Tuesday, April 17, 2018

BC7 encoding using weighted YCbCr colorspace metrics

I've written my second BC7 block encoder. My first was written in a straightforward way to gain experience with the format. My second was more focused on competing against the Fast ISPC Texture Compressor, but without using any SIMD, and was over 30x faster than my first attempt.

The BC7 encoders I've studied seem to be hyper focused on RGB PSNR metrics, which is just the wrong metric for many types of textures. Encoding authors that treat input textures as opaque arrays of 4x4 vectors are at a disadvantage in this domain. RGB PSNR tends to spread the error equally between the channels, which isn't what we want on sRGB textures. Instead, it's desirable to tradeoff a small amount of additional R/B error for less G error. This is what perceptual codecs like JPEG do: they transform the input into YCbCr space, then downsample and quantize the hell out of the CbCr coefficients because preserving chroma is a waste of bits.

Many other BC1 block compression codecs support weighted RGB metrics because in BC1 not doing so visually looks worse on sRGB photos/albedo textures/etc. Encoders using perceptual metrics look better on color gradients and with highly saturated blocks. Heavy usage of perceptual metrics dates back to at least NVidia's original nvdxt compressor, and it wasn't possible for crunch to compete against nvdxt without supporting perceptual metrics. The squish library recommends using perceptual metrics by default, because BC1 without perceptual metrics looks worse.

Anyhow, etc2comp by John Brooks takes things a step further and supports computing error metrics in weighted YCbCr space. Compared to vanilla RGB weighted metrics, this looks better in my experience writing Basis (especially with ETC1). I'm currently using weights (128,64,16).

Here's the REC 709 luma PSNR of 31 test textures encoded with ispc_texcomp (slow/highest quality - uses 7 modes) and my non-SIMD encoder in perceptual mode using just 4 modes:


The overall average PSNR for ispc_texcomp was 48.57, mine was 50.4. Even with ispc_texcomp's massive mode and SIMD advantages it does worse on this metric. ispc_texcomp doesn't support optimizing for perceptual metrics, which puts it at a huge disadvantage on many texture types.

I re-encoded the textures with linear metrics. My encoder used 6 modes: 0, 1, 3, 4, 5, and 6 (including all component rotations and the index flag).


ispc_texcomp's average PSNR was 46.77, mine was 46.50. My encoder can easily bridge this ~.25 dB gap (by using more modes and trying more partitions), but at a time penalty.

Note that ispc_texcomp in its best/slowest profile is pretty slow, and is much easier to compete against without SIMD code. It's just trying way too hard. It's faster in its lower quality "basic" profile, but it still doesn't support perceptual metrics so it'll continue to fight up a very steep hill.

For benchmarking, I ran each encoder in a single thread, and called ispc_texcomp with 64 blocks at a time.

Other findings: ispc_texcomp has a very weak mode 0 encoder, and it's weaker than it should be on grayscale textures. I'll blog examples soon.

Wednesday, April 4, 2018

Imaginary GPU formats

Every once in a while I wonder about alternative GPU texture format encodings. (Why not? It's fun.) There must be a sweet spot somewhere along the continuum between BC1 and BC7. Something that is more complex than BC1 but simpler than BC7. (I somewhat dislike ASTC, mostly because of its insanely complex encoding format.)

Here's one idea for an 128-bit per 4x4 block format (8 bits/texel) that mashes together ETC1+BC7. One thing I learned from ETC1 is that a lot of bits can be saved by forcing each subset's principle axis to always lie along the intensity direction. With a strong encoder, this constraint isn't as bad as one would think.

The format only has two modes: opaque and transparent. The opaque mode has 3 subsets, and the transparent mode has 2 subsets for RGB and 1 subset for alpha. Each color has 1 shared pbit, and each mode has 16 partitions for colors.

