Summary: | sys-cluster/ceph-10.2.5 - erasure-code/jerasure/gf-complete/src/gf_w16.c: /usr/lib/gcc/x86_64-pc-linux-gnu/4.9.3/include/tmmintrin.h:136:1: error: inlining failed in call to always_inline ‘_mm_shuffle_epi8’: target specific option mismatch | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Jeroen Roovers (RETIRED) <jer> |
Component: | Current packages | Assignee: | Patrick McLean <chutzpah> |
Status: | RESOLVED OBSOLETE | ||
Severity: | normal | CC: | cluster, dlan |
Priority: | Normal | ||
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | sys-cluster:ceph-9.2.0:20160217-062245.log |
Description
Jeroen Roovers (RETIRED)
2016-02-17 06:33:54 UTC
hi, i bet this is caused by your loooong CFLAGS, but sorry I couldn't tell which is the exact one.. could you try simple CFLAGS? here I'm using CFLAGS="-O2 -pipe -march=core2 " (In reply to Yixun Lan from comment #1) > hi, i bet this is caused by your loooong CFLAGS, but sorry I couldn't tell > which is the exact one.. Probably not. > could you try simple CFLAGS? here I'm using CFLAGS="-O2 -pipe -march=core2 " You mean, beside the endless compiler flags its build system already adds? Dropping target specific CFLAGS obviously avoids this issue. The ceph build system should not push its own CFLAGS like this. If there is a clash, then the make.conf set CFLAGS should win/override. The build system inherits any flags from make.conf then adds flags that upstream deems are necessary to make the package compile. If some flags from make.conf conflict with upstream's added flags, there is not much that can be done other than filter the offending flags. Dropping flags added by upstream is a bad idea, they are there for a reason. (In reply to Patrick McLean from comment #4) > The build system inherits any flags from make.conf then adds flags that > upstream deems are necessary to make the package compile. If some flags from > make.conf conflict with upstream's added flags, there is not much that can > be done other than filter the offending flags. > > Dropping flags added by upstream is a bad idea, they are there for a reason. We set feature specific architecture USE flags for vendor extensions all over the tree, so this package really should be no exception. I have no easy way of knowing which flags conflicting with which other flags. If you can figure out which flags are in conflict, I can probably add a filter-flags call to filter them out so ceph can compile. |