Will ICC do a better job than GCC on AMD targets?
Specifically, an Athlon (the old kind, Thunderbird core). I'm just wondering if ICC in pentium-pro mode will do a better job than GCC in march=athlon mode. I realize this is a somewhat subjective question, but maybe someone has some insights. It seems to me that the biggest thing is that GCC could potentially use 3DNow! in march=athlon, and ICC couldn't do anything past basic MMX optimizations...but I don't know how likely this is to help in practice.
Optimize for P2 with /Qw? (I forget on ICC, for the P2, look at ICC switches that enable MMX) I'd still bet ICC will generate somewhat better code than GCC almost always as it is tuned specifically to X86 (and IA64). Vectorizing sometimes works and sometimes doesn't, it generally does better on P4 optimization. PPro,P2,P3 are very similar, vectorize may or may not help usually it does not match hand optimized code in any case ever. If you need absolute speed recode in 3DNow! and forget it.
GCC is good because it compiles everywhere but it is not as efficient on x86 as iCC is. Have not used GCC in long while though.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement