More check elimination improvements including partial learning on false branches of CompareMap and …

More check elimination improvements including partial learning on false branches of CompareMap and better handling of unreachable blocks.

[Igor Sheludko, 2014-02-05 15:51:52]

PTAL (the last set of changes that I have)

[Toon Verwaest, 2014-02-11 14:54:03]

As discussed, I'd prefer an interface with just a Merge operation without a Copy
operation. That way unreachable blocks only need to be handled by pretending
that they are in the initial state.

https://codereview.chromium.org/140543004/diff/1/src/hydrogen-check-eliminati...
File src/hydrogen-check-elimination.cc (right):

https://codereview.chromium.org/140543004/diff/1/src/hydrogen-check-eliminati...
src/hydrogen-check-elimination.cc:167: if (pred_block->IsReachable() &&
succ->predecessors()->length() == 1) {
I think we shouldn't explicitly care about the number of predecessors, but
rather just copy information from all reachable blocks down and merge it in the
regular way. If the length of reachable predecessors is one, it will
automagically do what you expect.

[Igor Sheludko, 2014-02-11 22:44:34]

On 2014/02/11 14:54:03, Toon Verwaest wrote:
> As discussed, I'd prefer an interface with just a Merge operation without a
Copy
> operation. That way unreachable blocks only need to be handled by pretending
> that they are in the initial state.
> 
>
https://codereview.chromium.org/140543004/diff/1/src/hydrogen-check-eliminati...
> File src/hydrogen-check-elimination.cc (right):
> 
>
https://codereview.chromium.org/140543004/diff/1/src/hydrogen-check-eliminati...
> src/hydrogen-check-elimination.cc:167: if (pred_block->IsReachable() &&
> succ->predecessors()->length() == 1) {
> I think we shouldn't explicitly care about the number of predecessors, but
> rather just copy information from all reachable blocks down and merge it in
the
> regular way. If the length of reachable predecessors is one, it will
> automagically do what you expect.

I decided to redo my changes in a different CL that was already landed:
https://codereview.chromium.org/159963002/