Bug 122739 – FTL should use llvm.webkit.patchpoint for inline caches

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Filip Pizlo]

...

[Filip Pizlo]

Created attachment 214690 [details]
it begins

But every time I think I can start rage-hacking this into existence, I find another set of prerequisites that I have to handle in another patch.

[Filip Pizlo]

Created attachment 214779 [details]
so much fun still to be had!

[Filip Pizlo]

Created attachment 214795 [details]
more!

[Filip Pizlo]

Created attachment 214806 [details]
slow path call stuff is written

[Filip Pizlo]

Created attachment 214811 [details]
I think I wrote all of the code

[Filip Pizlo]

Created attachment 214812 [details]
for real this time

Haven't tried compiling it yet.

[Filip Pizlo]

Created attachment 214813 [details]
starting to compile

[Filip Pizlo]

Created attachment 214819 [details]
it compiles

[Filip Pizlo]

Created attachment 214821 [details]
starting to work?

[WebKit Commit Bot]

Attachment 214821 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/JavaScriptCore.xcodeproj/project.pbxproj', u'Source/JavaScriptCore/assembler/AbstractMacroAssembler.h', u'Source/JavaScriptCore/assembler/LinkBuffer.cpp', u'Source/JavaScriptCore/assembler/LinkBuffer.h', u'Source/JavaScriptCore/ftl/FTLAbbreviations.h', u'Source/JavaScriptCore/ftl/FTLCapabilities.cpp', u'Source/JavaScriptCore/ftl/FTLCompile.cpp', u'Source/JavaScriptCore/ftl/FTLInlineCacheDescriptor.h', u'Source/JavaScriptCore/ftl/FTLInlineCacheSize.cpp', u'Source/JavaScriptCore/ftl/FTLInlineCacheSize.h', u'Source/JavaScriptCore/ftl/FTLIntrinsicRepository.h', u'Source/JavaScriptCore/ftl/FTLJITFinalizer.cpp', u'Source/JavaScriptCore/ftl/FTLJITFinalizer.h', u'Source/JavaScriptCore/ftl/FTLLocation.cpp', u'Source/JavaScriptCore/ftl/FTLLocation.h', u'Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp', u'Source/JavaScriptCore/ftl/FTLOutput.h', u'Source/JavaScriptCore/ftl/FTLSlowPathCall.cpp', u'Source/JavaScriptCore/ftl/FTLSlowPathCall.h', u'Source/JavaScriptCore/ftl/FTLSlowPathCallKey.cpp', u'Source/JavaScriptCore/ftl/FTLSlowPathCallKey.h', u'Source/JavaScriptCore/ftl/FTLStackMaps.cpp', u'Source/JavaScriptCore/ftl/FTLStackMaps.h', u'Source/JavaScriptCore/ftl/FTLState.h', u'Source/JavaScriptCore/ftl/FTLThunks.cpp', u'Source/JavaScriptCore/ftl/FTLThunks.h', u'Source/JavaScriptCore/jit/CCallHelpers.h', u'Source/JavaScriptCore/jit/GPRInfo.h', u'Source/JavaScriptCore/jit/JITInlineCacheGenerator.cpp', u'Source/JavaScriptCore/jit/JITInlineCacheGenerator.h', u'Source/JavaScriptCore/jit/RegisterSet.cpp', u'Source/JavaScriptCore/jit/RegisterSet.h', u'Source/JavaScriptCore/runtime/Options.h', u'Source/WTF/wtf/BitVector.cpp', u'Source/WTF/wtf/BitVector.h', u'Source/WTF/wtf/HashTraits.h', u'Source/WTF/wtf/StdLibExtras.h']" exit_code: 1
Source/WTF/wtf/BitVector.h:271:  An else statement can be removed when the prior "if" concludes with a return, break, continue or goto statement.  [readability/control_flow] [4]
Source/JavaScriptCore/ftl/FTLSlowPathCall.h:60:  The parameter name "object" adds no information, so it should be removed.  [readability/parameter_name] [5]
Source/JavaScriptCore/ftl/FTLSlowPathCall.cpp:41:  When wrapping a line, only indent 4 spaces.  [whitespace/indent] [3]
Total errors found: 3 in 37 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Build Bot]

Comment on attachment 214821 [details]
starting to work?

Attachment 214821 [details] did not pass win-ews (win):
Output: http://webkit-queues.appspot.com/results/8858264

[Filip Pizlo]

This is going to take some effort to get right.  Just as I did with OSR, I think I'll first land the prototype that can do some simple inline cache and then file bugs for the things that need to be done before this passes all tests.

[Filip Pizlo]

This looks like it works as of http://trac.webkit.org/changeset/157872.

Leaving this bug open since the corresponding LLVM support hasn't landed yet.

[Filip Pizlo]

It's in trunk.  The remaining work to port the implementation to the slight changes between the original LLVM patch and what landed will be covered in https://bugs.webkit.org/show_bug.cgi?id=122318.