810946 - BaselineCompiler: Mark baseline frames

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan de Mooij [:jandem]]

Attachment: Patch

Stores the frame size without VMFunction arguments in the frame, then uses that to mark all locals and stack values.

Comment 1

[Jan de Mooij [:jandem]]

Attachment: Patch v2

Fix a typo.

Comment 2

[Kannan Vijayan [:djvj]]

Comment on attachment 680633 [details] [diff] [review]
Patch v2

Review of attachment 680633 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/ion/shared/IonFrames-x86-shared.h
@@ +450,5 @@
> +    static const uint32_t FramePointerOffset = sizeof(void *);
> +
> +    static inline size_t offsetOfFrameSize() {
> +        return -offsetof(BaselineFrame, frameSize) - sizeof(uint32_t);
> +    }

These offsets are intended to be used with reference to a pointer immediately above the BaselineFrame.  At least that's what my reading of the marking code seems to indicate.

This expression may be clearer as:
-(sizeof(BaselineFrame) - offsetOf(BaselineFrame, frameSize))

And maybe we should call these method backwardsOffsetOf... or reverseOffsetOf... ?  The offsetOf naming convention seems pretty solidly consistent to refer to the offset of things from a pointer to the start of a structure, not the end.

@@ +456,5 @@
> +        return -(sizeof(BaselineFrame) + index * sizeof(js::Value)) - sizeof(js::Value);
> +    }
> +    static inline size_t offsetOfArg(size_t index) {
> +        return FramePointerOffset + js::ion::IonJSFrameLayout::offsetOfActualArg(index);
> +    }

The above statement about renaming the offsetOf... methods wouldn't apply to offsetOfArg since it does compute a forward offset.

Comment 3

[Jan de Mooij [:jandem]]

Attachment: Dump baseline frames

Comment 4

[Jan de Mooij [:jandem]]

Pushed with the reverse* names + a comment.

https://hg.mozilla.org/projects/ionmonkey/rev/7c9cfaafa4a1

Comment 5

[Nicolas B. Pierron [:nbp]]

Comment on attachment 681014 [details] [diff] [review]
Dump baseline frames

Review of attachment 681014 [details] [diff] [review]:
-----------------------------------------------------------------

Sounds good. I am happy to see that adding the dumpBaseline function help clarify the bunch of JS_ASSERT of the Mark function.

Comment 6

[Jan de Mooij [:jandem]]

https://hg.mozilla.org/projects/ionmonkey/rev/f544c2afe9fb