Teach TurboFan to call C functions with result_size > 1.

Teach TurboFan to call C functions with result_size > 1.

Happily, this fixes WIN64 failures too.

BUG=
R=titzer@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=25216

[mvstanton, 2014-11-07 14:29:34]

mvstanton@chromium.org changed reviewers:
+ titzer@chromium.org

[mvstanton, 2014-11-07 14:29:34]

Hi Ben,
This fixes most of the Win64 turbofan failures. PTAL, thx!
--Michael

[titzer, 2014-11-07 14:41:12]

https://codereview.chromium.org/709033002/diff/1/src/compiler/js-generic-lowe...
File src/compiler/js-generic-lowering.cc (right):

https://codereview.chromium.org/709033002/diff/1/src/compiler/js-generic-lowe...
src/compiler/js-generic-lowering.cc:226: CEntryStub stub(isolate(),
fun->result_size);
Can you make the result size a parameter to the JSGraph::CEntryStubConstant()
method? Then you can keep the caching (which I think is pretty important, since
we call the CEntryStub a lot in generic code) inside the JSGraph. You could,
e.g. only cache the 1 result size case.

[mvstanton, 2014-11-07 14:49:38]

Right on, adjusted, thx,
--Michael

https://codereview.chromium.org/709033002/diff/1/src/compiler/js-generic-lowe...
File src/compiler/js-generic-lowering.cc (right):

https://codereview.chromium.org/709033002/diff/1/src/compiler/js-generic-lowe...
src/compiler/js-generic-lowering.cc:226: CEntryStub stub(isolate(),
fun->result_size);
On 2014/11/07 14:41:12, titzer wrote:
> Can you make the result size a parameter to the JSGraph::CEntryStubConstant()
> method? Then you can keep the caching (which I think is pretty important,
since
> we call the CEntryStub a lot in generic code) inside the JSGraph. You could,
> e.g. only cache the 1 result size case.

Done.

[titzer, 2014-11-07 14:53:06]

https://codereview.chromium.org/709033002/diff/20001/src/compiler/js-graph.cc
File src/compiler/js-graph.cc (right):

https://codereview.chromium.org/709033002/diff/20001/src/compiler/js-graph.cc...
src/compiler/js-graph.cc:29: DCHECK(result_size == 2);
Any risk in just passing the result_size through to the CEntryStub (i.e. skip
the assert)?

[mvstanton, 2014-11-07 15:00:00]

Another update, thx.
--Michael

https://codereview.chromium.org/709033002/diff/20001/src/compiler/js-graph.cc
File src/compiler/js-graph.cc (right):

https://codereview.chromium.org/709033002/diff/20001/src/compiler/js-graph.cc...
src/compiler/js-graph.cc:29: DCHECK(result_size == 2);
On 2014/11/07 14:53:06, titzer wrote:
> Any risk in just passing the result_size through to the CEntryStub (i.e. skip
> the assert)?

Done.

[titzer, 2014-11-07 15:33:32]

lgtm

[mvstanton, 2014-11-07 15:34:51]

Committed patchset #3 (id:40001) manually as 25216 (presubmit successful).