Re: [CMake] default release build flags
On Thu, Apr 28, 2016 at 2:21 PM, Burlen Loring wrote: > Hi Guys, > > I'd like to discuss changing the defaults of CMAKE_C/CXX_FLAGS_RELEASE on > gcc, and potentially gcc like compilers such as clang and intel. > > Currently the default is "-O3 -DNDEBUG". I would like to discuss changing > this to "-O3 -march=native -mtune=native -DNDEBUG". This change will enable > numerous optimizations and produce faster code targeted for the cpu in use > during the compilation. > As someone who was bitten by a project enabling -march=native in their CMake-based build system, and the days of debugging as this was a nested dependency built in a superbuild to create a package I hope we never make that a default. As a former Gentoo developer who has seen that -O3 is often not the best default flag for optimized builds I would actually suggest changing the default to "-O2 -DNDEBUG". The binaries are normally smaller, compile times faster, and the resulting code faster. Quoting from the wiki (https://wiki.gentoo.org/wiki/GCC_optimization): '-O3: the highest level of optimization possible. It enables optimizations that are expensive in terms of compile time and memory usage. Compiling with -O3 is not a guaranteed way to improve performance, and in fact, in many cases, can slow down a system due to larger binaries and increased memory usage. -O3 is also known to break several packages. Using -O3 is not recommended.' I think adding a RelWithNative or RelAgressiveOpt might be an option, but there are a number of things like distcc that makes applying -march=native undesirable. For those that know what they are doing certainly, but there are side effects and debugging can be tricky for the poor packagers too (who often have limited resources, and do not expect this). My $0.03 cents as someone who may have lost a little hair over this ;-) Marcus -- Powered by www.kitware.com Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Kitware offers various services to support the CMake community. For more information on each offering, please visit: CMake Support: http://cmake.org/cmake/help/support.html CMake Consulting: http://cmake.org/cmake/help/consulting.html CMake Training Courses: http://cmake.org/cmake/help/training.html Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Follow this link to subscribe/unsubscribe: http://public.kitware.com/mailman/listinfo/cmake
Re: [CMake] default release build flags
I think that changing the Release mode to generate non portable code is very nasty side-effect that would bite numerous projects down the road. I think a better avenue would be to think about adding another build configuration called ReleaseNative which would produce highly optimized non portable code. ( Name would need iteration ). Are you encountering this issue mainly when building locally or for HPC machines? I ask because maybe expanding the cross compilation infrastructure to set native flags would be a better avenue. On Thu, Apr 28, 2016 at 2:21 PM, Burlen Loring wrote: > Hi Guys, > > I'd like to discuss changing the defaults of CMAKE_C/CXX_FLAGS_RELEASE on > gcc, and potentially gcc like compilers such as clang and intel. > > Currently the default is "-O3 -DNDEBUG". I would like to discuss changing > this to "-O3 -march=native -mtune=native -DNDEBUG". This change will enable > numerous optimizations and produce faster code targeted for the cpu in use > during the compilation. > > The majority of people want cpu specific optimization enabled when compiling > code for use on their system. however many folks (specifically users of > scientific codes who are often not computer scientists) don't realize > requirement of passing march and mtune flags to enable cpu specific > instruction sets. Of course there are a small number of developers making > binaries for distribution who do not want to use cpu specific instructions, > but the number of developers make binaries for distribution is small > compared to folks who just need to compile code for their own use, and the > folks making binaries are well aware of the issues and can be expected to > know to disable such features. > > on my system here is the difference > > -03: > > $gcc -O3 -Q --help=target | grep enabled > -m64[enabled] > -m80387 [enabled] > -m96bit-long-double [enabled] > -malign-stringops [enabled] > -mfancy-math-387[enabled] > -mfentry[enabled] > -mfp-ret-in-387 [enabled] > -mglibc [enabled] > -mhard-float[enabled] > -mieee-fp [enabled] > -mlong-double-80[enabled] > -mno-sse4 [enabled] > -mpush-args [enabled] > -mred-zone [enabled] > -mstackrealign [enabled] > -mtls-direct-seg-refs [enabled] > > -O3 -march=native -mtune=native > > $gcc -O3 -march=native -mtune=native -Q --help=target | grep enabled > -m64[enabled] > -m80387 [enabled] > -m96bit-long-double [enabled] > -maes [enabled] > -malign-stringops [enabled] > -mavx [enabled] > -mcx16 [enabled] > -mf16c [enabled] > -mfancy-math-387[enabled] > -mfentry[enabled] > -mfp-ret-in-387 [enabled] > -mfsgsbase [enabled] > -mfxsr [enabled] > -mglibc [enabled] > -mhard-float[enabled] > -mieee-fp [enabled] > -mlong-double-80[enabled] > -mmmx [enabled] > -mpclmul[enabled] > -mpopcnt[enabled] > -mpush-args [enabled] > -mrdrnd [enabled] > -mred-zone [enabled] > -msahf [enabled] > -msse [enabled] > -msse2 [enabled] > -msse3 [enabled] > -msse4 [enabled] > -msse4.1[enabled] > -msse4.2[enabled] > -mssse3 [enabled] > -mstackrealign [enabled] > -mtls-direct-seg-refs [enabled] > -mxsave [enabled] > -mxsaveopt [enabled] > > notice specifically the lack of sse and avx in -O3! these instruction sets > play a major role in providing computational performance in modern cpus, and > that's why I think they should be enable by default in cmake Release builds. > clang and intel also support these flags, although there may be better > alternatives. > > What do you think? > > Burlen > > -- > > Powered by www.kitware.com > > Please keep messages on-topic and check the CMake FAQ at: > http://www.cmake.org/Wiki/CMake_FAQ > > Kitware offers various services to
[CMake] default release build flags
Hi Guys, I'd like to discuss changing the defaults of CMAKE_C/CXX_FLAGS_RELEASE on gcc, and potentially gcc like compilers such as clang and intel. Currently the default is "-O3 -DNDEBUG". I would like to discuss changing this to "-O3 -march=native -mtune=native -DNDEBUG". This change will enable numerous optimizations and produce faster code targeted for the cpu in use during the compilation. The majority of people want cpu specific optimization enabled when compiling code for use on their system. however many folks (specifically users of scientific codes who are often not computer scientists) don't realize requirement of passing march and mtune flags to enable cpu specific instruction sets. Of course there are a small number of developers making binaries for distribution who do not want to use cpu specific instructions, but the number of developers make binaries for distribution is small compared to folks who just need to compile code for their own use, and the folks making binaries are well aware of the issues and can be expected to know to disable such features. on my system here is the difference -03: $gcc -O3 -Q --help=target | grep enabled -m64[enabled] -m80387 [enabled] -m96bit-long-double [enabled] -malign-stringops [enabled] -mfancy-math-387[enabled] -mfentry[enabled] -mfp-ret-in-387 [enabled] -mglibc [enabled] -mhard-float[enabled] -mieee-fp [enabled] -mlong-double-80[enabled] -mno-sse4 [enabled] -mpush-args [enabled] -mred-zone [enabled] -mstackrealign [enabled] -mtls-direct-seg-refs [enabled] -O3 -march=native -mtune=native $gcc -O3 -march=native -mtune=native -Q --help=target | grep enabled -m64[enabled] -m80387 [enabled] -m96bit-long-double [enabled] -maes [enabled] -malign-stringops [enabled] -mavx [enabled] -mcx16 [enabled] -mf16c [enabled] -mfancy-math-387[enabled] -mfentry[enabled] -mfp-ret-in-387 [enabled] -mfsgsbase [enabled] -mfxsr [enabled] -mglibc [enabled] -mhard-float[enabled] -mieee-fp [enabled] -mlong-double-80[enabled] -mmmx [enabled] -mpclmul[enabled] -mpopcnt[enabled] -mpush-args [enabled] -mrdrnd [enabled] -mred-zone [enabled] -msahf [enabled] -msse [enabled] -msse2 [enabled] -msse3 [enabled] -msse4 [enabled] -msse4.1[enabled] -msse4.2[enabled] -mssse3 [enabled] -mstackrealign [enabled] -mtls-direct-seg-refs [enabled] -mxsave [enabled] -mxsaveopt [enabled] notice specifically the lack of sse and avx in -O3! these instruction sets play a major role in providing computational performance in modern cpus, and that's why I think they should be enable by default in cmake Release builds. clang and intel also support these flags, although there may be better alternatives. What do you think? Burlen -- Powered by www.kitware.com Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Kitware offers various services to support the CMake community. For more information on each offering, please visit: CMake Support: http://cmake.org/cmake/help/support.html CMake Consulting: http://cmake.org/cmake/help/consulting.html CMake Training Courses: http://cmake.org/cmake/help/training.html Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Follow this link to subscribe/unsubscribe: http://public.kitware.com/mailman/listinfo/cmake