On Saturday 15 March 2003 15:41, Dhruba Bandopadhyay wrote: > > (2) How can I make my Cflag above more aggressive to give better > performance? At the moment, it is quite basic and safe but I'm pretty > sure more can be done with it. Anyone have a P4 and can vouch for > additional optimisations? >
Sometimes speed increase doesnt mean better performance. For an example why would you want a speed increase in mozilla?Probably the answer is you wouldnt notice the difference since it doesnt use much cpu (coff coff) and the kind of speed you want is bandwidth dependent.In this case smaller code size generation is preferred because mozilla eats lots of mem. Also you should think if you want the-fastest-system-of-all-times-that-even-makes-my-cpu-work-as-a-lighter but where all the aplications crash all the time due to bad generated code or if you want a robust and fast system. <Now probably what you want to read :)> omit-frame-pointer increases code size but frees a register in the cpu letting gcc do more optimizations which is good. From my experience (and some old readings) unroll-loops doesnt mean speed or better performance.Maybe im wrong. prefetch-loop-arrays for some reason makes my system unstable.And I'm not the only one noticing it. If you're looking for speed... ...surprise!...-O3 isnt the best due to the overhead of instructions (push's pop's and jump's) and can make the code slower, bigger and memory-hungry.Functions that have inline stated in the code are inlined with =>-O1 so... Best Regards, Joćo Seabra -- [EMAIL PROTECTED] mailing list