The color encoding is "RGB PBit IntensityTable". The intensity tables could be borrowed from ETC1 and expanded to 8 entries. For the transparent blocks, two 8-bit alpha values are specified (like BC4), and by borrowing degeneracy breaking from BC7 we can shave one bit from the alpha selectors. "CompRot" is a BC7-style component rotation, so any of the channels can be encoded into alpha.

Some things I like about this format: equal precision for all components, and there are only two simple modes. The opaque mode is powerful but simple: always 3 subsets, with color and selector precision better than BC1 and even better than BC7's 3 subset modes. The transparent mode is more powerful than BC3 for RGB (better color precision, and 2 subsets), but weaker for alpha (2 bit selectors vs. 3).

The main downside is that each subset's endpoints are constrained to lie along the intensity axis. I've seen commercial games ship with normal maps encoded into ETC1 and DXT1 so I know this isn't a total deal breaker.

Opaque block:
ModeBit    1 
Partition  4
Color0     777 1 3 
Color1     777 1 3 
Color2     777 1 3

Color selectors;
3 3 3 3
3 3 3 3
3 3 3 3
3 3 3 3

Total bits: 128

Transparent block:
ModeBit    1 
Partition  4
Color0     666 3 
Color1     666 3 
AlphaLoHi  8 8
CompRot    2

Color selectors:
2 2 2 2
2 2 2 2
2 2 2 2
2 2 2 2

Alpha selectors:
1 2 2 2
2 2 2 2
2 2 2 2
2 2 2 2

Total bits: 128

A strong encoder would adaptively choose between opaque blocks and transparent blocks using various component rotations, to minimize overall error. Transparent blocks can be used even on all-opaque textures.

I have no idea if this format is useful. On a rainy day I'll make a simple encoder and compare it against BC1 and BC7.

Tuesday, April 3, 2018

Basis feature support

Here's what we support right now:
  • .basis universal format, which is transcodable to BC1-5, ETC1, PVRTC1 4bpp (currently opaque only), and BC7 (currently opaque only). Alpha support for PVRTC1/BC7 is coming, and enhanced quality for BC7 and ETC2 are on the way. This is a universal solution with two quality modes (baseline and BC7), so by its very nature it trade offs max achievable quality for GPU format support.
    • For really small images (think icon-size), .basis can switch to using fixed selector codebooks to cut down on selector codebook overhead
    • Format supports arbitrary resolution texture arrays, all referring to a single set of compressed codebooks.
  • RDO BC1-5 - Creates more compressible files than crunch's (RDO in crunch was an afterthought and was pretty dumb/low quality), but slower compression. We put the most effort into optimizing BC1's output for LZ coding. Supports up to 32K entry codebooks (vs. crunch's 8K).
  • RDO ETC1 - Supports all ETC1 features, and very high quality levels (up to 32K entry codebooks), usable even on complex normal maps.
  • ETC1 intermediate format (supports all features of the ETC1 format, i.e. flips and both differential and individual colors). 10-20% smaller files at same SSIM vs. Unity crunch's the last I checked.
All of these codecs have been utilized by customers for different purposes.

We don't support an intermediate file format exclusively for BC1-5, only ETC1. Instead, we're focusing on universal solutions first, and then we'll focus on an intermediate format solution for BC7, BC6H, and ASTC.

I get asked all the time how these solutions compare to crunch's. I'll be working on extensive benchmarks soon. I've learned a lot since I designed and wrote crunch in 2009.

Sunday, April 1, 2018

Basis GPU format support update

Our goal is to support all the GPU formats (literally). Here's an update on our format support:

We just added PVRTC1 4bpp and BC7 support. PVRTC1 quality is approximately equal to PVRTexTool's middle setting ("good"), and significantly better than its lower two settings. Max quality in BC7 mode is currently limited to BC1/ETC1-grade quality levels (what we're calling "baseline" quality).

We've devised several ways of improving the max quality to near-BC7 grade by storing extra data in the .basis file. (You can't get something for nothing!) This high quality data would be optional, so users that don't care about super high quality levels can disable it and the codec will just transcode the baseline data to BC7 instead.

