Add ScheduleVerifier.

Add ScheduleVerifier.

This adds a series of checks to the output of scheduling, including properties
of the RPO order, the dominance relation, phi placement, and the SSA property
that definitions dominate all their uses.

R=jarin@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=23310

[Jarin, 2014-08-22 11:55:32]

lgtm with some nits. Good stuff.

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc
File src/compiler/verifier.cc (right):

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:250: static bool FindDominatingDef(Schedule* schedule,
Node* node,
FindDominatingDef suggests that it returns the dominating def. How about
HasDominatingDef?

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:256: if (block->nodes_[use_pos] == node) return true;
It looks like the verifier would be happy if I connect a node's input to its
output. Is that ok?

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:261: use_pos = block->nodes_.size() - 1;
I guess this will not compile on Win64. Perhaps use (reverse) iterator instead
of int? Or cast...

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:275: use_pos = use_block->nodes_.size() - 1;
Again, this will not compile on Win64.

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:417: // schedules don't have control inputs.
Note: At some point we will have to fix RawMachineAssembler to produce regular
graphs (connect control and effect inputs).

[Jarin, 2014-08-22 12:05:20]

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc
File src/compiler/verifier.cc (right):

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:256: if (block->nodes_[use_pos] == node) return true;
On 2014/08/22 11:55:32, jarin wrote:
> It looks like the verifier would be happy if I connect a node's input to its
> output. Is that ok?

Ah, as discussed offline, ignore this comment. I have been confused.

[titzer, 2014-08-22 12:18:35]

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc
File src/compiler/verifier.cc (right):

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:250: static bool FindDominatingDef(Schedule* schedule,
Node* node,
On 2014/08/22 11:55:32, jarin wrote:
> FindDominatingDef suggests that it returns the dominating def. How about
> HasDominatingDef?

Done.

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:261: use_pos = block->nodes_.size() - 1;
On 2014/08/22 11:55:31, jarin wrote:
> I guess this will not compile on Win64. Perhaps use (reverse) iterator instead
> of int? Or cast...

Done.

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:275: use_pos = use_block->nodes_.size() - 1;
On 2014/08/22 11:55:31, jarin wrote:
> Again, this will not compile on Win64.

Done.

https://codereview.chromium.org/491303002/diff/1/src/compiler/verifier.cc#new...
src/compiler/verifier.cc:417: // schedules don't have control inputs.
On 2014/08/22 11:55:31, jarin wrote:
> Note: At some point we will have to fix RawMachineAssembler to produce regular
> graphs (connect control and effect inputs).

Yeah, I know.

[titzer, 2014-08-22 12:38:00]

Committed patchset #2 manually as 23310 (presubmit successful).