Bug 36753

Summary:	Some textures now rendered as completely black after register allocator rewrite.
Product:	Mesa	Reporter:	Chris Rankin <rankincj>
Component:	Drivers/Gallium/r300	Assignee:	Default DRI bug account <dri-devel>
Status:	RESOLVED FIXED	QA Contact:
Severity:	normal
Priority:	medium
Version:	git
Hardware:	x86 (IA32)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:
Attachments:	FP debug output Verbose patch Output of RADEON_DEBUG=nohiz,fp Possible Fix Output of RADEON_DEBUG=nohiz,fp

Description Chris Rankin 2011-05-01 15:35:48 UTC

Running World of Warcraft on my M66GL (RV530), and some textures are now rendered as completely black. However, my RV350 is unaffected.

git bisect says:

fe622bac0c1b5b9f2a9fcf9f35b51232a06bea42 is the first bad commit
commit fe622bac0c1b5b9f2a9fcf9f35b51232a06bea42
Author: Tom Stellard <tstellar@gmail.com>
Date:   Tue Jan 11 00:05:08 2011 -0800

    r300/compiler: Rewrite register allocator
    
    The new allocator uses ra and does swizzle packing.
    
    Also, a data structure (struct rc_variable) and associated functions have
    been added for generating UD and DU chains.

:040000 040000 545a13a6577705511d30b471d6901237e793679c f27924a4a5a40bab284ea7c03facab6c69875fb9 M	src

The following error messages are written to the output log:

r300 FP: Compiler Error:
Rewrite of inst 10 failed Can't allocate source for Inst 13 src_type=1 new_index=2 new_mask=1
Using a dummy shader instead.
r300 FP: Compiler Error:
Rewrite of inst 14 failed Can't allocate source for Inst 17 src_type=1 new_index=2 new_mask=1
Using a dummy shader instead.

Comment 1 Tom Stellard 2011-05-01 17:22:17 UTC

Can you run the program with RADEON_DEBUG=fp and post the output.

Comment 2 Chris Rankin 2011-05-02 06:04:46 UTC

Created attachment 46250 [details]
FP debug output

Strictly speaking, this output is from RADEON_DEBUG=nohiz,fp because HiZ is also broken with M66GL chips at the moment.

Comment 3 Tom Stellard 2011-05-04 00:28:04 UTC

(In reply to comment #2)
> Created an attachment (id=46250) [details]
> FP debug output
> 
> Strictly speaking, this output is from RADEON_DEBUG=nohiz,fp because HiZ is
> also broken with M66GL chips at the moment.

I'm working on a proper fix for this.  In the mean time, running with RADEON_DEBUG=noopt should fix the problem.

Comment 4 Chris Rankin 2011-05-05 13:14:55 UTC

(In reply to comment #3)
> In the mean time, running with RADEON_DEBUG=noopt should fix the problem.

Indeed it does, thanks.

Comment 5 Tom Stellard 2011-05-15 22:01:41 UTC

Created attachment 46753 [details] [review]
Verbose patch

The patch will give me some extra debug info.  Can you apply it and post the output of RADEON_DEBUG=fp.

Comment 6 Chris Rankin 2011-05-16 16:33:46 UTC

Created attachment 46786 [details]
Output of RADEON_DEBUG=nohiz,fp

(In reply to comment #5)
> The patch will give me some extra debug info.  Can you apply it and post the
> output of RADEON_DEBUG=fp.

As requested.

Comment 7 Tom Stellard 2011-05-17 00:59:25 UTC

Created attachment 46792 [details] [review]
Possible Fix

Can you try this patch.  If it doesn't fix the issue, can you post the output of RADEON_DEBUG=fp again.

Comment 8 Chris Rankin 2011-05-17 10:14:40 UTC

Created attachment 46833 [details]
Output of RADEON_DEBUG=nohiz,fp

(In reply to comment #7)
> Can you try this patch?

That seems to have fixed it, thanks. I've attached the RADEON_DEBUG output anyway, so that you can check that it did what you were expecting.

Comment 9 Tom Stellard 2011-05-17 19:18:57 UTC

Fixed in git by commit 217cd216eac65983004ca77a9e49dbfad1b720b6.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.