Unified texture encoder for BC7 and ASTC 4x4

So far, it looks possible to unify a very strong subset of BC7 and ASTC 4x4. Such an encoder would be very useful, even if it didn't support rate distortion optimization. I've been looking at this problem on and off for months, and I'm convinced that there's something really interesting here.

First, it turns out there are 30 2-subset partition patterns in common between ASTC 4x4 and BC7:

This is a collection of very strong patterns. Considering ASTC's partition pattern generator is basically a fancy rand() function, this is surprisingly good! We've got almost half of BC7's 64 patterns in there!

Secondly, the way ASTC and BC7 convert indices (weights) to 6-bit scales, and the way they both interpolate endpoints is extremely similar. So similar that I believe ASTC indices could be converted directly to BC7's with little to no expensive per-pixel CPU work required.

Finally, I wrote a little app that examines hundreds of valid ASTC configurations, trying to find which configurations resemble the strongest and most useful BC7 modes. Here they are:

So basically, all the important things between BC7 and ASTC 4x4 are basically the same or similar enough. ASTC's endpoint ranges are all over the map, but that's fine because in most cases BC7's endpoint precision is actually higher than ASTC's. A unified encoder could just optimize for lowest error across *both* formats simultaneously, and output both ASTC and BC7 endpoints with the same selectors. Or, we could only output ASTC's endpoints and just scale them to BC7's.

The next step is to write an ASTC encoder that is limited to these 12 configs and see how strong it is. After this, I need to see if this ASTC texture data can be directly and quickly converted to BC7 texture data with little loss. (Without any recompression to BC7, i.e. we just convert and copy over the endpoints/partition table index/per-pixel selector indices and that's it.) So far, I think this is all possible.

If all this actually works, it'll give us the foundation we need to build the next version of .basis. We could quantize the unified ASTC/BC7 selector/endpoint data and store it in a ".basis2" file. We then can convert this high-quality texture data to the other formats using fast block encoders (like we do already with PVRTC1 RGB/RGBA and PVRTC2 RGBA).

We could even store hints in the .basis2 file that accelerate conversion to some formats. For example we could store optimized BC1 endpoints in the endpoint codebook. Or we could store the optimal ETC1 base color/table indices, etc. Determining the per-pixel selectors for these formats is cheap once you have this info.

I think that with a strong ASTC 4x4 12-mode encoder that supports perceptual colorspace metrics, we could actually beat (or get really close) to ispc_texcomp BC7's encoder (which only supports linear RGB metrics). I think this encoder would get within a few dB of max achievable BC7.

If the system's quality isn't high enough, we could always tack on more ASTC modes, as long as they can be easily transcoded to one of the BC7 modes without expensive operations.

It's too bad that BC7 isn't well supported in WebGL yet. The extensions are there, but the browser support isn't yet. I have no idea why as the format is basically ubiquitous on desktop GPU's now, and it's the highest quality LDR texture format available. For WebGL we still need very strong BC1-5 support for desktops until this situation changes.

Idea for next texture compression experiment (originally published 9/11/16)

Right now, I've got a GPU texture in a simple ETC1 subset that is easily converted to most other GPU formats:

Base color: 15-bits, 5:5:5 RGB
Intensity table index: 3-bits
Selectors: 2-bits/texel

Most importantly, this is a "single subset" encoding, using BC7 terminology. BC7 supports between 1-3 subsets per block. A subset is just a colorspace line represented by two R,G,B endpoint colors.

This format is easily converted to DXT1 using a table lookup. It's also the "base" of the universal GPU texture format I've been thinking about, because it's the data needed for DXT1 support. The next step is to experiment with attempting to refine this base data to better take advantage of the full ETC1 specification. So let's try adding two subsets to each block, with two partitions (again using BC7 terminology), top/bottom or left/right, which are supported by both ETC1 and BC7.

For example, we can code this base color, then delta code the 2 subset colors relative to this base. We'll also add a couple more intensity indices, which can be delta coded against the base index. Another bit can indicate which ETC1 block color encoding "mode" should be used (individual 4:4:4 4:4:4 or differential 5:5:5 3:3:3) to represent the subset colors in the output block.

In DXT1 mode, we can ignore this extra delta coded data and just convert the basic (single subset) base format. In ETC1/BC7/ASTC modes, we can use the extra information to support 2 subsets and 2 partitions.

Currently, the idea is to share the same selector indices between the single subset (DXT1) and two subset (BC7/ASTC/full ETC1) encodings. This will constrain how well this idea works, but I think it's worth trying out.

