[Mesa-dev] [PATCH] r600g: fix vs export count

[Alex Deucher]

On Fri, Jul 29, 2011 at 10:04 AM, Alex Deucher <alexdeucher at gmail.com> wrote:
> On Fri, Jul 29, 2011 at 7:51 AM, Dave Airlie <airlied at gmail.com> wrote:
>> On Fri, Jul 29, 2011 at 12:41 PM, Vadim Girlin <vadimgirlin at gmail.com> wrote:
>>> On Fri, 2011-07-29 at 12:20 +0100, Dave Airlie wrote:
>>>> On Thu, Jul 28, 2011 at 9:33 PM, Vadim Girlin <vadimgirlin at gmail.com> wrote:
>>>> > Fixes https://bugs.freedesktop.org/show_bug.cgi?id=39572
>>>>
>>>> Hi Vadim,
>>>>
>>>> has this passed a full piglit run without regression?
>>>>
>>>> Dave.
>>>
>>> Hi Dave.
>>>
>>> I've tested this patch with r600.tests on evergreen without
>>> regressions.
>>>
>>> I have no r6xx/7xx hw to test, so I'm not sure about that part, but the
>>> patch is based on the info from R6xx_3D_Registers.pdf: "VS_EXPORT_COUNT
>>> 5:1 0x0 Number of vectors exported by the VS (value is minus 1)".
>>>
>>> I have forgotten to mention that it's tested on evergreen only in the
>>> patch comment, sorry.
>>>
>>> The problem in the bug was the shader with noutput==1. I don't think
>>> "noutput-2" is correct in that case for r6xx/7xx too. Though now I think
>>> it may be safer to handle noutput==1 as special case if this patch
>>> causes problems.
>>
>> Thanks, I just had some memory of fixing this before and it causing
>> problem, though it might be it uncovered some other issue that we've
>> since fixed.
>>
>> So we can push it to master and see if it causes any problem before
>> dropping it into 7.11 I suppose.
>
> I also did a piglit run on 6xx and evergreen and there were no
> regressions.  I went ahead and pushed it to 7.11 as well when I
> committed it since it fixed a GPU hang.
>

I dug into this a bit more and pushed an improved patch:
http://cgit.freedesktop.org/mesa/mesa/commit/?id=dc1c0ca22a1c7fcaef90b787290144d8e3d77c33

Alex