[Mesa-dev] [PATCH] i965: Don't use texture swizzling to force alpha to 1.0 if unnecessary.

[Kenneth Graunke]

On 03/16/2013 10:09 AM, Eric Anholt wrote:
> Kenneth Graunke <kenneth at whitecape.org> writes:
>
>> On 03/15/2013 04:54 PM, Eric Anholt wrote:
>>> Kenneth Graunke <kenneth at whitecape.org> writes:
>>>
>>>> Commit 33599433c7 began setting the texture swizzle mode to XYZ1 for
>>>> RED, RG, and RGB textures in order to force alpha to 1.0 in case we
>>>> actually stored the texture as RGBA.
>>>>
>>>> This had a unforseen performance implication: the shader precompile
>>>> assumes that the texture swizzle mode will be XYZW for non-shadow
>>>> sampler types.  By setting it to XYZ1, this means every shader used with
>>>> a RED, RG, or RGB texture has to be recompiled.  This is a very common
>>>> case.
>>>>
>>>> Unfortunately, there's no way to improve the precompile, since RGBA
>>>> textures still need XYZW, and there's no way to know by looking at
>>>> the shader source what texture formats might be used.
>>>>
>>>> However, we only need to smash alpha to 1.0 if the texture's memory
>>>> format actually has alpha bits.  If not, the sampler already returns 1.0
>>>> for us without any special swizzling.  XRGB8888, for example, is a very
>>>> common case where this occurs.
>>>>
>>>> This partially fixes a performance regression since commit 33599433c7.
>>>> More work is required to fully fix it in all cases.  This at least helps
>>>> Warsow.
>>>
>>> Now that we have MESA_FORMAT_XBGR16161616_FLOAT and company, we could
>>> potentially make this conditional just die by using those formats.
>>
>> Absolutely - I believe the real fix is to use XRGB formats when
>> sampling, and ARGB formats when rendering.  I'm pretty sure we have
>> formats for all of those now.  We should do that.
>>
>> In the meantime, this patch improves things and is simple and easily
>> cherry-pickable...
>
> Fair enough.  Are there some big performance changes?

No, not large in terms of FPS.  Generally, when we get this wrong, we 
recompile each shader once and then it's right from then on.  It does 
reduce stuttering quite a bit, though, since it means we don't have to 
recompile on first draw.  I suppose it might affect very short demos.

--Ken