Re: [gfortran,patch] Don't call library function for string copying, take 2

[FX Coudert <fxcoudert at gmail dot com>]

> The patch seems to work okay but I can only
> measure a ~13% speed improvement on the pb05 aeromod benchmark (compiled
> with -O3). Exactly how much of an improvement do you see on i686?

http://gcc.gnu.org/ml/fortran/2006-06/msg00399.html

I saw a 27% speedup in execution time for aermod compiled with -O3 
-ffast-math -march=pentium4 -funroll-loops. I don't have the time to do 
the benchmarks again, but the times I measured were not single runs, but 
averages over stable series of runs.

Most probably this speedup depends on the cost of a function call, the 
speed of access to the memory, etc. I'm not saying it's a 27% speedup 
everywhere; that's the case on my system, and furthermore I don't see 
how it could be a pessimization (we're simply getting rid of a function 
call and making more info known to the compiler).

FX