xref: /third_party/astc-encoder/Docs/Encoding.md (revision cc1dc7a3)
1# Effective ASTC Encoding
2
3Most texture compression schemes encode a single color format at single
4bitrate, so there are relatively few configuration options available to content
5creators beyond selecting which compressed format to use.
6
7ASTC on the other hand is an extremely flexible container format which can
8compress multiple color formats at multiple bit rates. Inevitably this
9flexibility gives rise to questions about how to best use ASTC to encode a
10specific color format, or what the equivalent settings are to get a close
11match to another compression format.
12
13This page aims to give some guidelines, but note that they are only guidelines
14and are not exhaustive so please deviate from them as needed.
15
16## Traditional format reference
17
18The most commonly used non-ASTC compressed formats, their color format, and
19their compressed bitrate are shown in the table below.
20
21| Name     | Color Format | Bits/Pixel | Notes            |
22| -------- | ------------ | ---------- | ---------------- |
23| BC1      | RGB+A        | 4          | RGB565 + 1-bit A |
24| BC3      | RGB+A        | 8          | BC1 RGB + BC4 A  |
25| BC3nm    | G+R          | 8          | BC1 G   + BC4 R  |
26| BC4      | R            | 4          | L8               |
27| BC5      | R+G          | 8          | BC1 R + BC1 G    |
28| BC6H     | RGB (HDR)    | 8          |                  |
29| BC7      | RGB / RGBA   | 8          |                  |
30| EAC_R11  | R            | 4          | R11              |
31| EAC_RG11 | RG           | 8          | RG11             |
32| ETC1     | RGB          | 4          | RGB565           |
33| ETC2     | RGB+A        | 4          | RGB565 + 1-bit A |
34| ETC2+EAC | RGB+A        | 8          | RGB565 + EAC A   |
35| PVRTC    | RGBA         | 2 or 4     |                  |
36
37**Note:** BC2 (RGB+A) is not included in the table because it's rarely used in
38practice due to poor quality alpha encoding; BC3 is nearly always used instead.
39
40**Note:** Color representations shown with a `+` symbol indicate non-correlated
41compression groups; e.g. an `RGB + A` format compresses `RGB` and `A`
42independently and does not assume the two signals are correlated. This can be
43a strength (it improves quality when compressing non-correlated signals), but
44also a weakness (it reduces quality when compressing correlated signals).
45
46# ASTC Format Mapping
47
48The main question which arises with the mapping of another format on to ASTC
49is how to handle cases where the input isn't a 4 component RGBA input. ASTC is
50a container format which always decompresses in to a 4 component RGBA result.
51However, the internal compressed representation is very flexible and can store
521-4 components as needed on a per-block basis.
53
54To get the best quality for a given bitrate, or the lowest bitrate for a given
55quality, it is important that as few components as possible are stored in the
56internal representation to avoid wasting coding space.
57
58Specific optimizations in the ASTC coding scheme exist for:
59
60* Encoding the RGB components as a single luminance component, so only a single
61  value needs to be stored in the coding instead of three.
62* Encoding the A component as a constant 1.0 value, so the coding doesn't
63  actually need to store a per-pixel alpha value at all.
64
65... so mapping your inputs given to the compressor to hit these paths is
66really important if you want to get the best output quality for your chosen
67bitrate.
68
69## Encoding 1-4 component data
70
71The table below shows the recommended component usage for data with different
72numbers of color components present in the data.
73
74The coding swizzle should be applied when compressing an image. This can be
75handled by the compressor when reading an uncompressed input image by
76specifying the swizzle using the `-esw` command line option.
77
78The sampling swizzle is what your should use in your shader programs to read
79the data from the compressed texture, assuming no additional API-level
80component swizzling is specified by the application.
81
82| Input components |  ASTC Endpoint | Coding Swizzle | Sampling Swizzle   |
83| -------------- |  ------------- | -------------- | ------------------ |
84| 1              |  L + 1         | `rrr1`         | `.g` <sup>1</sup>  |
85| 2              |  L + A         | `rrrg`         | `.ga` <sup>1</sup> |
86| 3              |  RGB + 1       | `rgb1`         | `.rgb`             |
87| 4              |  RGB + A       | `rgba`         | `.rgba`            |
88
89**1:** Sampling from `g` is preferred to sampling from `r` because it allows a
90single shader to be compatible with ASTC, BC1, or ETC formats. BC1 and ETC1
91store color endpoints as RGB565 data, so the `g` component will have higher
92precision. For ASTC it doesn't actually make any difference; the same single
93component luminance will be returned for all three of the `.rgb` components.
94
95## Equivalence with other formats
96
97Based on these component encoding requirements we can now derive the the ASTC
98coding equivalents for most of the other texture compression formats in common
99use today.
100
101| Formant  | ASTC Coding Swizzle | ASTC Sampling Swizzle | Notes            |
102| -------- | ------------------- | --------------------- | ---------------- |
103| BC1      | `rgba` <sup>1</sup> | `.rgba`               |                  |
104| BC3      | `rgba`              | `.rgba`               |                  |
105| BC3nm    | `gggr`              | `.ag`                 |                  |
106| BC4      | `rrr1`              | `.r`                  |                  |
107| BC5      | `rrrg`              | `.ra` <sup>2</sup>    |                  |
108| BC6H     | `rgb1`              | `.rgb` <sup>3</sup>   | HDR profile only |
109| BC7      | `rgba`              | `.rgba`               |                  |
110| EAC_R11  | `rrr1`              | `.r`                  |                  |
111| EAC_RG11 | `rrrg`              | `.ra` <sup>2</sup>    |                  |
112| ETC1     | `rgb1`              | `.rgb`                |                  |
113| ETC2     | `rgba` <sup>1</sup> | `.rgba`               |                  |
114| ETC2+EAC | `rgba`              | `.rgba`               |                  |
115| ETC2+EAC | `rgba`              | `.rgba`               |                  |
116
117**1:** ASTC has no equivalent of the 1-bit punch-through alpha encoding
118supported by BC1 or ETC2; if alpha is present it will be a full alpha
119component.
