Encode astc support

[wasimabbas-arm]

Updated version of #810.

Since #810 was adding a separate tool. Its easier to start another cleaner PR.

This PR has:
Commit to add --format option and make it exclusive with --codec
Add ASTC support to encode

[MarkCallow]

Thanks @wasimabbas-arm. Can you now get metrics for ASTC compressed files?

You will need to add tests of the new functionality to the tools CTS and possibly need to update some existing tests.

[MarkCallow]

Some nits.

[MarkCallow]

I've merged the CTS changes. They only update some messages. Please add tests that use encode to encode KTX2 files to ASTC.

[abbaswasim]

@MarkCallow thanks. I will make the rest of the changes.

Can I just ask. I know you have mentioned

Thanks @wasimabbas-arm. Can you now get metrics for ASTC compressed files?

You will need to add tests of the new functionality to the tools CTS and possibly need to update some existing tests.

But every time I have come back to this PR I wasn't sure how many things needs doing (apart from fixing those tests of course). And thats put me off a bit, and ended up procrastinating.

Is there a complete list of things that I need to do for this? I am hoping to go through all at once to reduce lots of back and forth.

At the moment I have

1. Write tests for new functionality (How many tests are enough? It's already tested in a way, as PR#31 in CTS). Please provide details of what else needs testing.
2. Get metrics for ASTC compress files (Could you describe what this exactly means?)
3. what else needs doing? (Maybe the doxygen docs are wrong?)

[MarkCallow]

Write tests for new functionality (How many tests are enough? It's already tested in a way, as PR#31 in CTS).

That only modifies 2 golden files to account for the changes in the text of some messages. We need to test the functionality of encoding to ASTC with ktx encode. Look at the tests for encoding UASTC and the tests for ASTC in ktx create to see the kinds of things to test. Test encoding from all valid input input types (R8, etc) encoding to various block dimensions. Probably, but please check, the tests of ASTC encoding in ktx create test the encoder sufficiently (various output block dimensions, etc) so we only need to test a few output types. Test invalid input types, e.g. R32 and make sure the errors are handled correctly. Also test invalid command line options, e.g --codec and --format, and make sure they are handled as expected. You should be able to find suitable input files, and maybe some golden files, in the existing collection under clitests/input.

I'll address the rest of the questions in your comment tomorrow. It's late here now.

[wasimabbas-arm]

Not sure if the last commit is correct or not. Maybe I need to wait for the KhronosGroup/KTX-Software-CTS#33 to be merged before I can update the submodule

[MarkCallow]

Not sure if the last commit is correct or not. Maybe I need to wait for the KhronosGroup/KTX-Software-CTS#33 to be merged before I can update the submodule

If you updated the submodule reference to the commit that is in the PR, it should be fine.

[MarkCallow]

I'll address the rest of the questions in your comment tomorrow.

Sorry. It's two days later. Back to your questions.

Get metrics for ASTC compress files (Could you describe what this exactly means?)

I was asking if you had implemented the feature of printing metrics on the loss from the encoding. Looking at the code, you seem to have done so. The important thing is decoding the ASTC-compressed images so they can be compared with the reference images. You've done that. The rest of the code in metrics_utils.h is the same regardless whether the images came from BasisU or ASTC encoding. We need tests though. Please add tests that encode to ASTC and print out the metrics. (It is possible you have done so and I missed them when reviewing the CTS PR.)

what else needs doing? (Maybe the doxygen docs are wrong?)

At the moment I do not think there is anything else beyond addressing the issues here and in the CTS PR. I'll review the docs again but what I looked at so far seemed fine.

[MarkCallow]

Please add tests that encode to ASTC and print out the metrics.

Please modify command_create.cpp as necessary so the metrics options work for ASTC as well as BasisU and add tests for the functionality. It previously only supported metrics for BasisU.

[MarkCallow]

I just found one more thing that needs to be done. The documentation for --normal-mode in encode_utils_common.h says "use --normalize if unsure that the input has unit normals". However there is no normalize option (that I could find. Something weird is going on with my Xcode find at present). The image class has implementations so it should just be a question of adding the option and when set calling the function on the image.

[MarkCallow]

List to track the requests I've made that aren't comments on existing code.

• Uncomment references in ktx create documentation to encode being able to encode to ASTC, now that it is.
• Enable printing metrics when encoding to ASTC in ktx create
• Add --normalize option to ktx create
• Add test to ensure passing an ASTC encoded KTX2 file to ktx encode is a no-op.

[wasimabbas-arm]

* [ ]  Add `--normalize` option to `ktx create`

@MarkCallow Please refer to #812. Its currently is still the case that both create and encode has lost the option.

I would argue this is out of scope for this PR. Lets conclude 812 before we proceed.

[MarkCallow]

The reuse issue has cleared up on its own. I've no idea what happened. The latest version ran fine for me locally.

[MarkCallow]

* [ ]  Add `--normalize` option to `ktx create`

@MarkCallow Please refer to #812. Its currently is still the case that both create and encode has lost the option.

I would argue this is out of scope for this PR. Lets conclude 812 before we proceed.

I agree it is out of scope for this PR. I've pinged Alexey again in #812. My view is normalize should be restored.

[MarkCallow]

Please update the CTS reference then I can merge this.

[wasimabbas-arm]

Please update the CTS reference then I can merge this.

Done.

[MarkCallow]

Thanks @wasimabbas-arm.