Here's what we support transcoding .basis into right now, in order of transcoding speed from fastest to slowest:
  • ETC1 
  • BC1 
  • BC3-5 
  • BC7: RGB 
  • PVRTC1 4bpp RGB 
Here are the formats we're going to eventually support in order of importance (with no changes to the .basis format needed):
  • PVRTC1 4bpp RGBA 
  • ETC2 RGBA 
  • BC7: RGBA 
  • PVRTC1 2bpp RGB/RGBA 
  • ASTC RGB/RGBA 
None of these formats require raw RGB/RGBA pixel processing during transcoding, i.e. we aren't just using real-time GPU format compressors here. Transcoding occurs at the level of GPU blocks, endpoints, and selector/modulation values.

At some point, we're going to boost quality above baseline, to better exploit BC7/ASTC. Most of our early users of this tech (which aren't native game apps) are happy with baseline quality, so the priority of doing this is relatively low. (Games will probably want BC7/ASTC specific codecs anyway.) We are designing the .basis format with this eventual goal, so when we add "enhanced quality" support we won't break compatibility with older baseline-only transcoders.

We'll be posting benchmarks comparing .basis to crunch (and Unity's crunch) and releasing WebAssembly (or asm.js) demos within the upcoming weeks.

Friday, March 30, 2018

Basis update - now with PVRTC support!

Basis (our new GPU texture compression product and the successor to our popular open source crunch lib) now supports PVRTC1, along with ETC1 and BC1-5 (DXTc). This means a .basis file can be utilized on pretty much every GPU in the universe that matters, independent of platform or API. A .basis file is conceptually like JPEG but for GPU texture data, and can be used on the web (using Emscripten and WebGL) or by native apps (using a small C++ transcoder library).

All textures are 1024x1024 (due to PVRTC1 limitations). Click on each one to see them at full-res (they are reduced in size on the page itself).

Each image below was transcoded directly to each GPU format from the .basis file, and then converted to 24bpp .PNG. On my desktop, ETC1 is fastest (~3ms), followed by BC1/4 (~7.9ms), then PVRTC (~37ms). The transcoders (particularly PVRTC) are not yet fully optimized, and are written in straightforward C++. (Update: PVRTC transcode at 1024x1024 is now ~15.4ms, without any SIMD or threading yet.)

The PVRTC transcoder really needs SIMD optimizations, which should give it a nice speed boost (probably around 2-3x). It would be trivial to thread the PVRTC transcoder too. The PVRTC's transcoder's quality is visually somewhere in between PVRTexTool's "Lower Quality" and "Good" settings. In many cases, it looks a little better than "Good", but it's a tossup.

Note that BC3 and BC5 formats are supported by calling the transcoder twice from different input image slices. So a RGBA GPU texture is encoded into two slices (sharing the same codebooks) in a single .basis file, and it transcodes to either two ETC1 textures, a ETC1 texture twice as high, or a single BC5 texture. PVRTC2 and ETC2 support will be very easy and transcode times will be comparable to ETC1 or BC1 (PVRTC1 will always be the most expensive). The PVRTC transcoder doesn't support alpha yet (it's next).

Image: laststarfighter_1024.basis, 133966 bytes, 1.022 bits/pixel

Original:


.basis->ETC1:


.basis->BC1:


.basis->BC4:


.basis->PVRTC:

Image: map_1024.basis, 180603 bytes, 1.38 bits/pixel

Original:


.basis->ETC1:


.basis->BC1:


.basis->BC4:


.basis->PVRTC:


Image: delorean_1024.png, 138894 bytes, 1.06 bits/pixel

Original:


.basis->ETC1:


.basis->BC1:


.basis->BC4:


.basis->PVRTC:



Monday, March 26, 2018

Basis v1.11 with universal GPU texture support has shipped

We've sent drops to two companies so far. This is the first version that supports fast block-level transcoding of .basis files to multiple formats: ETC1 (mobile) or BC1-5 (desktop). This is a major milestone for us, because Basis is the first system available to support efficient platform independent distribution of highly compressed GPU texture data. We've been working up to this release for over a year.

