On Sat, May 21, 2011 at 1:31 AM, Andreas Färber <andreas.faer...@web.de> wrote:
[...]
> Has anyone evaluated reusing LLVM optimization passes for TCG? Or maybe
> GIMPL if there's an equivalent?
IMHO the qemu_ld/st semantics and the size of TB blocks
will always limit the usefulness of more involved
optimizations than what already exists in QEMU.
Basically qemu_ld/st ops prevent any code change across
them which make optimization opportunities rather low.
Considering they always take the fast path would surely
change the situation, but then the slow path would have to
be able to rebuild information; certainly possible, but
quite involved too :-)
Once this qemu_ld/st change is done, you'll the need to
get to the problem of too short TB blocks; a way to do
this is to generate traces of frequently executed blocks
and consider them as a single block[1]; that would open
the door to many more code generation improvements.
This certainly doesn't mean Kirill and Aurélien changes
are not welcome, just that I think, in the current state,
such changes will always have a limited impact.
Laurent
[1] See for instance section 2.3 of Derek Bruening's
PhD thesis about DynamoRIO
http://www.burningcutlery.com/derek/phd.html
