• fpizlo@apple.com's avatar
    Teach DFG::Worklist and its clients that it may be reused for different kinds of compilations · 6931c476
    fpizlo@apple.com authored
    Reviewed by Geoffrey Garen.
    If the baseline JIT hits an OSR entry trigger into the DFG and we already have a
    DFG compilation but we've also started one or more FTL compilations, then we
    shouldn't get confused. Previously we would have gotten confused because we would
    see an in-process deferred compile (the FTL compile) and also an optimized
    replacement (the DFG code).
    If the baseline JIT hits an OSR entry trigger into the DFG and we previously
    did two things in this order: triggered a tier-up compilation from the DFG into
    the FTL, and then jettisoned the DFG code because it exited a bunch, then we
    shouldn't be confused by the presence of an in-process deferred compile (the FTL
    compile). Previously we would have waited for that compile to finish; but the more
    sensible thing to do is to let it complete and then invalidate it, while at the
    same time enqueueing a DFG compile to create a new, more valid, DFG code block.
    If the DFG JIT hits a loop OSR entry trigger (into the FTL) and it has already
    triggered an FTL compile for replacement, then it should fire off a second compile
    instead of thinking that it can wait for that one to finish. Or vice-versa. We
    need to allow for two FTL compiles to be enqueued at the same time (one for
    replacement and one for OSR entry in a loop).
    Then there's also the problem that DFG::compile() is almost certainly going to be
    the hook for triggering both DFG compiles and the two kinds of FTL compiles, but
    right now there is no way to tell it which one you want.
    This fixes these problems and removes a bunch of potential confusion by making the
    key for a compile in the DFG::Worklist be a CompilationMode (one of DFGMode,
    FTLMode, or FTLForOSREntryMode). That mode is also passed to DFG::compile().
    Awkwardly, this still leaves us in a no DFG->FTL tier-up situation - so
    DFG::compile() is always passed DFGMode and then it might do an FTL compile if
    possible. Fixing that is a bigger issue for a later changeset.
    * CMakeLists.txt:
    * GNUmakefile.list.am:
    * JavaScriptCore.vcxproj/JavaScriptCore.vcxproj:
    * JavaScriptCore.xcodeproj/project.pbxproj:
    * Target.pri:
    * bytecode/CodeBlock.cpp:
    * dfg/DFGCompilationKey.cpp: Added.
    * dfg/DFGCompilationKey.h: Added.
    * dfg/DFGCompilationMode.cpp: Added.
    * dfg/DFGCompilationMode.h: Added.
    * dfg/DFGDriver.cpp:
    * dfg/DFGDriver.h:
    * dfg/DFGPlan.cpp:
    * dfg/DFGPlan.h:
    * dfg/DFGWorklist.cpp:
    * dfg/DFGWorklist.h:
    * jit/JITStubs.cpp:
    git-svn-id: http://svn.webkit.org/repository/webkit/trunk@154854 268f45cc-cd09-0410-ab3c-d52691b4dbfc
JavaScriptCore.vcxproj 57.6 KB