Upstream code contains "always inline" instructions that require AVX Fatal compilation errors with -march=k8-sse3 Builds with -march=native on a Kaveri and bulldozer. This is an upstream bug, but I think the ebuild should test for the existence of the AVX instruction.
Thank you for the report. Please recompile and *attach* the logfiles and paste the emerge info as described on https://wiki.gentoo.org/wiki/Attach_the_logs_to_the_bug_ticket The logs must be part of the ticket, but not on external websites. We need to have all information at hand before ticket assignment, so that the maintainer can look after it in one session with minimal number of callbacks. Please reopen this ticket (Status:unconfirmed) afterwards. Please add also the output of cat /proc/cpuinfo
For what it's worth: I've built dev-qt/qtmultimedia-5.12.3 on an AMD Phenom II X6 1055T without issue and it does not have AVX instructions. -march=native on this machine as well. CPU_FLAGS_X86: mmx mmxext popcnt sse sse2 sse3 sse4a
Created attachment 582592 [details] build log Complete build log with CFLAGS="-march=k8-sse3 -O2 -pipe"
Created attachment 582594 [details] /proc/cpuinfo Builds as native but not as older cpu
(In reply to Brian Evans from comment #2) > For what it's worth: I've built dev-qt/qtmultimedia-5.12.3 on an AMD Phenom > II X6 1055T without issue and it does not have AVX instructions. > -march=native on this machine as well. CPU_FLAGS_X86: mmx mmxext popcnt sse > sse2 sse3 sse4a AVX assumption based on google search. Will attempt to build as k10. Repeat, will not build as k8, but will build as native (bulldozer/kaveri). ebuilds should build on base amd64 or check for instructions needed.
Also blows up building with march=amdfam10. Also blows up with no march (should default to minimal x86_64 works with march=bdver1
For me, COMMON_FLAGS="-O2 -pipe -march=ivybridge" -> success COMMON_FLAGS="-O2 -pipe -march=ivybridge -mno-sse3" -> fail