[Mesa-dev] [PATCH 1/4] xa: handle solid-fill src/mask

[Rob Clark]

btw, I think I've tracked down at least one of the rendercheck issues
I was having.  Which seems to be a pre-existing condition.  The
problem is a composite operation with xrgb src and a8 dst.  Ends up
generating a shader along the lines of:

  0: TEX TEMP[0], IN[0], SAMP[0], 2D
  1: MOV TEMP[0].w, IMM[..]
  2: MOV OUT[0], TEMP[0].wwww
  3: END

in this case, my compiler optimizes out the texture fetch.  But that
causes a bit of confusion in freedreno because half the driver thinks
there is a texture and sets things up accordingly.

I'm not entirely sure if XA should detect this.. or if that should be
fixed in the driver..

BR,
-R


On Thu, Apr 3, 2014 at 10:53 AM, Brian Paul <brianp at vmware.com> wrote:
> On 04/03/2014 08:41 AM, Brian Paul wrote:
>>
>> On 04/03/2014 08:13 AM, Rob Clark wrote:
>>>
>>> On Thu, Apr 3, 2014 at 10:07 AM, Brian Paul <brianp at vmware.com> wrote:
>>>>
>>>> On 04/02/2014 11:17 PM, Thomas Hellstrom wrote:
>>>>>
>>>>>
>>>>> On 04/01/2014 05:04 PM, Rob Clark wrote:
>>>>>>
>>>>>>
>>>>>> From: Rob Clark <robclark at freedesktop.org>
>>>>>>
>>>>>> Add support to property handle solid-fill src and/or mask.  Without
>>>>>> this
>>>>>> we fallback to sw a lot for common things like text rendering.
>>>>>>
>>>>>> Signed-off-by: Rob Clark <robclark at freedesktop.org>
>>>>>> ---
>>>>>>    src/gallium/state_trackers/xa/xa_composite.c |  88 ++++----
>>>>>>    src/gallium/state_trackers/xa/xa_priv.h      |   7 +-
>>>>>>    src/gallium/state_trackers/xa/xa_renderer.c  | 289
>>>>>> ++++++++++++++++-----------
>>>>>>    src/gallium/state_trackers/xa/xa_tgsi.c      |  31 ++-
>>>>>>    4 files changed, 242 insertions(+), 173 deletions(-)
>>>>>
>>>>>
>>>>>
>>>>> Rob,
>>>>> While testing this patch it looks like we sometimes set two samplers,
>>>>> and the first one is NULL.
>>>>> The SVGA driver asserts on that condition.
>>>>> We might need to move the active sampler to the first entry in that
>>>>> case, and adjust tex coords and shader accordingly.
>>>>>
>>>>> I'll discuss with BrianP.
>>>>
>>>>
>>>>
>>>> I think the root problem is a disagreement between texture samplers and
>>>> sampler views.  If a texture sampler is non-null, the corresponding
>>>> sampler
>>>> view be should be non-null too, and vice versa.
>>>>
>>>> We're tripping over an assertion when a a sampler view is non-null
>>>> but the
>>>> corresponding sampler is NULL.
>>>
>>>
>>> Right..
>>>
>>> Well, I think it would be not to hard to make things always start with
>>> SAMP[0], which would avoid this.  But I am pretty fuzzy about whether
>>> it is the driver or state tracker that is wrong here..  I started with
>>> the assumption that it was just a freedreno bug, but if it causes
>>> problem w/ vmware too, then maybe the state tracker is at fault.
>>
>>
>> I believe the general policy is that drivers can assume they're being
>> passed valid input and consistent state.  State trackers shouldn't pass
>> inconsistent state to drivers.
>>
>> In this case, the number of texture samplers and sampler views is
>> inconsistent.  The drivers could examine the current shader to determine
>> which samplers/views are actually used (which might be a subset of what
>> was specified by the state tracker).  But that's extra work that all
>> drivers would have to implement.  It seems much more economical to take
>> care of this in the state trackers.
>
>
> Though, with the DX10-level SAMPLE instructions, samplers and sampler views
> really are independent so the number could be different there. But with TEX
> instructions, the numbers should be the same.
>
> -Brian
>
>