Bug 151321 – B3 should be be clever about choosing which child to reuse for result in two-operand commutative operations

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Filip Pizlo]

If we have:

    Int32 @3 = Add(@1, @2)

and we know that @1 will die, then we should lower this to:

    Move %tmp1, %tmp3
    Add32 %tmp2, %tmp3

But if we know that @2 will die, then we should lower this to:

    Move %tmp2, %tmp3
    Add32 %tmp1, %tmp3

[Filip Pizlo]

Created attachment 266214 [details]
I wrote some code

[Filip Pizlo]

Created attachment 266227 [details]
almost happy with it

This is a regression on imaging-gaussian-blur because it commutes something that really should not have been commuted.  I'm thinking about what to do.

[Filip Pizlo]

Created attachment 266229 [details]
the patch

[Filip Pizlo]

Comment on attachment 266229 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=266229&action=review

> Source/JavaScriptCore/dfg/DFGCommon.h:41
> -#define FTL_USES_B3 0
> +#define FTL_USES_B3 1

I will revert, I promise!

[WebKit Commit Bot]

Attachment 266229 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/b3/B3PhiChildren.h:154:  Place brace on its own line for function definitions.  [whitespace/braces] [4]
Total errors found: 1 in 9 files


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

[Build Bot]

Comment on attachment 266229 [details]
the patch

Attachment 266229 [details] did not pass mac-ews (mac):
Output: http://webkit-queues.webkit.org/results/494819

Number of test failures exceeded the failure limit.

[Build Bot]

Created attachment 266231 [details]
Archive of layout-test-results from ews100 for mac-yosemite

The attached test failures were seen while running run-webkit-tests on the mac-ews.
Bot: ews100  Port: mac-yosemite  Platform: Mac OS X 10.10.5

[Build Bot]

Comment on attachment 266229 [details]
the patch

Attachment 266229 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://webkit-queues.webkit.org/results/494816

Number of test failures exceeded the failure limit.

[Build Bot]

Created attachment 266232 [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 266229 [details]
the patch

Attachment 266229 [details] did not pass mac-debug-ews (mac):
Output: http://webkit-queues.webkit.org/results/494812

Number of test failures exceeded the failure limit.

[Build Bot]

Created attachment 266233 [details]
Archive of layout-test-results from ews117 for mac-yosemite

The attached test failures were seen while running run-webkit-tests on the mac-debug-ews.
Bot: ews117  Port: mac-yosemite  Platform: Mac OS X 10.10.5

[Filip Pizlo]

Created attachment 266251 [details]
the patch

[WebKit Commit Bot]

Attachment 266251 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/b3/B3PhiChildren.h:154:  Place brace on its own line for function definitions.  [whitespace/braces] [4]
Total errors found: 1 in 8 files


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

[Geoffrey Garen]

Comment on attachment 266251 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=266251&action=review

r=me

> Source/JavaScriptCore/ChangeLog:8
> +        When lowering a commutative operation two a two-operand instruction, you have a choice of which

to

> Source/JavaScriptCore/b3/B3LowerToAir.cpp:542
> +        // then we can flip things around to held coalescing, which then kills the move instruction.

help

> Source/JavaScriptCore/b3/B3PhiChildren.h:109
> +        UpsilonCollection()
> +        {
> +        }

Do we use this null initializer? I couldn't find its use.

If we do use it, we should probably null-initialize our data members.

[Filip Pizlo]

(In reply to comment #14)
> Comment on attachment 266251 [details]
> the patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=266251&action=review
> 
> r=me
> 
> > Source/JavaScriptCore/ChangeLog:8
> > +        When lowering a commutative operation two a two-operand instruction, you have a choice of which
> 
> to
> 
> > Source/JavaScriptCore/b3/B3LowerToAir.cpp:542
> > +        // then we can flip things around to held coalescing, which then kills the move instruction.
> 
> help
> 
> > Source/JavaScriptCore/b3/B3PhiChildren.h:109
> > +        UpsilonCollection()
> > +        {
> > +        }
> 
> Do we use this null initializer? I couldn't find its use.
> 
> If we do use it, we should probably null-initialize our data members.

I added the new-style { } initializers.  I like it when these classes have a null initializer because then you can use them more freely, like:

PhiChildren::UpsilonCollection x;
if (things)
    x = phis[index];
else
    x = otherPhis[index];

[Filip Pizlo]

Landed in http://trac.webkit.org/changeset/192816