To add more quality to the 2 subset mode, we can delta code (maybe with some fancy per-pixel prediction) another array of selectors in some way. We can also add support for more partitions (derived from BC7's or ASTC's), too.

Few more random thoughts on a "universal" GPU texture format (originally published 9/9/16)

In my experiments, a simple but usable subset of ETC1 can be easily converted to DXT1, BC7, and ATC. And after studying the standard, it very much looks like the full ETC1 format can be converted into BC7 with very little loss. (And when I say "converted", I mean using very little CPU, just basically some table lookup operations over the endpoint and selector entries.)

ASTC seems to be (at first glance) around as powerful as BC7, so converting the full ETC1 format to ASTC with very little loss should be possible. (Unfortunately ASTC is so dense and complex that I don't have time to determine this for sure yet.)

So I'm pretty confident now that a universal format could be compatible with ASTC, BC7, DXT1, ETC1, and ATC. The only other major format that I can't fit into this scheme easily is my old nemesis, PVRTC.

Obviously this format won't look as good compared to a dedicated, single format encoder's output. So what? There are many valuable use cases that don't require super high quality levels. This scheme purposely trades off a drop in quality for interchange and distribution.

Additionally, with a crunch-style encoding method, only the endpoint (and possibly the selector) codebook entries (of which there are usually only hundreds, possibly up to a few thousand in a single texture) would need to be converted to the target format. So the GPU format conversion step doesn't actually need to be insanely fast.

Another idea is to just unify ASTC and BC7, two very high quality formats. The drop in quality due to unification would be relatively much less significant with this combination. (But how valuable is this combo?)

(This blog post was originally mirrored here: http://geldreich1.rssing.com/chan-32192436/all_p6.html#item116)

More universal GPU texture format stuff (originally published 9/9/16)

Some BC7 format references:
https://msdn.microsoft.com/en-us/library/hh308954(v=vs.85).aspx
https://msdn.microsoft.com/en-us/library/hh308953.aspx

Source to CPU and shader BC7 (and other format) encoders/decoders:
https://github.com/Microsoft/DirectXTex

Khronos texture format references, including BC6H and BC7:
https://www.khronos.org/registry/dataformat/specs/1.1/dataformat.1.1.pdf

It may be possible to add ETC1-style subblocks into a universal GPU texture format, in a way that can be compressed efficiently and still converted on the fly to DXT1. Converting full ETC1 (with subblocks and per-subblock base colors) directly to BC7 at high quality looks easy because of BC7's partition table support. BC7 tables 0 and 13 (in 2 subset mode) perfectly match the ETC1 subblock orientations.

Any DX11 class or better GPU supports BC7, so on these GPU's the preferred output format can be BC7. DXT1 can be viewed as a legacy lower quality fallback for older GPU's.

Also, I limited the per-block (or per-subblock) base colors to 5:5:5 to simplify the experiments in my previous posts. Maybe storing 5:5:5 (for ETC1/DXT1) with 1-3 bit per-component deltas could improve the output for BC7/ASTC.

Also, one idea for alpha channel support in a universal GPU format: Store a 2nd ETC1 texture, containing the alpha channel. There's nothing to do when converting to ETC1, because using two ETC1 textures for color+alpha is a common pattern. (And, this eats two samplers, which sucks.)

When converting to DXT5's alpha block (DXT5A blocks - and yes I know there are BCx format equivalents but I'm using crnlib terms here), just use another ETC1 block color/intensity selector index to DXT5A mapping table. This table will be optimized for grayscale conversion. BC7 has very flexible alpha support so it should be a straightforward conversion.

The final thing to figure out is ASTC, but OMG that format looks daunting. Reminds me of MPEG/JPEG specs.

Few more random thoughts on a "universal" GPU texture format (originally published 9/9/16)

In my experiments, a simple but usable subset of ETC1 can be easily converted to DXT1, BC7, and ATC. And after studying the standard, it very much looks like the full ETC1 format can be converted into BC7 with very little loss. (And when I say "converted", I mean using very little CPU, just basically some table lookup operations over the endpoint and selector entries.)

ASTC seems to be (at first glance) around as powerful as BC7, so converting the full ETC1 format to ASTC with very little loss should be possible. (Unfortunately ASTC is so dense and complex that I don't have time to determine this for sure yet.)

So I'm pretty confident now that a universal format could be compatible with ASTC, BC7, DXT1, ETC1, and ATC. The only other major format that I can't fit into this scheme easily is my old nemesis, PVRTC.

