Bug 122740 - FTL should have an inefficient but correct implementation of GetById
Summary: FTL should have an inefficient but correct implementation of GetById
Status: RESOLVED FIXED
Alias: None
Product: WebKit
Classification: Unclassified
Component: JavaScriptCore (show other bugs)
Version: 528+ (Nightly build)
Hardware: All All
: P2 Normal
Assignee: Filip Pizlo
URL:
Keywords:
Depends on:
Blocks: 122739
  Show dependency tree / graph
 
Reported: 2013-10-13 21:57 PDT by Filip Pizlo
Modified: 2013-10-14 11:06 PDT (History)
9 users (show)

See Also:

Attachments
the patch (6.01 KB, patch)
2013-10-13 22:21 PDT, Filip Pizlo 		mhahnenberg: review+
Details | Formatted Diff | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Filip Pizlo 2013-10-13 21:57:22 PDT 
Just to expand coverage and find bugs before I implement this for real.
Comment 1 Filip Pizlo 2013-10-13 22:21:38 PDT 
Created attachment 214126 [details]
the patch
Comment 2 Nadav Rotem 2013-10-13 23:07:45 PDT 
I see two changes: 1. remove prediction  2. implement getByID   Are these changes related ? 

What is J_JITOperation_EJI ?

The line below is really long. Can you extract the identifier number into a separate variable ?

 1231        setJSValue(vmCall(m_out.operation(operationGetById), m_callFrame, base, m_out.constIntPtr(m_graph.identifiers()[m_node->identifierNumber()])));
Comment 3 Mark Hahnenberg 2013-10-14 10:33:46 PDT 
Comment on attachment 214126 [details]
the patch

r=me
Comment 4 Filip Pizlo 2013-10-14 10:41:24 PDT 
(In reply to comment #2)
> I see two changes: 1. remove prediction  2. implement getByID   Are these changes related ? 

Loosely - I implemented the FTL version of GetById by looking at what the DFG version did, but of course removing all of the inline cache stuff.  Then I saw that the DFG had this assertion about predictions.  I had a choice of either adding such an assertion in the FTL, or removing it.  So I removed it.

> 
> What is J_JITOperation_EJI ?

It's a typedef:

typedef EncodedJSValue JIT_OPERATION (*J_JITOperation_EJI)(ExecState*, EncodedJSValue, StringImpl*);

See JITOperations.h.

> 
> The line below is really long. Can you extract the identifier number into a separate variable ?
> 
>  1231        setJSValue(vmCall(m_out.operation(operationGetById), m_callFrame, base, m_out.constIntPtr(m_graph.identifiers()[m_node->identifierNumber()])));

Ok!
Comment 5 Filip Pizlo 2013-10-14 11:06:52 PDT 
Landed in http://trac.webkit.org/changeset/157409