Bug 37724 - r300g with HyperZ/Z compression: occlusion queries are messed up in ut2004 (regression, bisected)
Summary: r300g with HyperZ/Z compression: occlusion queries are messed up in ut2004 (r...
Status: RESOLVED FIXED
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/Gallium/r300 (show other bugs)
Version: unspecified
Hardware: Other All
: medium normal
Assignee: Default DRI bug account
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-05-29 10:17 UTC by almos
Modified: 2017-03-11 08:44 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:

Attachments
possible fix (983 bytes, patch)
2011-05-29 10:57 UTC, Marek Olšák 		Details | Splinter Review
possible fix 2 (990 bytes, patch)
2011-05-30 04:03 UTC, Marek Olšák 		Details | Splinter Review
piglit w/o hyperz (1.29 MB, application/x-bzip2)
2017-02-26 16:55 UTC, cosiekvfj 		Details
piglit hyperz (1.29 MB, application/x-bzip2)
2017-02-27 13:05 UTC, cosiekvfj 		Details
mesa 17.0.1 kernel 4.10.1 (1.29 MB, application/x-bzip)
2017-03-11 08:43 UTC, cosiekvfj 		Details
mesa 17.0.1 kernel 4.10.1 hyperz on (1.29 MB, application/x-bzip)
2017-03-11 08:44 UTC, cosiekvfj 		Details
Show Obsolete (1) View All

Description almos 2011-05-29 10:17:25 UTC 
The majority of the 3d objects disappear periodically since this commit:
f76787b3eae3f0b8af839fabfb24b57715a017f6 is the first bad commit
commit f76787b3eae3f0b8af839fabfb24b57715a017f6
Author: Marek Olšák <maraeo@gmail.com>
Date:   Sun May 29 04:36:36 2011 +0200

    r300g: fix occlusion queries when depth test is disabled or zbuffer is missing
    
    From now on, depth test is always enabled in hardware.
    
    If depth test is disabled in Gallium, the hardware Z function is set to ALWAYS.
    
    If there is no zbuffer set, the colorbuffer0 memory is set as a zbuffer
    to silence the CS checker.
    
    This fixes piglit:
    - occlusion-query-discard
    - NV_conditional_render/bitmap
    - NV_conditional_render/drawpixels
    - NV_conditional_render/vertex_array

:040000 040000 baeff41ffed8952cbb1666d04941c6d5d01ca4fc cdb64f4b684804b818df4b65c04109eaad568e11 M	src
Comment 1 Marek Olšák 2011-05-29 10:57:11 UTC 
Created attachment 47288 [details] [review]
possible fix

Can you try this patch?
Comment 2 almos 2011-05-29 12:41:57 UTC 
(In reply to comment #1)
> Created an attachment (id=47288) [details]
> possible fix
> 
> Can you try this patch?
More things remain visible with the patch, but it doesn't fix the issue completely.
Comment 3 Marek Olšák 2011-05-30 04:03:13 UTC 
Created attachment 47307 [details] [review]
possible fix 2

Can you try this patch instead?
Comment 4 almos 2011-05-30 11:31:28 UTC 
(In reply to comment #3)
> Created an attachment (id=47307) [details]
> possible fix 2
> 
> Can you try this patch instead?
No apparent change compared to the first fix.

Now I see that I haven't mentioned that I use RADEON_HYPERZ=1. Without it there are no rendering errors.
Comment 5 Marek Olšák 2011-05-30 13:59:23 UTC 
Ah ok.

I don't consider Hyper-Z ready yet but thanks for testing it.
Comment 6 Fabio Pedretti 2011-06-09 04:37:34 UTC 
> I don't consider Hyper-Z ready yet but thanks for testing it.

It is enabled by default now, is it by accident?
Comment 7 Marek Olšák 2011-06-09 07:21:05 UTC 
(In reply to comment #6)
> > I don't consider Hyper-Z ready yet but thanks for testing it.
> 
> It is enabled by default now, is it by accident?

It's disabled. There are some annoying bugs and I have no clue how to fix them.
Comment 8 almos 2011-06-09 07:59:43 UTC 
(In reply to comment #6)
> > I don't consider Hyper-Z ready yet but thanks for testing it.
> 
> It is enabled by default now, is it by accident?
Perhaps you mean that it always reports this?
r300: AA compression RAM: YES, Z compression RAM: YES, HiZ RAM: NO

It used to mean that those features are enabled, but now it means that those are available in HW. Now if hyperz is enabled, these are printed:
radeon: Acquired Hyper-Z.
radeon: Released Hyper-Z.
Comment 9 Fabio Pedretti 2011-06-10 00:47:52 UTC 
(In reply to comment #8)
> (In reply to comment #6)
> > > I don't consider Hyper-Z ready yet but thanks for testing it.
> > 
> > It is enabled by default now, is it by accident?
> Perhaps you mean that it always reports this?
> r300: AA compression RAM: YES, Z compression RAM: YES, HiZ RAM: NO
> 
> It used to mean that those features are enabled, but now it means that those
> are available in HW. Now if hyperz is enabled, these are printed:
> radeon: Acquired Hyper-Z.
> radeon: Released Hyper-Z.

