Bug 69734 – Improve Null or Undefined test in 32_64 DFG

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Yuqiang Xian]

Currently Null or Undefined value test in 32_64 DFG will check Null and Undefined tag separately and introduce one more branch. It can be improved in the way how the baseline JIT is doing - by relying on the fact that "UndefinedTag + 1 == NullTag and NullTag & 1".

[Yuqiang Xian]

Created attachment 110315 [details]
the patch

[Darin Adler]

Comment on attachment 110315 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=110315&action=review

> Source/JavaScriptCore/dfg/DFGJITCodeGenerator32_64.cpp:1178
> +        ASSERT((JSValue::UndefinedTag + 1 == JSValue::NullTag) && (JSValue::NullTag & 0x1));

This assertion is written in an oblique way and could instead be written to directly mirror what the code relies on:

    COMPILE_ASSERT((JSValue::UndefinedTag | 1) == JSValue::NullTag);

> Source/JavaScriptCore/dfg/DFGSpeculativeJIT32_64.cpp:460
> +    ASSERT((JSValue::UndefinedTag + 1 == JSValue::NullTag) && (JSValue::NullTag & 0x1));

Ditto.

> Source/JavaScriptCore/dfg/DFGSpeculativeJIT32_64.cpp:560
> +    ASSERT((JSValue::UndefinedTag + 1 == JSValue::NullTag) && (JSValue::NullTag & 0x1));

Ditto.

[Filip Pizlo]

Comment on attachment 110315 [details]
the patch

Agree with Darin's comments, r=me otherwise. I'll be happy to cq+ if you can just fix that assertion.

[Yuqiang Xian]

Created attachment 110319 [details]
patch addressing Darin's comments

[WebKit Review Bot]

Comment on attachment 110319 [details]
patch addressing Darin's comments

Clearing flags on attachment: 110319

Committed r97039: <http://trac.webkit.org/changeset/97039>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.