Run ControlReducer early after graph building, then again later.

Run ControlReducer early after graph building, then again later.

The justification for doing this is to reduce the size of the graph and
therefore speedup later phases of compilation. The control reducer also
obviates the need to run the PhiReducer, since it subsumes it.

R=mstarzinger@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=24986

[Michael Starzinger, 2014-10-29 12:14:03]

LGTM. Just nits.

https://codereview.chromium.org/681263004/diff/1/src/compiler/ast-graph-build...
File src/compiler/ast-graph-builder.cc (right):

https://codereview.chromium.org/681263004/diff/1/src/compiler/ast-graph-build...
src/compiler/ast-graph-builder.cc:2064: // TODO(titzer): this should be in a
JSOperatorReducer.
Yes, this should definitely be factored out at some point. This is an important
TODO, let's not forget about it. :)

https://codereview.chromium.org/681263004/diff/1/src/compiler/ast-graph-build...
src/compiler/ast-graph-builder.cc:2076: // XXX: other constants.
nit: s/XXX/TODO(turbofan)/

https://codereview.chromium.org/681263004/diff/1/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/681263004/diff/1/src/compiler/scheduler.cc#ne...
src/compiler/scheduler.cc:248: BasicBlock* block = schedule_->block(loop);
nit: This probably should use the return value of BuildBlockForNode.

[titzer, 2014-10-29 15:17:09]

https://codereview.chromium.org/681263004/diff/1/src/compiler/ast-graph-build...
File src/compiler/ast-graph-builder.cc (right):

https://codereview.chromium.org/681263004/diff/1/src/compiler/ast-graph-build...
src/compiler/ast-graph-builder.cc:2076: // XXX: other constants.
On 2014/10/29 12:14:02, Michael Starzinger wrote:
> nit: s/XXX/TODO(turbofan)/

Done.

https://codereview.chromium.org/681263004/diff/1/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/681263004/diff/1/src/compiler/scheduler.cc#ne...
src/compiler/scheduler.cc:248: BasicBlock* block = schedule_->block(loop);
On 2014/10/29 12:14:02, Michael Starzinger wrote:
> nit: This probably should use the return value of BuildBlockForNode.

Done. While doing this I noticed a bug and had to fix the glitch. PTAL.

[Michael Starzinger, 2014-10-29 15:23:21]

LGTM, just one more nit.

https://codereview.chromium.org/681263004/diff/20001/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/681263004/diff/20001/src/compiler/scheduler.c...
src/compiler/scheduler.cc:259: FixNode(schedule_->end(),
scheduler_->graph_->end());
nit: Can we fix the end node in BuildBlocks as well?

[titzer, 2014-10-29 15:24:04]

https://codereview.chromium.org/681263004/diff/20001/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/681263004/diff/20001/src/compiler/scheduler.c...
src/compiler/scheduler.cc:259: FixNode(schedule_->end(),
scheduler_->graph_->end());
On 2014/10/29 15:23:21, Michael Starzinger wrote:
> nit: Can we fix the end node in BuildBlocks as well?

Done.

[titzer, 2014-10-29 15:27:38]

Committed patchset #3 manually as 24986 (presubmit successful).