Lawrence Gold wrote: > > On Fri, Sep 28, 2001 at 04:13:11PM +0000, Adam D. Moss wrote: > > > > When I get a working build I'll submit a patch for > > appropriate makefile.unix options to Lawrence (if he > > doesn't mind :)). > > Please do. I'm very interested in giving this compiler a try.
I spent a couple of days on this and met with success but results were rather mixed! Assorted notes: * icc doesn't support the same inline asm syntax as gcc, so we lose the inline assembly (m68k asm core is okay though). * icc doesn't seem to like having its objects ranlib'd (at least for the more aggressive optimisations). * 'static __inline__' is bad in icc, so I reverted to 'static'. Haven't tried simply 'static inline' yet. * I just couldn't get a successful link with icc's -wp_ipo option. That's a shame because that's where icc's code really starts to out-run gcc's. -ipo is working, but isn't giving the same sorts of wins. * Letting icc use MMX doesn't seem to give any sort of a win. * I'm not using an intel processor, which icc seems somewhat geared towards. However, I *do* see performance wins with icc in most apps except for xmame! Performance notes: * Some games (irobot, cps2 games) get a 6-15% frame-rate boost. * Some games (eprom, paperboy, smashtv) get a 5-25% frame-rate drop. * irobot's performance in DGA1 is much poorer, despite its performance in an X window being improved. Odd. Seems specific to irobot. On the whole I'd say that more games get a drop than a boost. Why would this be? One reason would be the lack of inline assembly and __inline__. Another would be that one of icc's most interesting optimizations (-wp_ipo) just doesn't work with xmame. Finally, [x]mame has simply been written very much with djpp/gcc's particular optimization foibles in mind. I'll submit makefile.unix patches to Lawrence anyway, so people can experiment for themselves! --Adam _______________________________________________ Xmame mailing list [EMAIL PROTECTED] http://toybox.twisted.org.uk/mailman/listinfo/xmame
