Use stability to only conditionally flush information from the CheckMaps table.

Use stability to only conditionally flush information from the CheckMaps table.

BUG=
R=ishell@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=19330

[Toon Verwaest, 2014-02-11 18:07:45]

PTAL

[Igor Sheludko, 2014-02-12 11:39:40]

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-instruction...
src/hydrogen-instructions.h:2707: }
If all but the last map we are adding are stable then we will add dependency
even if the instruction is not stable. Is it ok?

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
File test/cctest/test-deoptimization.cc (right):

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
test/cctest/test-deoptimization.cc:616: // CHECK_EQ(0,
Deoptimizer::GetDeoptimizedCodeCount(CcTest::i_isolate()));
What if we disable check elimination for this test instead of commenting this?

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
test/cctest/test-deoptimization.cc:698: // CHECK_EQ(0,
Deoptimizer::GetDeoptimizedCodeCount(CcTest::i_isolate()));
Same as above.

[Igor Sheludko, 2014-02-12 12:58:20]

More comments.

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
File src/hydrogen-check-elimination.cc (right):

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:212: // If the other state is empty, simply
reset.
Reset();

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:443: void Kill() {
KillNonStableEntries(); ?

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:638: bool requires_reset_ : 1;
Consider replacing these flags with GVNFlagSet as was recently done in load
elimination (r19322).

[Toon Verwaest, 2014-02-12 14:53:33]

Addressed comments

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
File src/hydrogen-check-elimination.cc (right):

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:212: // If the other state is empty, simply
reset.
On 2014/02/12 12:58:20, Igor Sheludko wrote:
> Reset();

Done.

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:443: void Kill() {
On 2014/02/12 12:58:20, Igor Sheludko wrote:
> KillNonStableEntries(); ?

Done.

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-check-elimi...
src/hydrogen-check-elimination.cc:638: bool requires_reset_ : 1;
On 2014/02/12 12:58:20, Igor Sheludko wrote:
> Consider replacing these flags with GVNFlagSet as was recently done in load
> elimination (r19322).

Done.

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/153823003/diff/60001/src/hydrogen-instruction...
src/hydrogen-instructions.h:2707: }
I think it's a fine approximation for now. We can try to optimize later when it
becomes necessary.

On 2014/02/12 11:39:40, Igor Sheludko wrote:
> If all but the last map we are adding are stable then we will add dependency
> even if the instruction is not stable. Is it ok?

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
File test/cctest/test-deoptimization.cc (right):

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
test/cctest/test-deoptimization.cc:616: // CHECK_EQ(0,
Deoptimizer::GetDeoptimizedCodeCount(CcTest::i_isolate()));
On 2014/02/12 11:39:40, Igor Sheludko wrote:
> What if we disable check elimination for this test instead of commenting this?

Done.

https://codereview.chromium.org/153823003/diff/60001/test/cctest/test-deoptim...
test/cctest/test-deoptimization.cc:698: // CHECK_EQ(0,
Deoptimizer::GetDeoptimizedCodeCount(CcTest::i_isolate()));
On 2014/02/12 11:39:40, Igor Sheludko wrote:
> Same as above.

Done.

[Igor Sheludko, 2014-02-12 15:01:39]

lgtm

[Toon Verwaest, 2014-02-12 15:07:50]

Committed patchset #5 manually as r19330 (presubmit successful).