On Thu, May 1, 2025 at 11:41 AM Toon Moene <t...@moene.org> wrote: > > Compare: > > https://gcc.gnu.org/pipermail/gcc-testresults/2025-April/845489.html > > (checking=yes,extra,rtl,df,gcac - shortly over 2 hours and 15 minutes) > > with: > > https://gcc.gnu.org/pipermail/gcc-testresults/2025-May/845705.html > > (checking=yes,extra,rtl,df,gcac,fold - almost 7 hours wall clock time). > > A large part of the difference is explained by one of the emit-insn-N.cc > files takes over an hour to compile (twice, of course) - mostly on its > own ... > > In the manual "checking=fold" is not indicated as one of the checking > options that takes a lot of time ... but that was perhaps in the dim and > distant past.
Yes, that is a dim and very distant past. Before tree-ssa really. fold/fold_build is called a lot more instead of once per statement/expression. Plus fold checking is expensive due to the way it needs to compare each field if the hash is different. and it has to do a deep copy of the tree. It is definitely not cheap. Thanks, Andrew Pinski > > Aside from that, the split of emit-insn.cc on AArch64 seems suboptimal - > perhaps someone can point me to the algorithm behind the splitting so > that I can experiment with it ? > > Thanks in advance, > > -- > Toon Moene - e-mail: t...@moene.org - phone: +31 346 214290 > Saturnushof 14, 3738 XG Maartensdijk, The Netherlands >
