Created attachment 683443 [details] build.log Both 1.8.7 & 1.8.6 exhibit the same errors. Clang/LLVM 11.0.1 /var/tmp/portage/dev-libs/libgcrypt-1.8.7/work/libgcrypt-1.8.7/cipher/sha512-avx-amd64.S:341:2: error: invalid operand for instruction add [digest + 8*(7)], h_64 ^ /var/tmp/portage/dev-libs/libgcrypt-1.8.7/work/libgcrypt-1.8.7/cipher/sha512-avx-amd64.S:344:6: error: ambiguous operand size for instruction 'add' add msg, 16*8
Created attachment 683446 [details] emerge --info
I can only reproduce if I add -flto=thin to a amd64 build (x86 not using this asm, so). Builds fine otherwise, I did have to add USE=o-flag-munging though.
This should really go upstream.
I did test without -flto=thin, but I didn't check if the error was any different (emerge --keep-going). Looks like it just turns into https://bugs.gentoo.org/707330. Is it worth opening -flto* bugs here?
(In reply to Theo Anderson from comment #4) > I did test without -flto=thin, but I didn't check if the error was any > different (emerge --keep-going). Looks like it just turns into > https://bugs.gentoo.org/707330. Yes, that's what USE=o-flag-munging I mentioned fixes. > Is it worth opening -flto* bugs here? clang/lld issues alone doesn't seem so bad (to me), but lto or other exotic flag may be a bit much to ask for. Worse if it's asm issues. Not that I can speak for every maintainers. Still better to try and get these fixed upstream though, unless it's an issue gentoo is introducing.
Closing as fixed in 1.9.1