[webkit-reviews] review granted: [Bug 82155] DFG OSR exit value recoveries should be computed lazily : [Attachment 134256] the patch

[bugzilla-daemon at webkit.org]

Gavin Barraclough <barraclough at apple.com> has granted Filip Pizlo
<fpizlo at apple.com>'s request for review:
Bug 82155: DFG OSR exit value recoveries should be computed lazily
https://bugs.webkit.org/show_bug.cgi?id=82155

Attachment 134256: the patch
https://bugs.webkit.org/attachment.cgi?id=134256&action=review

------- Additional Comments from Gavin Barraclough <barraclough at apple.com>
View in context: https://bugs.webkit.org/attachment.cgi?id=134256&action=review


It seems that it might be cleaner at some point to refactor GenerationInfo, so
that rather than being a class representing a single value & having a Vector of
them, to instead have GenerationInfo be a container holding info for N
variables, and make all methods be passed the virtual register to be operated
on.  Then the generation info could hold a reference to m_stream as a member,
which would better encapsulate the implementation.

Apologies again on the delay on reviewing this, r+, please re-test performance
before landing.

> Source/JavaScriptCore/bytecode/Operands.h:199
> +    // Use const-cast because:

:-(

> Source/JavaScriptCore/dfg/DFGMinifiedNode.h:42
> +    switch (type) {

Given that this happens around the node generation loop, I wonder if it might
be worth trying to make this a faster test than a switch?