Yes, I see this:
r300: DRM version: 2.8.0, Name: ATI RV530, ID: 0x71c5, GB: 1, Z: 2
r300: GART size: 509 MB, VRAM size: 256 MB
r300: AA compression RAM: YES, Z compression RAM: YES, HiZ RAM: YES

I didn't noticed the
radeon: Acquired Hyper-Z.
radeon: Released Hyper-Z.
that indeed get printed when RADEON_HYPERZ=1 is set, however I also noticed that with openarena I get the same FPS with or without it.
Comment 10 Pavel Ondračka 2012-01-05 14:04:19 UTC 
This isn't actually limited to ut2004. I did notice almost completely black scenes in Lightsmark, broken lightning and black objects in Cogs and occasional black screen in HoN with RADEON_HYPERZ=1, all broken by f76787b3eae3f0b8af839fabfb24b57715a017f6. (RV530 here)

Any chance a revert can happen?
Comment 11 Marek Olšák 2012-01-05 15:14:59 UTC 
The problematic commit fixes occlusion queries for some cases, so I'm sorry but I can't revert it just because it breaks an optimization which is not even enabled by default. I can take a look at the issue and see if Z compression is fixable.
Comment 12 Pavel Ondračka 2012-01-05 15:54:31 UTC 
(In reply to comment #11)
> The problematic commit fixes occlusion queries for some cases, so I'm sorry but
> I can't revert it just because it breaks an optimization which is not even
> enabled by default. I can take a look at the issue and see if Z compression is
> fixable.

OK, I see revert is not an option. It still would be nice if this could be fixed, but it's not a priority. 

BTW I did quick check of r300g bugs and did find only three about HyperZ including this one. This doesn't seems like a large number but it may be only a lack of testing. Is there a possibility to enable just some features on some generations? For example there are no bug reports about hiz on r500 cards... I could do more testing if needed, maybe some piglit runs.
Comment 13 Marek Olšák 2012-01-05 16:03:28 UTC 
I am sure bug 35095 also affects r500.
Comment 14 Marek Olšák 2012-01-05 16:18:14 UTC 
Well maybe not, I am not sure.

I recall there was another bug with HiZ, which could be triggered by changing the depth function. The workaround was to flush the zbuffer cache when the depth function was changed. However flushing the cache is a very costly operation, so it wasn't even an option.

I just gave up after that, because it made no ******* sense. Zbuffer cache should have nothing to with HiZ.
Comment 15 Marek Olšák 2012-12-02 06:11:29 UTC 
I reverted the commit and I also fixed the bug with Lightsmark. So far everything seems to work. Maybe it's time to enable HyperZ by default?
Comment 16 cosiekvfj 2017-02-06 23:23:48 UTC 
Is it enabled by default now?
Comment 17 Pavel Ondračka 2017-02-07 06:43:59 UTC 
(In reply to cosiekvfj from comment #16)
> Is it enabled by default now?

IIRC the HyperZ should be enabled by default on r500 but was never enabled on r300 and r400 due to lack of testing.
Comment 18 cosiekvfj 2017-02-26 16:55:58 UTC 
Created attachment 129931 [details]
piglit w/o hyperz

I'm testing mesa with patch for this bug: https://bugs.freedesktop.org/show_bug.cgi?id=98869
And also dri3 is related to: https://bugs.freedesktop.org/show_bug.cgi?id=99970

Test results:
dri3 w/o patch 890 fps
dri3 w/o patch 1113 fps hyperz=1
dri3 patch 912 fps
dri3 patch 1115 hyperz=1
dri2 w/o patch 1136
dri2 w/o patch 1417 hyperz=1
dri2 patch 1138
dri2 patch 1417 hyperz=1
Comment 19 cosiekvfj 2017-02-27 13:05:02 UTC 
Created attachment 129952 [details]
piglit hyperz

Above results are from glxgears.
Comment 20 cosiekvfj 2017-02-27 19:57:42 UTC 
Is there anything else needed?
If you want more tests or you want me to redo previous tests or if you are unsure about results, I'm happy to help.
Comment 21 cosiekvfj 2017-03-11 08:43:50 UTC 
Created attachment 130162 [details]
mesa 17.0.1 kernel 4.10.1
Comment 22 cosiekvfj 2017-03-11 08:44:52 UTC 
Created attachment 130163 [details]
mesa 17.0.1 kernel 4.10.1 hyperz on

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.