------- Comment #18 from bernds at gcc dot gnu dot org 2010-04-27 22:27 ------- Thanks for all the information. However, I'm still puzzled. Here's the situation.
Thanks to your information, I think I can reproduce how the assembly files are generated: ./cc1 -feliminate-unused-debug-symbols -mmacosx-version-min=10.5.8 -mdynamic-no-pic -W -Wall -Wwrite-strings -Wcast-qual -Wstrict-prototypes -Wmissing-prototypes -Wmissing-format-attribute -pedantic -Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -Werror -Wold-style-definition -Wc++-compat -fno-common -g -gtoggle -fworking-directory -fpreprocessed -O2 omega_w.i -o omega_w.s However, both in the case of real_w.i and omega_w.i, it does not seem to matter whether the ifcvt patch is changed as in comment 7: I get identical assembly from both compilers, and interestingly it's identical to your omega_f.s and real_f.s. Two questions - are you sure you didn't reverse the _w and _f files (i.e. maybe _f is the working version and _w the failing one)? Are you sure this isn't a problem with your 4.0.1 bootstrap compiler? Can you try building a different working gcc (e.g. from the 4.5 branch), and bootstrapping with that? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43858