SkBitmapProcShader: if the bitmap already has a texture, just use it.

SkBitmapProcShader: if the bitmap already has a texture, just use it.

BUG=367163

[Stephen White, 2014-05-01 15:15:47]

Brian: PTAL. Thanks!

[Stephen White, 2014-05-01 15:17:02]

Now that I think about this some more, perhaps we need to check that the
bitmap's texture has matching params.

[bsalomon, 2014-05-01 15:22:14]

On 2014/05/01 15:17:02, Stephen White wrote:
> Now that I think about this some more, perhaps we need to check that the
> bitmap's texture has matching params.

The risk is that the bitmap is not a power-of-two, the npot extension isn't
available, and the shader wants to repeat/mirror the bitmap. In that case we'd
have to go through the code path of stretching it to the next pot.

[Stephen White, 2014-05-01 15:30:21]

On 2014/05/01 15:22:14, bsalomon wrote:
> On 2014/05/01 15:17:02, Stephen White wrote:
> > Now that I think about this some more, perhaps we need to check that the
> > bitmap's texture has matching params.
> 
> The risk is that the bitmap is not a power-of-two, the npot extension isn't
> available, and the shader wants to repeat/mirror the bitmap. In that case we'd
> have to go through the code path of stretching it to the next pot.

Right. That should be doable on-GPU, no?

The check seems to be made in get_texture_flags() in GrTexture.cpp; should we
make
this public? There doesn't seem to be a way of querying the texture for the
params
that were used to create it, but maybe we can cons up the flags using that
function
and compare the texture's flags against it?

[Stephen White, 2014-10-19 18:34:54]

On 2014/05/01 15:30:21, Stephen White wrote:
> On 2014/05/01 15:22:14, bsalomon wrote:
> > On 2014/05/01 15:17:02, Stephen White wrote:
> > > Now that I think about this some more, perhaps we need to check that the
> > > bitmap's texture has matching params.
> > 
> > The risk is that the bitmap is not a power-of-two, the npot extension isn't
> > available, and the shader wants to repeat/mirror the bitmap. In that case
we'd
> > have to go through the code path of stretching it to the next pot.
> 
> Right. That should be doable on-GPU, no?
> 
> The check seems to be made in get_texture_flags() in GrTexture.cpp; should we
> make
> this public? There doesn't seem to be a way of querying the texture for the
> params
> that were used to create it, but maybe we can cons up the flags using that
> function
> and compare the texture's flags against it?

Brian: dunno if you're still working on (or have finished) this, but this my
attempt to fix this issue (never landed).