Tile Compression

Tile Compression

This is almost the complete source code of the tile texture compression
feature developed at Opera. It's quite large as everything has been
squashed to a single commit. The only thing omitted is the heuristics
for automatically turning compression on or off for individual tiles. It
will need to be handled separately due to the added complexity.

This review has been created for evaluation purposes. Individual
components will be uploaded and reviewed separately.

For reference the design document can be found here:
https://docs.google.com/document/d/1WdA0ir5J5gzhJ1yizDOigVG1lIQRolaLtI4CWEWcj5g/edit?usp=sharing

The bug in question is: 434699


BUG=

[reveman, 2014-12-12 16:52:24]

reveman@chromium.org changed reviewers:
+ reveman@chromium.org

[reveman, 2014-12-12 16:52:25]

GPU memory buffer related changes look good. I'm not sure it makes sense to
include async upload support if it's only supported by the idle uploader.

[christiank, 2014-12-15 07:09:57]

On 2014/12/12 16:52:25, reveman wrote:
> GPU memory buffer related changes look good. I'm not sure it makes sense to
> include async upload support if it's only supported by the idle uploader.

True, I included it mostly to show what we currently have. I'll start with the
GPU memory buffer parts and we can take it from there.

Thanks for having a look!

[jchen10, 2015-04-08 07:29:34]

On 2014/12/15 07:09:57, christiank wrote:
> On 2014/12/12 16:52:25, reveman wrote:
> > GPU memory buffer related changes look good. I'm not sure it makes sense to
> > include async upload support if it's only supported by the idle uploader.
> 
> True, I included it mostly to show what we currently have. I'll start with the
> GPU memory buffer parts and we can take it from there.
> 
> Thanks for having a look!

Hi Christian,
    You have optimized the encoders using NEON for ARM in this CL. So, do you
have plan to do the same for IA using SIMD intrinsics? If possible, will you
welcome such contribution?

[christiank, 2015-04-08 07:40:10]

On 2015/04/08 07:29:34, jchen10 wrote:
> On 2014/12/15 07:09:57, christiank wrote:
> > On 2014/12/12 16:52:25, reveman wrote:
> > > GPU memory buffer related changes look good. I'm not sure it makes sense
to
> > > include async upload support if it's only supported by the idle uploader.
> > 
> > True, I included it mostly to show what we currently have. I'll start with
the
> > GPU memory buffer parts and we can take it from there.
> > 
> > Thanks for having a look!
> 
> Hi Christian,
>     You have optimized the encoders using NEON for ARM in this CL. So, do you
> have plan to do the same for IA using SIMD intrinsics? If possible, will you
> welcome such contribution?

Hi,

We're currently focused on ARM and have no immediate plans on implementing Intel
SIMD optimized encoders. Any contributions in this area are welcome.

Please note that this is a somewhat outdated CL. We have gradually been
upstreaming parts of this feature and it have diverged somewhat from this CL.

Currently we only have only landed the non-SIMD optimized ETC1 encoder. The NEON
optimized version will follow soon.

[jchen10, 2015-04-09 01:06:36]

On 2015/04/08 07:40:10, christiank wrote:
> On 2015/04/08 07:29:34, jchen10 wrote:
> > On 2014/12/15 07:09:57, christiank wrote:
> > > On 2014/12/12 16:52:25, reveman wrote:
> > > > GPU memory buffer related changes look good. I'm not sure it makes sense
> to
> > > > include async upload support if it's only supported by the idle
uploader.
> > > 
> > > True, I included it mostly to show what we currently have. I'll start with
> the
> > > GPU memory buffer parts and we can take it from there.
> > > 
> > > Thanks for having a look!
> > 
> > Hi Christian,
> >     You have optimized the encoders using NEON for ARM in this CL. So, do
you
> > have plan to do the same for IA using SIMD intrinsics? If possible, will you
> > welcome such contribution?
> 
> Hi,
> 
> We're currently focused on ARM and have no immediate plans on implementing
Intel
> SIMD optimized encoders. Any contributions in this area are welcome.
> 
> Please note that this is a somewhat outdated CL. We have gradually been
> upstreaming parts of this feature and it have diverged somewhat from this CL.
> 
> Currently we only have only landed the non-SIMD optimized ETC1 encoder. The
NEON
> optimized version will follow soon.

Great! Thanks for your information. I will start to work on IA following your
pace of upstreaming.