Pass PC offset into runtime when compiling for OSR.

Pass PC offset into runtime when compiling for OSR.

R=titzer@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=16649

[Yang, 2013-09-10 09:30:16]

Rebased. Please take a look.

[titzer, 2013-09-10 11:00:46]

https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc
File src/compiler.cc (right):

https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc#newcode1145
src/compiler.cc:1145: Deoptimizer::RevertInterruptCode(isolate, *unoptimized);
Please move this reversion of the interrupt code back into the runtime. The
compiler should not be responsible for doing so.

https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc#newcode1152
src/compiler.cc:1152: if (IsSuitableForOnStackReplacement(isolate, function,
unoptimized)) {
And this as well. The compiler should just compile code, not enforce policy
about when/where.

[Yang, 2013-09-10 11:19:37]

On 2013/09/10 11:00:46, titzer wrote:
> https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc
> File src/compiler.cc (right):
> 
> https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc#newcode1145
> src/compiler.cc:1145: Deoptimizer::RevertInterruptCode(isolate, *unoptimized);
> Please move this reversion of the interrupt code back into the runtime. The
> compiler should not be responsible for doing so.
> 
> https://codereview.chromium.org/23842004/diff/7010/src/compiler.cc#newcode1152
> src/compiler.cc:1152: if (IsSuitableForOnStackReplacement(isolate, function,
> unoptimized)) {
> And this as well. The compiler should just compile code, not enforce policy
> about when/where.

Those suggested refactorings are not scope of this CL, so I would like to move
that to another one.

Moving Deoptimizer::RevertInterruptCode and IsSuitableForOnStackReplacement from
CompileForOnStackReplacement to runtime is simple, but not that simple for
CompileForConcurrentOSR, since we call both functions depending on which stage
of concurrent OSR we are at. I would split CompileForConcurrentOSR into several
functions, one for each stage we are at, and call
Deoptimizer::RevertInterruptCode and IsSuitableForOnStackReplacement where
appropriate.

[Yang, 2013-09-10 14:27:50]

Addressed comment and refactored a bunch of stuff. Please take a look.

https://codereview.chromium.org/23842004/diff/19001/src/runtime-profiler.cc
File src/runtime-profiler.cc (right):

https://codereview.chromium.org/23842004/diff/19001/src/runtime-profiler.cc#n...
src/runtime-profiler.cc:150: return;
This is not actually a change in heuristic. We used to return
Handle<Code>::null() in Compiler::CompileForConcurrentOSR when recompilation
didn't finish. Runtime_CompileForOnStackReplacement would then overwrite the
builtin marker by shared code. Since that's gone now, I just decided to simulate
this by not marking a function for regular recompilation if it's already queued
for OSR.

This won't be any issue once OSR becomes one-shot, since we don't install the
optimized code onto the function in that case.

[titzer, 2013-09-11 11:02:28]

lgtm

Ahh...so much better :)

https://codereview.chromium.org/23842004/diff/19001/src/runtime-profiler.cc
File src/runtime-profiler.cc (right):

https://codereview.chromium.org/23842004/diff/19001/src/runtime-profiler.cc#n...
src/runtime-profiler.cc:246: 
Whitespace change

[Yang, 2013-09-11 12:39:26]

Committed patchset #6 manually as r16649 (presubmit successful).