Enable optimization of functions with generic switches.

Enable optimization of functions with generic switches.

R=jkummerow@chromium.org, titzer@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=18347

[Toon Verwaest, 2013-12-09 10:33:20]

PTAL

[titzer, 2013-12-09 10:37:26]

Sending over to Jakob, as I am Sheriff this week.

[Jakob Kummerow, 2013-12-10 09:44:53]

lgtm

[Jakob Kummerow, 2013-12-10 12:23:17]

Platform ports LGTM.

[titzer, 2013-12-10 14:27:08]

https://codereview.chromium.org/110123002/diff/20001/src/hydrogen.cc
File src/hydrogen.cc (right):

https://codereview.chromium.org/110123002/diff/20001/src/hydrogen.cc#newcode4174
src/hydrogen.cc:4174: for (int i = 0; i < clause_count; ++i) {
It seems like this loop can be rewritten to go over the clauses once, generating
the compares and the body blocks in one pass. If so, then we can get rid of this
body_blocks array altogether.

https://codereview.chromium.org/110123002/diff/20001/test/mjsunit/switch-opt.js
File test/mjsunit/switch-opt.js (right):

https://codereview.chromium.org/110123002/diff/20001/test/mjsunit/switch-opt....
test/mjsunit/switch-opt.js:40: 
I would like to see several more tests for this optimization, as there seem to
be many things that can go wrong. E.g. this test doesn't even have any
fallthrough between blocks.

I would have more confidence if I saw ~10 or more different switch statements in
this test.

[titzer, 2013-12-10 17:07:27]

On 2013/12/10 14:27:08, titzer wrote:
> https://codereview.chromium.org/110123002/diff/20001/src/hydrogen.cc
> File src/hydrogen.cc (right):
> 
>
https://codereview.chromium.org/110123002/diff/20001/src/hydrogen.cc#newcode4174
> src/hydrogen.cc:4174: for (int i = 0; i < clause_count; ++i) {
> It seems like this loop can be rewritten to go over the clauses once,
generating
> the compares and the body blocks in one pass. If so, then we can get rid of
this
> body_blocks array altogether.
> 
>
https://codereview.chromium.org/110123002/diff/20001/test/mjsunit/switch-opt.js
> File test/mjsunit/switch-opt.js (right):
> 
>
https://codereview.chromium.org/110123002/diff/20001/test/mjsunit/switch-opt....
> test/mjsunit/switch-opt.js:40: 
> I would like to see several more tests for this optimization, as there seem to
> be many things that can go wrong. E.g. this test doesn't even have any
> fallthrough between blocks.
> 
> I would have more confidence if I saw ~10 or more different switch statements
in
> this test.

LGTM

But please don't land until after the Chrome branch.

[Toon Verwaest, 2013-12-18 11:44:16]

This has been sitting in our stash for too long. Landing.

[Toon Verwaest, 2013-12-18 11:45:00]

Committed patchset #3 manually as r18347.