Bug 184236 – [DFG] Remove duplicate 32bit code more

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Yusuke Suzuki]

[DFG] Remove duplicate 32bit code more

[Yusuke Suzuki]

Created attachment 336990 [details]
Patch

[EWS Watchlist]

Attachment 336990 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:43:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:44:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
Total errors found: 2 in 12 files


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

[Yusuke Suzuki]

Created attachment 336993 [details]
Patch

[EWS Watchlist]

Attachment 336993 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:43:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:44:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
Total errors found: 2 in 12 files


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

[Yusuke Suzuki]

Created attachment 336998 [details]
Patch

[EWS Watchlist]

Attachment 336998 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:43:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/dfg/DFGCompareSlowPathGenerator.h:44:  Wrong number of spaces before statement. (expected: 12)  [whitespace/indent] [4]
Total errors found: 2 in 12 files


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

[Yusuke Suzuki]

Ping?

[Mark Lam]

Comment on attachment 336998 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=336998&action=review

> Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp:12920
> +    unblessedBooleanResult(resultGPR, node, UseChildrenCalledExplicitly);

This is not equivalent to the original.  From my reading of the code, this merely performs a bool to JSBoolean conversion.  The original code which calls jsValueResult() (see DFGSpeculativeJIT.h:855-867) which adds JIT asserts, do some tracking for register spilling, and capture GenerationInfo.  There appear to be a lot of work which are missing here.

> Source/JavaScriptCore/dfg/DFGSpeculativeJIT64.cpp:438
> +    JITCompiler::Jump twoCellsCase = m_jit.branchTest64(JITCompiler::Zero, resultGPR, GPRInfo::tagMaskRegister);

I don't see how this can work for 32-bit targets.  The tagMaskRegister only exits for x86_64 and ARM64.  What am I missing?

[Mark Lam]

(In reply to Mark Lam from comment #8)
> Comment on attachment 336998 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=336998&action=review
> 
> > Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp:12920
> > +    unblessedBooleanResult(resultGPR, node, UseChildrenCalledExplicitly);
> 
> This is not equivalent to the original.  From my reading of the code, this
> merely performs a bool to JSBoolean conversion.  The original code which
> calls jsValueResult() (see DFGSpeculativeJIT.h:855-867) which adds JIT
> asserts, do some tracking for register spilling, and capture GenerationInfo.
> There appear to be a lot of work which are missing here.

I misread the code.  I thought unblessedBooleanResult() only did blessBoolean() part for 64-bit, and not the blessedBooleanResult() part.  It actually does both.

> > Source/JavaScriptCore/dfg/DFGSpeculativeJIT64.cpp:438
> > +    JITCompiler::Jump twoCellsCase = m_jit.branchTest64(JITCompiler::Zero, resultGPR, GPRInfo::tagMaskRegister);
> 
> I don't see how this can work for 32-bit targets.  The tagMaskRegister only
> exits for x86_64 and ARM64.  What am I missing?

I misread this part too.  I thought that there's only one consolidated version of nonSpeculativeNonPeepholeStrictEq().  That's not the case: this is just the 64-bit version.  Hence, there's no issue here.

Sorry for the noise.

[Mark Lam]

Comment on attachment 336998 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=336998&action=review

r=me.  Thanks for the hard work.  This is a big patch.

> Source/JavaScriptCore/jit/RegisterSet.h:210
> +    void setAny(JSValueRegs reg) { set(reg); }

nit: lets call the argument "regs" instead of "reg".  I think this is the convention used everywhere else.

[Yusuke Suzuki]

Comment on attachment 336998 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=336998&action=review

Thank you for your review!

>> Source/JavaScriptCore/jit/RegisterSet.h:210
>> +    void setAny(JSValueRegs reg) { set(reg); }
> 
> nit: lets call the argument "regs" instead of "reg".  I think this is the convention used everywhere else.

Fixed!

[Yusuke Suzuki]

Committed r230517: <https://trac.webkit.org/changeset/230517>

[Radar WebKit Bug Importer]

<rdar://problem/39340504>

[Yusuke Suzuki]

Committed r230520: <https://trac.webkit.org/changeset/230520>