[cfe-users] Question casually profiling Clang vs GCC
Hi, I am a long term Gentoo Linux user, which, as a source-based meta distribution, involves frequent (re-)compilation of a lot of packages from scratch. Curious how Clang would perform against the system's gcc-7.3.0-r3 in terms of compilation speed, I installed clang-5.0.2 (latest stable on Gentoo) and compiled a handful of random packages: Python, unzip, libopus, exim, nginx, postresql. Setting was -O2, LTO was disabled. Surprisingly, Clang consistently took considerably longer to compile these packages than GCC, ranging from +28% (unzip) to +54% (postgresql), with a median increase of +40% in compile time. That contradicts the notion that Clang would be on par or quicker than GCC in terms of compilation time. Is that expected behaviour? May I have been missing some vital switch or setup? Any hints appreciated, and I'll gladly provide more information. Kind regards, Florian ___ cfe-users mailing list cfe-users@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-users
Re: [cfe-users] Question casually profiling Clang vs GCC
Hi, very interesting. My experience with normal compilation is, that clang is faster. Could it be, that there are some precompiled headers or similar present on the system, that GCC utilizes? Best Regards, Jonas Am 01.08.2018 um 13:23 schrieb Florian Berger via cfe-users: > Hi, > > I am a long term Gentoo Linux user, which, as a source-based meta > distribution, involves frequent (re-)compilation of a lot of packages > from scratch. > > Curious how Clang would perform against the system's gcc-7.3.0-r3 in > terms of compilation speed, I installed clang-5.0.2 (latest stable on > Gentoo) and compiled a handful of random packages: Python, unzip, > libopus, exim, nginx, postresql. Setting was -O2, LTO was disabled. > > Surprisingly, Clang consistently took considerably longer to compile > these packages than GCC, ranging from +28% (unzip) to +54% (postgresql), > with a median increase of +40% in compile time. That contradicts the > notion that Clang would be on par or quicker than GCC in terms of > compilation time. > > Is that expected behaviour? May I have been missing some vital switch or > setup? > > Any hints appreciated, and I'll gladly provide more information. > > Kind regards, > > Florian > ___ > cfe-users mailing list > cfe-users@lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-users ___ cfe-users mailing list cfe-users@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-users
Re: [cfe-users] Question casually profiling Clang vs GCC
On 01.08.2018 16:29, Jonas Toth via cfe-users wrote: > > very interesting. My experience with normal compilation is, that clang > is faster. Here's the data. It's from a single run each, so it's nowhere near a statistically sound sample. Package GCC Clang Difference dev-lang/python-2.7.152:31,640s 3:36,891s +43% app-arch/unzip-6.0_p21-r2 0:07,859s 0:10,094s +28% media-libs/opus-1.2.1 0:34,218s 0:46,750s +37% mail-mta/exim-4.91-r2 0:39,779s 0:54,672s +37% www-servers/nginx-1.12.2-r1 0:43,378s 1:03,537s +46% dev-db/postgresql-10.34:03,518s 6:14,859s +54% media-gfx/inkscape-0.92.228:21,146s 34:52,015s +23% > Could it be, that there are some precompiled headers or similar present > on the system, that GCC utilizes? That is possible. Where and how would I check that? Cheers, Florian ___ cfe-users mailing list cfe-users@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-users
