Compile error with fat runtime on gcc 13.2

[todd-richmond]

Just enabled fat runtime and now seeing a compile error that didn't occur when just AVX cmake options were enabled. We have custom compile flags, but these ones were ok with stock Hyperscan and it's default fat build

[ 74%] Building CXX object CMakeFiles/hs_exec_shared_avx512.dir/src/hwlm/noodle_engine.cpp.o
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/cmake/build_wrapper.sh avx512 /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/cmake/keep.syms.in /tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/bin/g++ -DNDEBUG -I/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/obj.Linux_3.10.0-x86_64 -I/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11 -I/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src -isystem /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/include -isystem /tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/boost-1_82_0/include -fPIC -O3 -pipe -ffunction-sections -fdata-sections -march=nehalem -mtune=broadwell -DNDEBUG -fvisibility-inlines-hidden -Wno-shadow -Wno-unused-local-typedefs  -I/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/include -msse4.2 -O3 -std=c++17 -Wall -Wextra -Wshadow -Wswitch -Wreturn-type -Wcast-qual -Wno-deprecated -Wnon-virtual-dtor -fno-strict-aliasing -fno-new-ttp-matching -DNDEBUG -Wno-maybe-uninitialized -Wno-abi -fno-omit-frame-pointer -fvisibility=hidden -Wno-unused-const-variable -Wno-unused-but-set-variable -Wno-ignored-attributes -Wno-redundant-move -Wmissing-declarations -Wno-stringop-overflow  -DNDEBUG -fPIC -fvisibility-inlines-hidden -march=skylake-avx512 -MD -MT CMakeFiles/hs_exec_shared_avx512.dir/src/hwlm/noodle_engine.cpp.o -MF CMakeFiles/hs_exec_shared_avx512.dir/src/hwlm/noodle_engine.cpp.o.d -o CMakeFiles/hs_exec_shared_avx512.dir/src/hwlm/noodle_engine.cpp.o -c /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/hwlm/noodle_engine.cpp

[ 74%] Building CXX object CMakeFiles/hs_exec_shared_avx512.dir/src/hwlm/noodle_engine.cpp.o
In file included from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/supervector.hpp:386,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/hwlm/noodle_engine_simd.hpp:32,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/hwlm/noodle_engine.cpp:102:
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp: In constructor ‘SuperVector<SIZE>::SuperVector(T) [with T = m256; short unsigned int SIZE = 64]’:
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp:1232:40: error: cannot convert ‘const m256’ to ‘__m256i’
 1232 |     u.v512[0] = _mm512_broadcast_i64x4(v);
      |                                        ^
      |                                        |
      |                                        const m256
In file included from /tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/gcc-13.2.0/lib64/gcc/x86_64-pc-linux-gnu/13.2.0/include/immintrin.h:53,
                 from /tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/gcc-13.2.0/lib64/gcc/x86_64-pc-linux-gnu/13.2.0/include/x86intrin.h:32,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/intrinsics.h:67,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/simd_types.h:34,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/ue2common.h:68,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/hwlm/hwlm.h:36,
                 from /var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/hwlm/noodle_engine.cpp:34:
/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/gcc-13.2.0/lib64/gcc/x86_64-pc-linux-gnu/13.2.0/include/avx512fintrin.h:4346:33: note:   initializing argument 1 of ‘__m512i _mm512_broadcast_i64x4(__m256i)’
 4346 | _mm512_broadcast_i64x4 (__m256i __A)
      |                         ~~~~~~~~^~~
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp: At global scope:
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp:1477:63: error: no member function ‘vshl_256_imm’ declared in ‘SuperVector<64>’
 1477 | really_inline SuperVector<64> SuperVector<64>::vshl_256_imm() const
      |                                                               ^~~~~
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp:1533:63: error: no member function ‘vshr_256_imm’ declared in ‘SuperVector<64>’
 1533 | really_inline SuperVector<64> SuperVector<64>::vshr_256_imm() const
      |                                                               ^~~~~
/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp:1612:74: error: no member function ‘vshl_256’ declared in ‘SuperVector<64>’
 1612 | nline SuperVector<64> SuperVector<64>::vshl_256(uint8_t const N) const
      |                                                                  ^~~~~

/var/tmp/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11/src/util/supervector/arch/x86/impl.cpp:1679:74: error: no member function ‘vshr_256’ declared in ‘SuperVector<64>’
 1679 | nline SuperVector<64> SuperVector<64>::vshr_256(uint8_t const N) const
      |                                                                  ^~~~~

make[2]: *** [CMakeFiles/hs_exec_shared_avx512.dir/build.make:160: CMakeFiles/h

[markos]

could you please paste the exact cmake command?

[markos]

Keep in mind many things have changed in terms of compilation, just because some compile flags were ok for Hyperscan, it does not mean they will work ok for Vectorscan, some of the code is in the process of heavy refactoring -like the file you just pasted- and the rest is in the list. The main reason for that is better maintainance and higher performing code, but also more portable. So, while we promise API and ABI compatibility with the original Hyperscan, there is no guarantee that you will be able to continue to use the same compile flags as Hyperscan or even the same build method. I hope you understand.

[todd-richmond]

Nothing too fancy. I use these compiler flags for 200+ open source projects we compile

CFLAGS='-fPIC -O3 -pipe -ffunction-sections -fdata-sections -march=nehalem -mtune=broadwell -DNDEBUG -Wno-shadow -Wno-unused-local-typedefs  -I/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/include'

cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/vectorscan-5.4.11  -DCMAKE_SYSTEM_INCLUDE_PATH=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/include -DCMAKE_SYSTEM_LIBRARY_PATH=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/lib64 -DCMAKE_AR=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/bin/gcc-ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_NM=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/bin/gcc-nm -DCMAKE_RANLIB=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/bin/gcc-ranlib -DCMAKE_VISIBILITY_INLINES_HIDDEN=ON -DCMAKE_POLICY_DEFAULT_CMP0063=NEW -DBUILD_AVX512=ON -DBUILD_AVX512VBMI=ON -DBUILD_BENCHMARKS=OFF -DBUILD_SHARED_LIBS=ON -DBOOST_ROOT=/tools/arch/Linux_3.10.0-x86_64/gcc-13.2.0-pps-8.21.0/boost -DFAT_RUNTIME=ON ..

[markos]

@todd-richmond ok, apologies for the delay in replying, I was able to reproduce the build failure with your cmake. You need to add -DBUILD_AVX2=ON in the cmake command, not that it matters, but I would add it right before -DBUILD_AVX512=ON. I should make this more robust so that if AVX512 is enabled, AVX2 is also enabled by default.

[markos]

Fixed in #220