• fpizlo@apple.com's avatar
    Get rid of SetMyScope/SetCallee; use normal variables for the scope and callee... · 48cdafdc
    fpizlo@apple.com authored
    Get rid of SetMyScope/SetCallee; use normal variables for the scope and callee of inlined call frames of closures
    Reviewed by Oliver Hunt.
    Currently we have the DFG reserve space for inline call frames at exactly the same stack
    offsets that you would have gotten if the baseline interpreter/JIT had made the calls.
    We need to get rid of that. One of the weirder parts of this is that we have special DFG
    operations for accessing these inlined call frame headers. It's really hard for any
    analysis of DFG IR to see what the liveness of any of those frame header "variables" is;
    the liveness behaves like flushed arguments (it's all live until end of the inlinee) but
    we don't have anything like a Flush node for those special variables.
    This patch gets rid of the special operations for accessing inline call frame headers.
    GetMyScope and GetCallee still remain, and are only for accessing the machine call
    frame's scope/callee entries. The inline call frame's scope/callee now behave like
    normal variables, and have Flush behavior just like inline arguments.
    * dfg/DFGAbstractInterpreterInlines.h:
    * dfg/DFGByteCodeParser.cpp:
    * dfg/DFGCSEPhase.cpp:
    * dfg/DFGClobberize.h:
    * dfg/DFGFixupPhase.cpp:
    * dfg/DFGNodeType.h:
    * dfg/DFGPredictionPropagationPhase.cpp:
    * dfg/DFGSafeToExecute.h:
    * dfg/DFGSpeculativeJIT32_64.cpp:
    * dfg/DFGSpeculativeJIT64.cpp:
    git-svn-id: http://svn.webkit.org/repository/webkit/trunk@156594 268f45cc-cd09-0410-ab3c-d52691b4dbfc
DFGByteCodeParser.cpp 161 KB