Quoting Sylvain J. [email protected]:
For those who are interested, I already measured the effect of GCC
optimization flags on Ruby speed using the MatrixBenchmark, which was the
only one I had at the time. Results here:
http://www.jhaampe.org/software/ruby-gcc
Very interesting … 22 percent speedup overall. I believe, though I’d
need to
go back and check the scripts, that when I ran the original tests, I
compiled
with O3 and “march” for my machine, which is my default setting for
Gentoo
Linux. But I didn’t know about the GCC feature of recompiling after
profiling,
so I didn’t try that. It’s clearly worth doing, and when I get some free
time
I’ll put it in the benchmark.
While we’re on the subject, I haven’t forgotten the needs of our Windows
brethren. 
Does anyone here know if either the Visual Studio Express
C++ or
MinGW compilers are capable of
a) Tuning code to the architecture a la GCC’s “march”, and/or
b) Inserting profiling counters and recompiling on the basis of the
profiled
runs?