This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]


Joern RENNECKE <> wrote:
> Well, you could run a benchmark.
> Unless there are actually significant performance differences, hardware
> would probably not be much use - the differences are likely to be in the
> noise.  sh-elf-run can give you an approximate cycle count, for 
> something that
> is more or less the SH2 or SH3 pipeline, I think.  It doesn't emulate 
> caches,
> which is actually good for our purposes, since random alignments cause
> extra noise.
> I also made a patch to the simulator some time ago to support approximate
> cycle counting for the SH4 pipeline.  It does slow down the general 
> operation
> of the simulator slightly, though.  I thought I had sent it, but 
> unfortunately,
> I couldn't find it with a web search - which might or might not be due to a
> hole in the archives.  If I have sent it to a relevant FSF mainling list,
> it's Copyright FSF, otherwise, it is Copyright STM / Renesas.  At any
> rate, you can use it under the GPL.

It's very useful for not only this particular case but also
more general class of problems.  Thanks!  I'll come up with
some benchmark numbers.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]