Here's a tiny demo (our first), using the Basis transcoder compiled to Javascript using Emscripten: http://lzham.info/decode_test.html

For some encoded example images created during development, see thisthis, or this post.

You encode your textures/images a single time, store a single set of .basis files (which are approximately the size of JPEG files), download the file on the remote device, and then transcode to the format you need for that device. Our transcoder converts the block-level data to DXT or ETC format GPU texture bits on the fly. The encoder is aware of all the formats and balances the quality levels of each.

.basis files consist of one or more 2D texture "slices", where each slice can be any dimension. Slices can be mipmap levels, tiles, cubemap faces, video frames, etc. - whatever you want.

We think the primary use case for .basis files are web apps of various types, or any kind of app that needs to distribute GPU texture data across a wide range of GPU devices. We've tested this solution on normal maps, diffuse maps, gloss maps, satellite photos, photographs, grayscale images, flight navigation maps, etc.

Anyhow, here's what you get with Basis:

  • bin, bin_linux, bin_osx: DLL/so/dylib's containing the precompiled encoder library (which is closed source) and several command line tools. The main tools are basiscomp (our new .basis file encoder, and our RDO ETC1 compressor) and rdodxt (uses our new RDO BC1-5 encoders that are around 10-25% better than crunch's).
  • basisexample: Shows how to use the encoder DLL to encode .basis or RDO .KTX files.
  • inc: Transcoder library source code/headers.
  • lib: static import library for encoder DLL
  • transcoding_demo: Sample that uses the included transcoder library (provided in source code form in the 
  • rdodxt: Sample that uses the encoder DLL to do RDO BC1-5 compression
I've tried to keep the few API's in the product as simple as possible, so not much documentation is needed for them. The readme file covers them. Encoding involves filling out a struct and calling a single C function in the DLL. Transcoding slices is similar, except you use a couple simple methods in inc/basis_decoder.h.


