[webkit-reviews] review granted: [Bug 124181] FTL should have an explicit notion of bytecode liveness : [Attachment 217073] the patch
bugzilla-daemon at webkit.org
bugzilla-daemon at webkit.org
Fri Nov 15 14:22:09 PST 2013
Mark Hahnenberg <mhahnenberg at apple.com> has granted Filip Pizlo
<fpizlo at apple.com>'s request for review:
Bug 124181: FTL should have an explicit notion of bytecode liveness
https://bugs.webkit.org/show_bug.cgi?id=124181
Attachment 217073: the patch
https://bugs.webkit.org/attachment.cgi?id=217073&action=review
------- Additional Comments from Mark Hahnenberg <mhahnenberg at apple.com>
View in context: https://bugs.webkit.org/attachment.cgi?id=217073&action=review
r=me
> Source/JavaScriptCore/dfg/DFGGraph.cpp:676
> + if (reg.offset() >= JSStack::CallFrameHeaderSize)
> + return true;
Maybe we should ASSERT this instead, since it should be impossible to take this
branch.
> Source/JavaScriptCore/dfg/DFGMovHintCountingPhase.h:45
> +// Computes VariableAccessData::m_movHintCount. Variables that are assigned
only once can get
> +// treated specially. For example, OSR exit in the FTL can allow the format
used for such
> +// nodes BasicBlock::ssa->availabiltiyAtHead/Tail. This is a forward flow
type inference
> +// over MovHints. At the time when MovHints are introduced, we were certain
that at any
> +// point in the program where a bytecode variable had been live, all of the
MovHints that
> +// flowed into that point for that variable had the same FlushFormat. This
analysis will tell
> +// you what that format would have been.
Comment seems whacky.
> Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp:3960
> + else if (direction == ForwardSpeculation) {
else?
> Source/JavaScriptCore/ftl/FTLOSRExit.cpp:76
> -
> +
Revert!
More information about the webkit-reviews
mailing list