LGTM!
http://codereview.chromium.org/16538/diff/206/209
File src/regexp-macro-assembler-ia32.cc (right):
http://codereview.chromium.org/16538/diff/206/209#newcode69
Line 69: * - register n ebp[-4*(n+1)]
I would prefer n - 1 here which would correspond to num_registers.
http://codereview.
I am reverting Erik's LGTM. I am sorry but I honestly think this change
is on a road to big train wreck. We should discuss in person as I think
there are multiple issues not just with this change but with some
assumptions that lead to this.
Here are some of my concerns:
- The excessive use of mem
Unfortunately neither Lasse nor I have time for a meeting after work today.
We will put this change on hold and instead Lasse will investigate what
performance hit we can expect if we stop using esp for the regexp
backtracking stack. It could well be that the penalty is much less for us
than it wa
On Wed, Jan 7, 2009 at 1:46 AM, wrote:
> I am reverting Erik's LGTM. I am sorry but I honestly think this change
> is on a road to big train wreck. We should discuss in person as I think
> there are multiple issues not just with this change but with some
> assumptions that lead to this.
>
> Here
It is perhaps needlessly complex.
Let's try to see if we can get it working using an external stack for
only the backtracking information (the thing that can grow unbounded),
and keep the system stack for small temporary data.
It'll cost us another register, to keep an extra stack pointer, push
a