wilson at tuliptree dot org writes: > The resulting patch works for this testcase. I suspect we will get > better PIC FP code also, since we aren't needlessly forcing non-XFmode > constants into memory. This gives a medium size patch that will need a > lot of testing, and may not be appropriate for the gcc-3.3.1 release. > > I can't really test this with m68k crosses, and I don't have access to > native m68k hardware.
bootstrapped on m68k-linux and i386-linux (together with your patch for PR10021) with no new regressions on the 3.3 branch.