720925 - IonMonkey: Add Object comparison

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[David Anderson [:dvander] - inactive, e-mail if emergency]

This is needed to run kraken-ai-astar.

Comment 1

[David Anderson [:dvander] - inactive, e-mail if emergency]

Attachment: patch

Comment 2

[David Anderson [:dvander] - inactive, e-mail if emergency]

Attachment: actual patch

Comment 3

[Sean Stangl [:sstangl]]

Comment on attachment 591548 [details] [diff] [review]
actual patch

Review of attachment 591548 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/ion/LIR-Common.h
@@ +385,1 @@
>  class LCompareI : public LInstructionHelper<1, 2, 0>

Keeping the I suffix seems misleading. LCompareIO? LCompareIvO (the "v" being "or")? Same with LCompareIAndBranch.

::: js/src/ion/Lowering.cpp
@@ +223,5 @@
>          MCompare *comp = opd->toCompare();
>          MDefinition *left = comp->getOperand(0);
>          MDefinition *right = comp->getOperand(1);
>  
> +        if (comp->specialization() == MIRType_Int32|| comp->specialization() == MIRType_Object) {

nit: space between MIRType_Int32 and ||.

::: js/src/ion/TypePolicy.h
@@ +118,5 @@
>  
>    public:
> +    ComparePolicy() : specialization_(MIRType_None)
> +    {
> +    }

nit:
ComparePolicy()
  : specialization_(MIRType_None)
{ }

::: js/src/ion/shared/CodeGenerator-x86-shared.cpp
@@ +226,5 @@
>      }
>  }
>  
> +void
> +CodeGeneratorX86Shared::emitCompare(MIRType type, const LAllocation *left, const LAllocation *right)

emitCompareIvO?

@@ +227,5 @@
>  }
>  
> +void
> +CodeGeneratorX86Shared::emitCompare(MIRType type, const LAllocation *left, const LAllocation *right)
> +{

JS_ASSERT(type == MIRType_Object || type == MIRType_Int32);

Comment 4

[David Anderson [:dvander] - inactive, e-mail if emergency]

http://hg.mozilla.org/projects/ionmonkey/rev/b2a4c499d052

Comment 5

[Guilherme Lima]

It's a bug or this patch really made IM 5.4x times faster than JM+TI? (looking at awfy)

Comment 6

[Guilherme Lima]

On 652377 testcase.

Comment 7

[Sean Stangl [:sstangl]]

(In reply to Guilherme Lima from comment #5)
> It's a bug or this patch really made IM 5.4x times faster than JM+TI?
> (looking at awfy)

It's a bug: debug builds trip an assert.

Comment 8

[David Anderson [:dvander] - inactive, e-mail if emergency]

It's also unrelated to this patch (awfy only tests on the hour) - bug 721272 fixed a typo that caused more paths to compile, exposing more bugs.