[Mesa-dev] [PATCH V2 00/22] Add ARB_texture_multisample support

Paul Berry stereotype441 at gmail.com
Wed Feb 6 07:04:47 PST 2013


On 5 February 2013 11:34, Paul Berry <stereotype441 at gmail.com> wrote:

> On 4 February 2013 21:48, Chris Forbes <chrisf at ijw.co.nz> wrote:
>
>> This series adds the core mesa bits and i965 driver support for
>> ARB_texture_multisample.
>>
>> Big changes from V1:
>>    - Generally cleaner
>>    - GLSL texelFetch() with a multisample sampler is converted to
>>       a new ir_txf_ms opcode in the GLSL IR, which makes the driver side
>>       a lot tidier.
>>    - Most of the review comments have been addressed.
>>    - Intel driver support includes both Gen6 and Gen7 now.
>>    - Lots of bugs fixed.
>>
>> Remaining issues:
>>    - Stencil support is still a mess. This is disabled in the intel driver
>>      for now by exposing only GL_MAX_DEPTH_TEXTURE_SAMPLES=1. To do this
>>      properly requires some more small changes in core mesa to cope with
>>      stencil-only texture formats a bit better, and Intel-specific support
>>      to cope with the strange W-tiled stencil buffer format.
>>
>>    - The IvyBridge support forces uncompressed multisample layout, which
>>      is wasteful of memory bandwidth. Allowing CMS here requires emitting
>>      specialized code based on the multisample layout, since fetching from
>>      an undefined MCS surface isn't safe.
>>
>>    - It's unclear whether deallocating multisample textures via zero
>>      dimensions is legal, or whether the texture object must be deleted.
>>
>> I'm going to continue working on these issues, but would appreciate
>> another
>> round of review comments in the mean time.
>>
>> -- Chris
>>
>
> I'm about halfway through reviewing the series, and out of time for patch
> review today.  I'll pick it up again tomorrow.  So far it looks really
> good!  I think we are nearing the home stretch on this series.  Thanks
> again for your continued efforts, Chris.  I'm going to be really glad to
> see this feature land.
>

Ok, finished my review.  I sent comments on patches 10, 15, 18, 20, and
21.  Also, I second Eric's comment that patches 1-3 need to be squashed--we
try to keep Mesa buildable (and passing "make check") after every commit,
for ease in future bisecting.

The remaining patches are:

Reviewed-by: Paul Berry <stereotype441 at gmail.com>

Note: considering the complexity of this series, I tihnk it would be good
to try to land the Piglit tests at the same time as the feature.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130206/1f97104b/attachment.html>


More information about the mesa-dev mailing list