On Tue, May 18, 2021 at 10:51:23PM +0200, Arnd Bergmann wrote: > > > zstd looks very similar to lz4. > > > End result: at a minimum, I'd suggest using > > "-fno-tree-loop-vectorize", although somebody should check that NEON > > case. > > > And I still think that using O3 for anything halfway complicated > > should be considered odd and need some strong numbers to enable. > > Agreed. I think there is a fairly strong case for just using -O2 on lz4 > and backport that to stable. > Searching for lz4 bugs with -O3 also finds several reports including > one that I sent myself: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65709 > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69702 > > I see that user space zstd is built with -O3 in Debian, but it the changelog > also lists "Improved : better speed on clang and gcc -O2, thanks to Eric > Biggers", so maybe Eric has some useful ideas on whether we should > just use -O2 for the in-kernel version. > In my opinion, -O2 is a good default even for compression code. I generally don't see any benefit from -O3 in compression code I've written. That being said, -O2 is what I usually use during development. Other people could write code that relies on -O3 to be optimized well. The Makefiles for lz4 and zstd use -O3 by default, which is a little concerning. I do expect that they're still well-written enough to do well with -O2 too, but it would require doing benchmarks to tell for sure. (As Arnd noted, it happens that I did do such benchmarks on zstd about 5 years ago, and I found an issue where some functions weren't marked inline when they should be, causing them to be inlined at -O3 but not at -O2. That got fixed.) - Eric
