cse = Variants[variant] ^ aTable2[variant];
printf("Variants[%d] = 0x%2.2X, aTable2[%d] = 0x%2.2X\n",
variant, Variants[variant], variant, aTable2[variant]);
printf("cse = 0x%2.2X\n", cse);
Variants = 0xCC, aTable2 = 0xDC
cse = 0x73
/me tries to work out how to attach the testcase tarball..... sod it, it's
Compiling without -fomit-frame-pointer makes it better, as does compiling
with -O2 -fomit-frame-pointer -fschedule-insns. The tarball's makefile uses
the former to produce 'test-good'.
Created attachment 4327 [details]
tarball containing gcc bug testcase.
Thanks. Verified it happens with mainline CVS as well and is ia32 specific.
Will look into what's going on.
Bernd Schmidt fixed this, the fix is in gcc-2.96-61.