Wednesday, February 28, 2018

On Age DE's pathing/movement

Typical Age DE forum post:
The pathfinding in the game is terrible.
First off, Age of Empires Definitive Edition is a remaster of Age of Empires. It's not a rewrite, it's not a new engine, and that's what we've been saying for almost a year. Age of Empires 1's path finding was really bad, as most reviews point out:
This is the code we started with in DE. Not Age 2, not new code, but the original code which had a ton of flaws and quirks we had to learn about the hard way. This code was almost a quarter of a century old, and it showed. The original movement/pathing code was very weak to say the least. Yet entire complex systems above it (combat, AI, etc.) depended on this super quirky movement/pathing code. It took multiple Ensemble engineers several years of development to go from Age 1 to Age 2 level pathing.
We made a number of improvements to the path finding and unit movement code without breaking the original system. It must be emphasized that Age1's pather and movement code is extremely tricky and hard to change without breaking a hundred things about the game or AI (sometimes in subtle ways). It was a very tricky balance. The current system still has problems with chokepoints, which can be fixed with more work, but we instead had to focus on multiplayer which had to basically be 85% rewritten.
Here's a list of fixes made so far to DE's pathing and movement code in the time I had, which was only like 2 months:
  • DE's pathing system's findPath() function was speeded up by approx 3-4x faster vs. Age1's
    I performed around a dozen separate optimizations passes on the core pather. I implemented the A* early exploration optimization (eliminating 1 open list insertion/removal per iteration), and massively tuned the C++ code to generate reasonably efficient x64 assembly. I transformed the inner loops so much that they barely resemble the original code. We retested the pather and game thoroughly after each major optimization pass. 
  • Age1's pather's A* implementation was outright broken (the open list management was flawed, so the cheapest node wasn't always expanded upon during each iteration). DE's pather fixes all these bugs and is a proper implementation of A*. (I have no idea how the original code shipped, but it was 1997!)
  • DE's pather gives up if after many thousands of iterations it can't make forward progress towards the goal, to avoid spending CPU cycles on hopeless pathing unnecessarily. (It's more complex than this, but that's the gist of it.)
  • Added multiple lane support to villager pathing.
  • Villagers can use one of two collision sizes (either small or large), so if a villager bumps into another friendly villager we can immediately switch to the smaller radius to avoid stopping. So basically, villagers can get very close to each other, avoiding gathering slowdowns. Villager vs. military combat was preserved because vills use large radii by default, and if a vill vs. vill collision doesn't occur after a set period of time the villager returns back to the large radius. (An unfortunate side effect of this: It's possible to group together a ton of villagers - way more than the original game - and use them to attack other units. Villager Mobs are a game unbalancing issue in DE.) Note that another engineer (not associated with FE) attempted to "fix" villagers by allowing them to collide/overlap and totally broke the entire game, and that code did not and could not ship because it broke the engine's assumptions in multiple ways.
  • Movement of units through single-tile openings was greatly improved and tested with all unit types. Age 1's handling of single tile openings was so bad that players would exploit it:
    http://artho.com/age/placeb.html
  • The DE pather was modified to have a much higher max iteration count than Age1's, so longer and more complex routes can be found.
  • The per-turn pathing cap in Age1 was switched to short and long range pathing categories in DE. 8 short range paths can occur per turn, and for long range paths it supports up to 4 findPaths() per turn.
  • For short range paths, straight line paths are preferred vs. the tile path returned by findPath() if the straight line path is safe to traverse.
  • Boat movement was modified to have deceleration.
  • Waypoints along a path can be skipped if a unit can safely move from its current position to the next waypoint
  • Added support for 32-facing angles vs. Age's original 8. Also, the unit direction/facing angle is interpolated in DE, instead of "snapped" to like in Age1. The interpolation is purposely disabled when units switch angles during combat.
  • Added stuck unit detection logic to DE's movement code, to automatically detect and fix permanently stuck units (rare, but possible).
  • We ported Age2's entire obstruction manager into DE, replacing the old bitmap system. Units use circular obstructions, and buildings use square obstructions.
  • Added several new behaviors to the movement code to help with chokepoints: A "wait" behavior, that checks every second or so for up to 45 seconds to see if the unit can be moved to the destination, and a stuck unit "watchdog", which watches to see if a unit hasn't made forward progress and tries to switch behaviors to get the unit unstuck. Age1's code would just give up at the slightest problem.
  • The pather tries to path starting from the center of each tile, but this sometimes fails in tight spaces or with lots of units around. DE tries harder to find a good starting position, so movement through single tile openings isn't broken.
  • Path caching system: Villagers and boats can reuse previously found paths in DE, for efficiency.
  • In situations that Age1's pather would just outright give up and stop, DE's pather tries a lot harder to get the unit where it needs to go using several randomized fallback behaviors.
  • Age1's waypoint detection code was very janky and fixed in DE. Age1's code would sometimes cause units to oscillate around their current waypoint until it figured out it was ok to go to the next one.
  • The original Age1 devs used text log files to debug the pather/movement systems. I had to write a 2D/3D debug primitive system so I could see what was actually going on. Here are some development screenshots - notice how complex this stuff is:



Age1's pathing/movement systems implements a form of randomized, emergent behavior. The units are basically like dumb ants. It's imperfect in chokepoints, but it's a continuation of the essence of what made Age 1 what it was. If all the units are moving in the same direction it can usually handle chokepoints (I tested this over and over with a wide variety of units on a pathing torture test scenario from MS before release). The fundamental behaviors the AI and combat systems expected were accurately preserved in DE's pather, which was our goal.
Instead of people saying "the pathing in DE sucks!", I would much rather hear about the specific issues with movement/pathing, and actually constructive suggestions on how to improve the system without breaking the game or turning it into Age 2.