[lra] patch to solve most scalability problems for LRA

[Steven Bosscher]

On Wed, Oct 10, 2012 at 10:14 PM, Vladimir Makarov <vmakarov@redhat.com> wrote:
>   It is also interesting that your IRA range patch results in
> different code generation (i can not explain it too now). I saw the same
> on a small test (black jack playing and betting strategy).

I haven't looked into this, but I'm guessing this is because functions
whose semantics depend on the number of program points, like
build_conflict_bit_table, will behave differently.

For example, with my patch there are significantly fewer program
points and as a result OBJECT_MIN/OBJECT_MAX take on smaller values.
For build_conflict_bit_table this leads to a different cost metric to
decide whether the conflict table is small enough to build. Probably
other functions respond similarly to my patch, but I don't know IRA
well enough to say :-)

In any case, bootstrap&test passed (with and without checking enabled)
on x86_64-unknown-linux-gnu and powerpc64-unknown-linux-gnu so I
intend to commit the patch (with a few comment typos fixed) today or
tomorrow.

Ciao!
Steven