All versions of gcc that we commonly use for building are subject to bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90348. To work around that, set -fstack-reuse=none for all builds. See #15959 (comment)
This option does not exist for clang https://clang.llvm.org/docs/ClangCommandLineReference.html#target-independent-compilation-options, but it does for gcc https://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html
Performance came back negligibly different, though I’m not totally sure this PR worked for me as intended. I double-checked to see if -fstack-reuse=none made it into the build parameters, and I don’t see it in any Makefiles after configuration (gcc (Ubuntu 5.4.0-6ubuntu1~16.04.10) 5.4.0 20160609):
0Wed 8 20:32 james/.bitcoinperf/fad88bd6c9f85e6e7f8fb66a94aa75c67d26b7d8
1$ grep -R 'fstack-reuse' .
2
3./autom4te.cache/output.0:{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether C++ compiler accepts -fstack-reuse=none" >&5
4./autom4te.cache/output.0:$as_echo_n "checking whether C++ compiler accepts -fstack-reuse=none... " >&6; }
5./autom4te.cache/output.0: CXXFLAGS="$CXXFLAGS -fstack-reuse=none"
6./autom4te.cache/output.0: HARDENED_CXXFLAGS="$HARDENED_CXXFLAGS -fstack-reuse=none"
7./autom4te.cache/traces.0:m4trace:configure.ac:737: -1- AX_CHECK_COMPILE_FLAG([-fstack-reuse=none], [HARDENED_CXXFLAGS="$HARDENED_CXXFLAGS -fstack-reuse=none"])
In any case, here’s the bench data:
| bench name | x | 1905-buildStackReuseNone | master |
|---|---|---|---|
| ibd.local.500005.505000.dbcache=2048 | 3 | 304.0416 (± 3.5953) | 312.8965 (± 4.2451) |
| ibd.local.500005.505000.dbcache=2048.mem-usage | 3 | 2463438.6667 (± 3039.5620) | 2462478.0000 (± 5874.0000) |
| build.make.7.gcc | 1 | 302.9981 (± 0.0000) | |
| build.make.7.gcc.mem-usage | 1 | 1024568.0000 (± 0.0000) | |
| ibd.local.500000.505000.dbcache=2048 | 1 | 315.0404 (± 0.0000) | |
| ibd.local.500000.505000.dbcache=2048.mem-usage | 1 | 2458576.0000 (± 0.0000) |
| bench name | x | 1905-buildStackReuseNone | master |
|---|---|---|---|
| ibd.local.500005.505000.dbcache=2048 | 3 | 1.00 | 1.029 |
| ibd.local.500005.505000.dbcache=2048.mem-usage | 3 | 1.00 | 1.000 |
| build.make.7.gcc | 1 | 1.00 | |
| build.make.7.gcc.mem-usage | 1 | 1.00 | |
| ibd.local.500000.505000.dbcache=2048 | 1 | 1.00 | |
| ibd.local.500000.505000.dbcache=2048.mem-usage | 1 | 1.00 |
It should be supported for all versions of gcc we support: https://gcc.gnu.org/onlinedocs/gcc-4.8.0/gcc/Code-Gen-Options.html
I had to run ./autogen.sh && ./configure for it to end up in the CXXFLAGS (they are printed at the end of ./configure)
Also, the results of the micro benchmarks would be useful.
0configure:23543: checking whether C++ compiler accepts -fstack-reuse=none
1configure:23562: /opt/clang90/bin/clang++ -std=c++11 -c -g -O2 -fstack-reuse=none -ggdb -DLIBEVENT_EXPERIMENTAL -DHAVE_BUILD_INFO -D__STDC_FORMAT_MACROS conftest.cpp >&5
2clang-9: error: unknown argument: '-fstack-reuse=none'
3configure:23562: $? = 1
gcc (Ubuntu 7.4.0-1ubuntu1~18.04) 7.4.00configure:23543: checking whether C++ compiler accepts -fstack-reuse=none
1configure:23562: g++ -std=c++11 -c -O2 -g -fstack-reuse=none -DHAVE_BUILD_INFO -D__STDC_FORMAT_MACROS conftest.cpp >&5
2configure:23562: $? = 0
3configure:23570: result: yes
0$ make V=1
1Making all in src
2...
3/bin/bash ../libtool --tag=CXX --preserve-dup-deps --mode=compile /usr/bin/ccache g++ -std=c++11 ...-fstack-reuse=none ...`crypto/chacha20.cpp
733@@ -734,6 +734,7 @@ if test x$TARGET_OS != xwindows; then
734 AX_CHECK_COMPILE_FLAG([-fPIC],[PIC_FLAGS="-fPIC"])
735 fi
736
737+AX_CHECK_COMPILE_FLAG([-fstack-reuse=none],[HARDENED_CXXFLAGS="$HARDENED_CXXFLAGS -fstack-reuse=none"])
733@@ -734,6 +734,10 @@ if test x$TARGET_OS != xwindows; then
734 AX_CHECK_COMPILE_FLAG([-fPIC],[PIC_FLAGS="-fPIC"])
735 fi
736
737+# All versions of gcc that we commonly use for building are subject to bug
738+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90348. To work around that, set
739+# -fstack-reuse=none for all builds.
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.
Reviewers, this pull request conflicts with the following ones:
If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.
