so on any MIPS R10000 system, -march=r10000 automatically implies -mfix-r10000, which causes branch-likely instructions to be emitted to deal with a known silicon erratum in early R10000 processors. It's been this way ever since I sent the patch in to add the feature to gcc in 2008. However, newer R10000's and the R12000+ CPUs shouldn't be affected by this bug anymore, so I am experimenting by disabling the fix in both the kernel and in the userland by passing -mno-fix-r10000 in CFLAGS. Problem is, both gcc and glibc strip this flag out, and I need it to stay. Longterm fix is to send a patch in to gcc to make -march=r12000 and up not enable the flag by default, but still turn it on for -march=r10000. That is on my long TODO list for MIPS.
Better to borrow from the title of #519120, as the problem is toolchain.eclass (for gcc) and glibc is likely affected by an eblit.
pretty sure neither is doing this. you most likely want to update flag-o-matic.eclass to list these random flags.
(In reply to SpanKY from comment #2) > pretty sure neither is doing this. you most likely want to update > flag-o-matic.eclass to list these random flags. That's right! I forgot we had that eclass. I was looking at gcc_do_filter_flags() in toolchain.eclass as being responsible for the flag-stripping.
Fixed in CVS.
