This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Exploiting dual mode operation, implementation.
On Mon, 2005-08-01 at 11:38 -0700, Richard Henderson wrote:
> On Mon, Aug 01, 2005 at 02:30:32PM -0400, Daniel Berlin wrote:
> > This is exactly what redundancy elimination would have come up with.
>
> To be fair, se3 and se4 are not identical. se3 was hi->si
> and se4 was si->di.
>
> His problem is that he's forgotten that each of the conversions are
> the same cost (typically one cycle), thus the original se2 (hi->di)
> conversion is optimial.
Yeah.
If it had gotten more complex in terms of costing (non-unit costs), it
could still be modeled as a redundancy elimination problem, but solving
it would be equivalent to Stone's problem (Max-flow/min-cut).
Leehod, see Speculative PRE papers, like
http://www.cs.uvic.ca/~nigelh/Publications/LCTES04.pdf
For how they go about doing this.
Its still poly-time, but a bit expensive for what we want to do. It may
give you ideas on what to do.
Someone actually sent me gcc 3.4.3 with GCSE replaced with a speculative
PRE implementation, so it is possible (and i'm happy to share that with
whoever wants to see it. It uses a public domain stones problem solver
to do the computation).
>
>
> r~