This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
RE: 2.19 -enable-targets=all one error in m68k-dis.c due to longjmp, easily fixed with volatile
I'll try the patch "later".
This is on Debian/GNU/Linux/IA64/etch/4.0r?4? => gcc 4.1.x I think.
I understand the deoptimization of volatile..but actually..if you mark just a parameter or local variable as volatile..on a non-register-starved platforms..i.e. anything but x86..it's not a deoptimization at all, eh? Or hardly? I'll have to look later. You know, if the canonical location of the variable is in a registry and not on the stack, then every unoptimized use will just be to use that register. Granted, register pressure..but really...and for m68k-dis? I think just putting in volatile an acceptable change. I'll check later if non-x86 code is even deoptimized by it.
I understand x86 would suffer from lack of enregistration.
Possibly some of the uses could use the non-volatile version.
Thanks,
- Jay