Bug 151747 – Polymorphic operand types for DFG and FTL div.

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Mark Lam]

Patch coming.

[Radar WebKit Bug Importer]

<rdar://problem/23725121>

[Mark Lam]

*** Bug 151749 has been marked as a duplicate of this bug. ***

[Mark Lam]

Created attachment 266846 [details]
x86_64 benchmark results with FTL.

[Mark Lam]

Created attachment 266847 [details]
x86_64 benchmark results with DFG only.

[Mark Lam]

Created attachment 266848 [details]
x86 benchmark results with DFG.

[Mark Lam]

Created attachment 266851 [details]
proposed patch.

[WebKit Commit Bot]

Attachment 266851 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Wrong number of spaces before statement. (expected: 8)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Comma should be at the beginning of the line in a member initialization list.  [whitespace/init] [4]
Total errors found: 2 in 20 files


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

[Build Bot]

Comment on attachment 266851 [details]
proposed patch.

Attachment 266851 [details] did not pass mac-ews (mac):
Output: http://webkit-queues.webkit.org/results/531793

New failing tests:
svg/in-html/sizing/svg-inline.html

[Build Bot]

Created attachment 266856 [details]
Archive of layout-test-results from ews102 for mac-yosemite

The attached test failures were seen while running run-webkit-tests on the mac-ews.
Bot: ews102  Port: mac-yosemite  Platform: Mac OS X 10.10.5

[Build Bot]

Comment on attachment 266851 [details]
proposed patch.

Attachment 266851 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-queues.webkit.org/results/531797

New failing tests:
svg/in-html/sizing/svg-inline.html

[Build Bot]

Created attachment 266858 [details]
Archive of layout-test-results from ews105 for mac-yosemite-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews105  Port: mac-yosemite-wk2  Platform: Mac OS X 10.10.5

[Build Bot]

Comment on attachment 266851 [details]
proposed patch.

Attachment 266851 [details] did not pass mac-debug-ews (mac):
Output: http://webkit-queues.webkit.org/results/531826

New failing tests:
svg/in-html/sizing/svg-inline.html

[Build Bot]

Created attachment 266860 [details]
Archive of layout-test-results from ews114 for mac-yosemite

The attached test failures were seen while running run-webkit-tests on the mac-debug-ews.
Bot: ews114  Port: mac-yosemite  Platform: Mac OS X 10.10.5

[Mark Lam]

Comment on attachment 266851 [details]
proposed patch.

Taking off review while I investigate the EWS failure.

[Mark Lam]

Created attachment 266918 [details]
proposed patch 2: fixed EWS failure.

The EWS crashed test is due https://bugs.webkit.org/show_bug.cgi?id=151997.  Applied a workaround in this patch.

[WebKit Commit Bot]

Attachment 266918 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp:3754:  Should have only a single space after a punctuation in a comment.  [whitespace/comments] [5]
ERROR: Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp:1985:  Should have only a single space after a punctuation in a comment.  [whitespace/comments] [5]
ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Wrong number of spaces before statement. (expected: 8)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Comma should be at the beginning of the line in a member initialization list.  [whitespace/init] [4]
Total errors found: 4 in 20 files


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

[Geoffrey Garen]

Comment on attachment 266918 [details]
proposed patch 2: fixed EWS failure.

View in context: https://bugs.webkit.org/attachment.cgi?id=266918&action=review

> Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp:1972
> +            if (leftOperand.isConst()) {
> +                // FIXME: https://bugs.webkit.org/show_bug.cgi?id=151997
> +                // The DFG should have folded a division of a const by a const into a const result,
> +                // but doesn't.  For now, we work around it by treating the rightChild as not const.
> +                // We only saw this issue manifest with the DFG backend, but this check is added here
> +                // as a conservative defense until the issue is in the DFG front end.
> +            } else if (rightChild->isInt32Constant())
> +                rightOperand.setConstInt32(rightChild->asInt32());
> +#if USE(JSVALUE64)
> +            else if (rightChild->isDoubleConstant())
> +                rightOperand.setConstDouble(rightChild->asNumber());

Do all of our snippets make this assumption that they will never see two constants?

I don't think this is a good assumption to make. Thought the DFG often optimizes constants, it does not promise to.

[Mark Lam]

Created attachment 266920 [details]
patch 3: fixed valid style issues.

[WebKit Commit Bot]

Attachment 266920 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Wrong number of spaces before statement. (expected: 8)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Comma should be at the beginning of the line in a member initialization list.  [whitespace/init] [4]
Total errors found: 2 in 20 files


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

[Mark Lam]

Created attachment 266923 [details]
patch 4: Changed comment about 2 operands being const.

[WebKit Commit Bot]

Attachment 266923 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Wrong number of spaces before statement. (expected: 8)  [whitespace/indent] [4]
ERROR: Source/JavaScriptCore/ftl/FTLInlineCacheDescriptorInlines.h:40:  Comma should be at the beginning of the line in a member initialization list.  [whitespace/init] [4]
Total errors found: 2 in 20 files


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

[Geoffrey Garen]

Comment on attachment 266923 [details]
patch 4: Changed comment about 2 operands being const.

r=me

[Mark Lam]

Thanks for the review.  Landed in r193781: <http://trac.webkit.org/r193781>.