|
From: Dirk M. <dm...@gm...> - 2005-06-16 11:59:34
|
On Thursday 16 June 2005 13:26, Julian Seward wrote: > Thanks for these. I forgot the glibc-2.4.supp file btw, its still in the works. > I see it decides if gcc understands -msse/-mmmx (what about > -msse2 ?) and I know that -msse2 engages gcc4's autovectoriser, > but how does that help with the regtests? the fpu and sse regression tests contain assembler sequences which reference MMX/SSE registers. gcc 4.x does not accept usage of those registers unless explicitely told about it with -mmmx / -msse. of course we could use something like -march=i[56]86 as well, because then it knows that those registers exist on the target platform. explicitely hardcoding -march=i686 is rather bad though, if you want to compile e.g. -march=pentium-m (which is "better" than i686). Dirk |