[Piglit] [PATCH] Add a test of the accuracy of MSAA rendering.

Anuj Phogat anuj.phogat at gmail.com
Thu Apr 19 09:11:31 PDT 2012

On Thu, Apr 19, 2012 at 8:30 PM, Paul Berry <stereotype441 at gmail.com> wrote:

> On 19 April 2012 06:41, Anuj Phogat <anuj.phogat at gmail.com> wrote:
>> On Thu, Apr 19, 2012 at 12:09 AM, Paul Berry <stereotype441 at gmail.com>wrote:
>>> This is the test I've been using to prototype MSAA on the i965
>>> driver.  Unlike the existing EXT_framebuffer_multisample
>>> tests (which so far mostly check corner cases at the API level),
>>> this test verifies that correct pixels are rendered when MSAA is
>>> enabled.  It's a bit long for a typical piglit test, since in
>>> order to verify that the MSAA color, depth, and stencil buffers
>>> are all configured correctly, it has to render some fairly
>>> complex scenes.
>>> ---
>>> This test partially overlaps the functionality of the test Anuj
>>> sent to the list on Monday ([PATCH] Add test to switch on/off
>>> MSAA in FBO).  Anuj: I'm interested in hearing how my test
>>> performs on your AMD system.
>> Here is what I get on my AMD system:
>> ./bin/ext_framebuffer_multisample-accuracy 4 color
>> Framebuffer not complete
>> PIGLIT: { 'result' : 'skip' }
>> similar error with other test_type options.
> Do you also get this result with the "depthstencil" flag?  I added that
> flag so that the test would work on my nVidia test machine, which doesn't
> seem to support separating the depth and stencil buffers.

Now I don't get "Framebuffer not complete"
$ ./bin/ext_framebuffer_multisample-accuracy 4 color depthstencil -auto
Pixels that should be unlit
  count = 214644
  Perfect output
Pixels that should be totally lit
  count = 28968
  Perfect output
Pixels that should be partially lit
  count = 18532
  RMS error = 0.260300
The error threshold for this test is 0.119880
PIGLIT: {'result': 'fail' }

I get similar RMS error value for all other cases.

>> No such error is observed on SandyBridge. But MSAA was absent as expected.
>> I couldn't get MSAA working on AMD with my test case as well.
>> I've installed ATI's catalyst drivers (Version: 12.2) on Ubuntu 10.04
>> Couldn't successfully install catalyst drivers on latest Ubuntu version:
>> 11.10.
>> Executing fglrxinfo outputs:
>> display: :0  screen: 0
>> OpenGL vendor string: Advanced Micro Devices, Inc.
>> OpenGL renderer string: AMD Radeon(TM) HD 6480G
>> OpenGL version string: 4.2.11554 Compatibility Profile Context
>> I'll try updating my system and see if that helps.
> I'd also be curious to see the output of glxinfo on your system--this will
> show which framebuffer configurations the driver claims to support MSAA
> with.
 Here is my glxinfo output: http://pastebin.com/7YeVMDC2

>>> Everyone: I'm interested in hearing
>>> opinions about whether this test is too big, and if so, how to
>>> effectively split it up into smaller tests.
>> This is a fairly comprehensive test to measure the accuracy of MSAA in
>> FBO.
>> I liked the way you are generating reference image. In my opinion there
>> are lot
>> of functions in this test which could be utilized by upcoming multisample
>> test
>> cases. e.g. the current piglit test I'm working on (Test switching MSAA
>> on/off in
>> FBO), can utilize draw_reference_image() and measure_accuracy() functions
>> to
>> verify if MSAA is actually switched ON in FBO. Every new multisample test
>> case
>> would require these functions to verify if MSAA is happening as expected.
>> So, it would be helpful to split it in to two files: common.cpp,
>> accuracy.cpp
> That's a good thought.  I'll try to figure out a sensible way to factor
> out the reusable stuff into a separate file.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/piglit/attachments/20120419/67e6ff20/attachment.htm>

More information about the Piglit mailing list