Bug 85269 – DFG should be able to compute dominators

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Filip Pizlo]

This is starting to be essential because of the need to optimize create_arguments, create_activation, among others. The point is that if these opcodes appear on paths that dominate the exit, then we should hoist them and have them execute unconditionally.

[Filip Pizlo]

Created attachment 139599 [details]
the patch

[Oliver Hunt]

Comment on attachment 139599 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=139599&action=review

r=me, but i'd prefer it if you made the m_array member of FastBitVector an OwnArrayPtr

> Source/WTF/wtf/FastBitVector.h:163
> +    uint32_t* m_array;

Can't you use an OwnArrayPtr here, and save the manual management of this array's lifetime?

[Filip Pizlo]

Landed with Oliver's suggestions in http://trac.webkit.org/changeset/115754

[Filip Pizlo]

Merged in http://trac.webkit.org/changeset/117861

[Csaba Osztrogonác]

- It broke the 32 bit build, so I fixed it - http://trac.webkit.org/changeset/117905
- It made almost all tests crash on Qt, new bug report on it: https://bugs.webkit.org/show_bug.cgi?id=87082 - Could you check it, please?

[Filip Pizlo]

(In reply to comment #5)
> - It broke the 32 bit build, so I fixed it - http://trac.webkit.org/changeset/117905
> - It made almost all tests crash on Qt, new bug report on it: https://bugs.webkit.org/show_bug.cgi?id=87082 - Could you check it, please?

Looking!

[Filip Pizlo]

(In reply to comment #2)
> (From update of attachment 139599 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=139599&action=review
> 
> r=me, but i'd prefer it if you made the m_array member of FastBitVector an OwnArrayPtr
> 
> > Source/WTF/wtf/FastBitVector.h:163
> > +    uint32_t* m_array;
> 
> Can't you use an OwnArrayPtr here, and save the manual management of this array's lifetime?

No, I can't, because OwnArrayPtr uses delete[].