC 2023 x86_64 GNU/Linux
Using built-in specs.
COLLECT_GCC=cpp
Target: x86_64-pc-linux-gnu
Configured with: /build/gcc/src/gcc/configure --enable-languages=c,c++,ada,fortran,go,lto,objc,obj-c++,d --enable-bootstrap --prefix=/usr --libdir=/usr/lib --libexecdir=/usr/lib --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=https://bugs.archlinux.org/ --with-build-config=bootstrap-lto --with-linker-hash-style=gnu --with-system-zlib --enable-__cxa_atexit --enable-cet=auto --enable-checking=release --enable-clocale=gnu --enable-default-pie --enable-default-ssp --enable-gnu-indirect-function --enable-gnu-unique-object --enable-libstdcxx-backtrace --enable-link-serialization=1 --enable-linker-build-id --enable-lto --enable-multilib --enable-plugin --enable-shared --enable-threads=posix --disable-libssp --disable-libstdcxx-pch --disable-werror
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.2.1 20230201 (GCC)
COLLECT_GCC_OPTIONS='-E' '-v' '-mtune=generic' '-march=x86-64'
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/cc1 -E -quiet -v - -mtune=generic -march=x86-64 -dumpbase -
ignoring nonexistent directory "/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/../../../../x86_64-pc-linux-gnu/include"
#include "..." search starts here:
#include <...> search starts here:
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include
/usr/local/include
/usr/lib/gcc/x86_64-pc-linux-gnu/12.2.1/include-fixed
/usr/include
q10 commented
Hi @jayagami, thanks for reporting this issue!
Currently, our offiicial build instructions specify gcc 10, because it supports the lowest common denominator in terms of glibc and glibcxx versions on older Linux distributions (namely Ubuntu 20.04 and CentOS 7). Since these errors appear to be just uninitialized-variable errors, we will fix within the next few days (currently we are backlogged with other developments).
In the meantime, we recommend building inside a Conda environment, in which you can install a local version of gcc 10 (see our full build instructions here). Please let us know if you have any questions or concerns.
q10 commented
This uninitialized variable issue is being addressed with #1697. However, this alone will not fix the build because there a known regression in GCC 12 (see here). To work around that, you will need to add the following to your environment prior to re-running CMake:
export CFLAGS+=" -Wno-error=maybe-uninitialized -Wno-error=uninitialized -Wno-error=restrict"
export CXXFLAGS+=" -Wno-error=maybe-uninitialized -Wno-error=uninitialized -Wno-error=restrict"jayagami commented
This uninitialized variable issue is being addressed with #1697. However, this alone will not fix the build because there a known regression in GCC 12 (see here). To work around that, you will need to add the following to your environment prior to re-running CMake:
export CFLAGS+=" -Wno-error=maybe-uninitialized -Wno-error=uninitialized -Wno-error=restrict" export CXXFLAGS+=" -Wno-error=maybe-uninitialized -Wno-error=uninitialized -Wno-error=restrict"
Thanks, that's great to hear, it's not a serious problem, but I think it's better to report.
Maybe I should close it.
q10 commented
Closing this issue, as we have added a note to the build instructions in the README, which should unblock users trying to build FBGEMM on gcc-12. Feel free to re-file a ticket if the issue is blocking your builds.