Obviously this format won't look as good compared to a dedicated, single format encoder's output. So what? There are many valuable use cases that don't require super high quality levels. This scheme purposely trades off a drop in quality for interchange and distribution.

Additionally, with a crunch-style encoding method, only the endpoint (and possibly the selector) codebook entries (of which there are usually only hundreds, possibly up to a few thousand in a single texture) would need to be converted to the target format. So the GPU format conversion step doesn't actually need to be insanely fast.

Another idea is to just unify ASTC and BC7, two very high quality formats. The drop in quality due to unification would be relatively much less significant with this combination. (But how valuable is this combo?)

Universal texture compression: 5th experiment (originally published 9/15/16)

I outlined a plan for my next texture compression experiment in a previous post, here. I modified my ETC1 packer so it accepts an optional parameter which forces the encoder to use a set of predetermined selectors, instead of allowing it to use whatever selectors it likes.

The idea is, I can take an ETC1 texture using a subset of the full-format (no flips and only a single base color/intensity index - basically a single partition/single subset format using BC7 terminology) and "upgrade" it to higher quality without modifying the selector indices. I think this is one critical step to making a practical universal texture format that supports both DXT1 and ETC1.

Turns out, this idea works better than I thought it would. The ETC1 subset encoding gets 33.265 dB, while the "upgraded" version (using the same selectors as the subset encoding) gets 34.315 dB, a big gain. (Which isn't surprising, because the ETC1 subset encoding doesn't take full advantage of the format.) The nearly-optimal ETC1 encoding gets 35.475 dB, so there is still some quality left on the table here.

The ETC1 subset to DXT1 converted texture is 32.971 dB. I'm not worried about having the best DXT1 quality, because I'm going to support ASTC and BC7 too and (at the minimum) they can be directly converted from the "upgraded" ETC1 encoding that this experiment is about.

I need to think about the next step from here. I now know I can build a crunch-like format that supports DXT1, ETC1, and ATC. These experiments have opened up a bunch of interesting product and open source library ideas. Proving that BC7 support is also practical to add should be easy. ASTC is so darned complex that I'm hesitant to do it for "fun".

1. ETC1 (subset):

Max:  80, Mean: 3.809, MSE: 30.663, RMSE: 5.537, PSNR: 33.265

Its selectors:

2. ETC1 (full format, constrained selectors) - optimizer was constrained to always use the subset encoding's selectors:

Max:  85, Mean: 3.435, MSE: 24.076, RMSE: 4.907, PSNR: 34.315

Its selectors (should be the same as #1's):

Biased delta between the ETC1 subset and ETC1 full encoding with constrained selectors - so we can see what pixels have benefited from the "upgrade" pass:

3. ETC1 (full format, unconstrained selectors) - packed using an improved version of rg_etc1 in highest quality mode:

Max:  80, Mean: 3.007, MSE: 18.432, RMSE: 4.293, PSNR: 35.475

Delta between the best ETC1 encoding (#3) and the ETC1 encoding using constrained selectors (#2):

Direct conversion of ETC1 to DXT1 texture data: 4th experiment (originally published 9/11/16)

In this experiment, I've worked on reducing the size of the lookup table used to quickly convert a subset of ETC1 texture data (using only a single 5:5:5 base color, one 3-bit intensity table index, and 2-bit selectors) directly to DXT1 texture data. Now the ETC1 encoder is able to simultaneously optimize for both formats, and due to this I can reduce the size of the conversion table. To accomplish this, I've modified the ETC1 base color/intensity optimizer function so it also factors in the DXT1 block encoding error into each trial's computed ETC1 error.

The overall trial error reported back to the encoder in this experiment was etc_error*16+dxt_error. The ETC1->DXT1 lookup table is now 3.75MB, with precomputed DXT1 low/high endpoints for three used selector ranges: 0-3, 0-2, 1-3. My previous experiment had 10 precomputed ranges, which seemed impractically large. I'm unsure which set of ranges is really needed or optimal yet. Even just one (0-3) seems to work OK, but with more artifacts on very high contrast blocks.

Anyhow, here's kodim18.

ETC1 subset:

Max:  80, Mean: 3.809, MSE: 30.663, RMSE: 5.537, PSNR: 33.265

DXT1:

Max:  76, Mean: 3.952, MSE: 32.806, RMSE: 5.728, PSNR: 32.971

ETC1 block selector range usage histogram:
0-3: 19161
1-3: 3012
0-2: 2403