Bug 21984

Summary:

Slow cases of opcodes should use the same machine registers for their final result as fast cases

Product:

WebKit

Reporter:

Cameron Zwarich (cpst) <zwarich>

Component:

JavaScriptCore

Assignee:

Cameron Zwarich (cpst) <zwarich>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

zwarich

Priority:

Version:

528+ (Nightly build)

Hardware:

All

OS:

All

Attachments:

Description	Flags
Proposed patch	none

Cameron Zwarich (cpst)

Reported 2008-10-30 13:13:10 PDT

In order to properly resolve bug 21943, the slow cases of opcodes should use the same machine registers for their final destination as the fast cases. As far as I can tell, only op_get_by_id and op_get_by_val violate this rule, although some other opcodes without slow cases unnecessarily avoid using eax.

Attachments
Proposed patch (1.37 KB, patch) 2008-11-10 15:39 PST, Cameron Zwarich (cpst)	no flags	Details Formatted Diff Diff
View All Add attachment proposed patch, testcase, etc.

Cameron Zwarich (cpst)

Comment 1 2008-10-30 15:51:29 PDT

I am trying to fix the op_get_by_id case first, but changing the write to ecx there to use eax makes everything explode, even though it shouldn't matter. I will investigate and then attempt to fix this.

Cameron Zwarich (cpst)

Comment 2 2008-11-10 15:39:30 PST

Created attachment 25031 [details] Proposed patch Here is a patch for changing the last two cases that I know about. I will probably wait to land it with the speedup.

Cameron Zwarich (cpst)

Comment 3 2008-11-19 13:07:41 PST

This is now fixed, due to a number of changes.

Note You need to log in before you can comment on or make changes to this bug.