[turbofan] Optimize certain chains of Branch into a Switch.

[turbofan] Optimize certain chains of Branch into a Switch.

This adds a new ControlFlowOptimizer that - for now - recognizes chains
of Branches generated by the SwitchBuilder for a subset of javascript
switches into Switch nodes. Those Switch nodes are then lowered to
either table or lookup switches.

Also rename Case to IfValue (and introduce IfDefault) for consistency.

BUG=v8:3872
LOG=n

Committed: https://crrev.com/acd9c46ca7a0ada723d717f2545fd484c824eaa2
Cr-Commit-Position: refs/heads/master@{#26691}

[Benedikt Meurer, 2015-02-17 09:24:45]

bmeurer@chromium.org changed reviewers:
+ jarin@chromium.org, mstarzinger@chromium.org

[Benedikt Meurer, 2015-02-17 09:24:46]

PTAL

[Michael Starzinger, 2015-02-17 12:23:35]

Looking good from my end. Just a couple of comments.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/control-flo...
File src/compiler/control-flow-optimizer.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/control-flo...
src/compiler/control-flow-optimizer.cc:77: if (if_true->opcode() !=
IrOpcode::kIfTrue) std::swap(if_true, if_false);
At some point I think we want to move CollectSuccessorProjections from the
Scheduler to NodeProperties so that we can use it in places like this. But that
is for a follow-up CL. Should we leave a TODO?

https://codereview.chromium.org/931623002/diff/60001/src/compiler/instruction...
File src/compiler/instruction-selector.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/instruction...
src/compiler/instruction-selector.cc:529: CheckNoPhis(default_branch);
FYI: In my CL I am hoisting the check out of the switch.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/operator.h
File src/compiler/operator.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/operator.h#...
src/compiler/operator.h:88: // TODO(bmeurer): Use bit fields below?
Please don't address this TODO in the near future before we know the saved
memory footprint is worth it. Let's keep it simple for now.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/schedule.h
File src/compiler/schedule.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/schedule.h#...
src/compiler/schedule.h:104: typedef value_type& reference;
Do we really need these two typedefs?

https://codereview.chromium.org/931623002/diff/60001/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/scheduler.c...
src/compiler/scheduler.cc:384: size_t const kIfFalseIndex = 1;
Can we just inline these values below as it was before? IMHO that was more
readable.

https://codereview.chromium.org/931623002/diff/60001/src/flag-definitions.h
File src/flag-definitions.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/flag-definitions.h#n...
src/flag-definitions.h:427: DEFINE_BOOL(turbo_cf_optimization, true, "optimize
control flow in TurboFan")
nit: Can we just call this --turbo-switch? I assume it is a temporary flag and
for now it only affects switches.

[Benedikt Meurer, 2015-02-17 12:30:39]

Comments addressed.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/control-flo...
File src/compiler/control-flow-optimizer.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/control-flo...
src/compiler/control-flow-optimizer.cc:77: if (if_true->opcode() !=
IrOpcode::kIfTrue) std::swap(if_true, if_false);
On 2015/02/17 12:23:35, Michael Starzinger wrote:
> At some point I think we want to move CollectSuccessorProjections from the
> Scheduler to NodeProperties so that we can use it in places like this. But
that
> is for a follow-up CL. Should we leave a TODO?

Done.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/instruction...
File src/compiler/instruction-selector.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/instruction...
src/compiler/instruction-selector.cc:529: CheckNoPhis(default_branch);
On 2015/02/17 12:23:35, Michael Starzinger wrote:
> FYI: In my CL I am hoisting the check out of the switch.

Acknowledged.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/operator.h
File src/compiler/operator.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/operator.h#...
src/compiler/operator.h:88: // TODO(bmeurer): Use bit fields below?
On 2015/02/17 12:23:35, Michael Starzinger wrote:
> Please don't address this TODO in the near future before we know the saved
> memory footprint is worth it. Let's keep it simple for now.

Acknowledged.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/schedule.h
File src/compiler/schedule.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/schedule.h#...
src/compiler/schedule.h:104: typedef value_type& reference;
No, removed them.

https://codereview.chromium.org/931623002/diff/60001/src/compiler/scheduler.cc
File src/compiler/scheduler.cc (right):

https://codereview.chromium.org/931623002/diff/60001/src/compiler/scheduler.c...
src/compiler/scheduler.cc:384: size_t const kIfFalseIndex = 1;
On 2015/02/17 12:23:35, Michael Starzinger wrote:
> Can we just inline these values below as it was before? IMHO that was more
> readable.

Done.

https://codereview.chromium.org/931623002/diff/60001/src/flag-definitions.h
File src/flag-definitions.h (right):

https://codereview.chromium.org/931623002/diff/60001/src/flag-definitions.h#n...
src/flag-definitions.h:427: DEFINE_BOOL(turbo_cf_optimization, true, "optimize
control flow in TurboFan")
On 2015/02/17 12:23:35, Michael Starzinger wrote:
> nit: Can we just call this --turbo-switch? I assume it is a temporary flag and
> for now it only affects switches.

Done.

[Michael Starzinger, 2015-02-17 12:53:38]

LGTM from my end.

[Jarin, 2015-02-17 13:27:20]

lgtm

[Benedikt Meurer, 2015-02-17 13:27:34]

The CQ bit was checked by bmeurer@chromium.org

[commit-bot: I haz the power, 2015-02-17 13:27:48]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/931623002/100001

[commit-bot: I haz the power, 2015-02-17 13:29:39]

Committed patchset #6 (id:100001)

[commit-bot: I haz the power, 2015-02-17 13:29:55]

Patchset 6 (id:??) landed as
https://crrev.com/acd9c46ca7a0ada723d717f2545fd484c824eaa2
Cr-Commit-Position: refs/heads/master@{#26691}