Measuring gcc optimizations: are they well balanced?

[Horst von Brand]

Jan Hubicka <hubicka@atrey.karlin.mff.cuni.cz> said:

[...]

> It is somewhat non-naturally optimized, but equivalent in number
> of cycles on most modern IA-32 implementations to the lcc's code.

[I have been hacking a bit on lcc myself, so I understand a bit of what is
 going on here]

Even if so, it would be a shame if all the complex analysis gcc does gives
you the same performance as the very simplistic code generator in lcc. But
this case is not exactly representative code, gcc does global analysis of
the functions, which lcc doesn't do at all (and which won't help a bit for
one simple loop). For the other, lcc is absolutely unable to assign
variables to registers for the duration (been there, tried to do that, gave
up when bugs elsewhere (AFAICS) made reasonably simple schemes
unworkable). For another data point, the rcc executable (the compiler
proper for lcc, somewhat like gcc's cc1) compiled on Linux ia32 is
something like a third of the size the one lcc itself gives.

Any comparison should be made with a full benchamrk for the compilers, is
there any out there?
-- 
Horst von Brand                             vonbrand@sleipnir.valparaiso.cl
Casilla 9G, Viña del Mar, Chile                               +56 32 672616