Bug 80779 – All DFG nodes should have a mutable set of flags

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

[Filip Pizlo]

Currently some nodes have a mutable set of arithmetic use flags, and all nodes have a mostly-immutable set of flags baked into their opcode.

Instead, the opcode should be a 16-bit quantity that does not include any flags, and separately, there should be a 16-bit flags that combines the current flags-that-are-in-the-opcode and the arithmetic flags.

This will facilitate richer backward flow analyses and should make manipulating the DFG IR generally easier.

[Filip Pizlo]

<rdar://problem/11026218>

[Filip Pizlo]

Created attachment 131205 [details]
the patch

[WebKit Review Bot]

Attachment 131205 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/CMakeLists.txt', u'S..." exit_code: 1
Source/JavaScriptCore/dfg/DFGNode.h:114:  The parameter name "flags" adds no information, so it should be removed.  [readability/parameter_name] [5]
Total errors found: 1 in 17 files


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

[Early Warning System Bot]

Comment on attachment 131205 [details]
the patch

Attachment 131205 [details] did not pass qt-wk2-ews (qt):
Output: http://queues.webkit.org/results/11936304

[Early Warning System Bot]

Comment on attachment 131205 [details]
the patch

Attachment 131205 [details] did not pass qt-ews (qt):
Output: http://queues.webkit.org/results/11939311

[Filip Pizlo]

Created attachment 131206 [details]
the patch

Ooops, forgot about 32_64.  This patch fixes that.

[Filip Pizlo]

Created attachment 131273 [details]
the patch

Just rebased it.

[Gavin Barraclough]

There is an awful lot of static_cast<NodeType>(node.op) in this patch.  It would be really good to make op private (& probably should be called m_op), and change op() to be an accessor function returning a NodeType.  Would probably be nice to do the same for flags.  What do you think?

[Filip Pizlo]

(In reply to comment #8)
> There is an awful lot of static_cast<NodeType>(node.op) in this patch.  It would be really good to make op private (& probably should be called m_op), and change op() to be an accessor function returning a NodeType.  Would probably be nice to do the same for flags.  What do you think?

I'm with you!

I had avoided doing that because I had already racked up 50kb of changes.  I feared that privatizing Node::op will introduce even more changes.

I want to make that change and would be happy to do it in this patch.  But how would you feel about that being a follow-on patch?

[Gavin Barraclough]

> I want to make that change and would be happy to do it in this patch.  But how would you feel about that being a follow-on patch?

That works for me.

[Filip Pizlo]

Landed in http://trac.webkit.org/changeset/110419