On Thu, 17 Oct 2024 at 20:05, Benjamin Berg <benja...@sipsolutions.net> wrote: > > Hi, > > On Thu, 2024-10-17 at 10:18 +0200, Johannes Berg wrote: > > [SNIP] > > > > I wonder now if the SSE instructions generated are memset() and that > > goes away with the patches that Nathan just sent to not have the memset > > (which was due to -ftrivial-auto-var-init) in the first place? > > I am guessing it would not fix it. I can reproduce the problem if I add > __attribute__((aligned(16))) > to the "stack" variable. That is sufficient for GCC to emit an > offending movaps instruction for me. The -ftrivial-auto-var-init= > option will not affect that initialization. >
Yeah, that seems to be what's happening here: the SSE use is all in the 'stack' and 'sa' variable initialisation. Nathan's patch doesn't fix it for me, but then again, I can't reproduce it with clang, just gcc 13.2.0. > > But anyway all of these pretty much sound reasonable. We don't need much > > to happen in the stub here, simpler is better. > > Yeah, and I guess c) really is sufficient by itself. I've sent (c) out here: https://lore.kernel.org/linux-um/20241017231007.1500497-2-david...@google.com/T/#u Thanks, -- David
smime.p7s
Description: S/MIME Cryptographic Signature
