gcc -O2 -march=pentium3 -mfpmath=sse -msse -ffast-math -fomit-frame-pointer -DETCDIR=\"/etc/postfish\" -DXTHREADS -D_REENTRANT -DXUSE_MTSAFE_API -I/usr/include/gtk-2.0 -I/usr/lib/gtk-2.0/include -I/usr/X11R6/include -I/usr/include/atk-1.0 -I/usr/include/pango-1.0 -I/usr/include/freetype2 -I/usr/include/freetype2/config -I/usr/include/glib-2.0 -I/usr/lib/glib-2.0/include -DG_DISABLE_DEPRECATED -DGDK_DISABLE_DEPRECATED -DGTK_DISABLE_DEPRECATED -DGDK_PIXBUF_DISABLE_DEPRECATED -DUGLY_IEEE754_FLOAT32_HACK=1 -c -o singlecomp.o singlecomp.c singlecomp.c: In function `work_and_lapping': singlecomp.c:483: error: insn does not satisfy its constraints: (insn 2597 1866 545 27 0x406476e0 (set (reg/v:SF 29 emm0 [312]) (const_double:SF 0 [0x0] 0.0 [0x0.0p+0])) 61 {*movsf_1} (nil) (nil)) singlecomp.c:483: internal compiler error: in copyprop_hardreg_forward_1, at regrename.c:1601 USE="gtk gnome qt kde dvd alsa cdr matrox mmx sse apache2 f77 java gcj mbox cups foomaticdb ppds usb" Reproducible: Always Steps to Reproduce: 1. Download the code tarball from the above URL with the edited Makefile 2. make 3. error comes out Actual Results: see "Details" Expected Results: compile through, of course :-) The error does not occur when omitting -mfpmath=sse.
Forgot to mention gcc version: gcc (GCC) 3.3.4 20040623 (Gentoo Linux 3.3.4-r1, ssp-3.3.2-2, pie-8.7.6)
See bug 63922. I think that this bug will have exactly the same solution: filter the -msse and -msse2 flags.
Hmm. That's not what I meant. The distribution of the application comes without -msse flag in the Makefile. But it heavily uses floating point calculations, all single-precision, and I wanted to use the extra speed by utilizing SSE. Thus I added -msse and -mfpmath=sse. I expected that it behaves the same in compilation and runtime but perhaps with better performance. The bug that was mentioned in the comment above seems to have been fixed in a newer release, which is not yet the default revision of gcc in gentoo. So I will probably have to wait until it's out.
I'll say that this bug then depends on a newer gcc being unmasked. Any virtual bugs we can make this depend on related to that (the same kinds of things bugs.gentoo.org does with gnome-2.8.0/2.8.1 for instance)
Assuming it works with latest versions. Please reopen if not.