120
121**2:** ASTC relies on using the L+A color endpoint type for coding efficiency
122for two component data. It therefore has no direct equivalent of a two-plane
123format sampled though the `.rg` components such as BC5 or EAC_RG11. This can
124be emulated by setting texture component swizzles in the runtime API - e.g. via
125`glTexParameteri()` for OpenGL ES - although it has been noted that API
126controlled swizzles are not available in WebGL.
127
128**3:** ASTC can only store unsigned values, and has no equivalent of the BC6
129signed endpoint mode.
130
131# Other Considerations
132
133This section outlines some of the other things to consider when encoding
134textures using ASTC.
135
136## Decode mode extensions
137
138ASTC is specified to decompress into a 16-bit per component RGBA output by
139default, with the exception of the sRGB format which uses an 8-bit value for the
140RGB components.
141
142Decompressing in to a 16-bit per component output format is often higher than
143many use cases require, especially for LDR textures which originally came from
144an 8-bit per component source image. Most implementations of ASTC support the
145decode mode extensions, which allow an application to opt-in to a lower
146precision decompressed format (RGBA8 for LDR, RGB9E5 for HDR). Using these
147extensions can improve GPU texture cache efficiency, and even improve texturing
148filtering throughput, for use cases that do not need the higher precision.
149
150The ASTC format uses different data rounding rules when the decode mode
151extensions are used. To ensure that the compressor chooses the best encodings
152for the RGBA8 rounding rules, you can specify `-decode_unorm8` when compressing
153textures that will be decompressed into the RGBA8 intermediate. This gives a
154small image quality boost.
155
156**Note:** This mode is automatically enabled if you use the `astcenc`
157decompressor to write an 8-bit per component output image.
158
159## Encoding non-correlated components
160
161Most other texture compression formats have a static component assignment in
162terms of the expected data correlation. For example, ETC2+EAC assumes that RGB
163are always correlated and that alpha is non-correlated. ASTC can automatically
164encode data as either fully correlated across all 4 components, or with any one
165component assigned to a separate non-correlated partition to the other three.
166
167The non-correlated component can be changed on a block-by-block basis, so the
168compressor can dynamically adjust the coding based on the data present in the
169image. This means that there is no need for non-correlated data to be stored
170in a specific component in the input image.
171
172It is however worth noting that the alpha component is treated differently to
173the RGB color components in some circumstances:
174
175* When coding for sRGB the alpha component will always be stored in linear
176  space.
177* When coding for HDR the alpha component can optionally be kept as LDR data.
178
179## Encoding normal maps
180
181The best way to store normal maps using ASTC is similar to the scheme used by
182BC5; store the X and Y components of a unit-length normal. The Z component of
183the normal can be reconstructed in shader code based on the knowledge that the
184vector is unit length.
185
186To encode this we need to store only two input components in the compressed
187data, and therefore use the `rrrg` coding swizzle to align the data with the
188ASTC luminance+alpha endpoint. We can sample this in shader code using the
189`.ga` sampling swizzle, and reconstruct the Z value with:
190
191    vec3 nml;
192    nml.xy = texture(...).ga;                // Load normals (range 0 to 1)
193    nml.xy = nml.xy * 2.0 - 1.0;             // Unpack normals (range -1 to +1)
194    nml.z = sqrt(1 - dot(nml.xy, nml.xy));   // Compute Z, given unit length
195
196The encoding swizzle and appropriate component weighting is enabled by using
197the `-normal` command line option. If you wish to use a different pair of
198components you can specify a custom swizzle after setting the `-normal`
199parameter. For example, to match BC5n component ordering use
200`-normal -esw gggr` for compression and `-normal -dsw arz1` for decompression.
201
202## Encoding sRGB data
203
204The ASTC LDR profile can compress sRGB encoded color, which is a more
205efficient use of bits than storing linear encoded color because the gamma
206corrected value distribution more closely matches human perception of
207luminance.
208
209For color data it is nearly always a perceptual quality win to use sRGB input
210source textures that are then compressed using the ASTC sRGB compression mode
211(compress using the `-cs` command line option rather than the `-cl` command
212line option). Note that sRGB gamma correction is only applied to the RGB
213components during decode; the alpha component is always treated as linear
214encoded data.
215
216*Important:* The uncompressed input texture provided on the command line must
217be stored in the sRGB color space for `-cs` to function correctly.
218
219## Encoding HDR data
220
221HDR data can be encoded just like LDR data, but with some caveats around
222handling the alpha component.
223
224For many use cases the alpha component is an actual alpha opacity component and
225is therefore used for storing an LDR value between 0 and 1. For these cases use
226the `-ch` compressor option which will treat the RGB components as HDR, but the
227A component as LDR.
228
229For other use cases the alpha component is simply a fourth data component which
230is also storing an HDR value. For these cases use the `-cH` compressor option
231which will treat all components as HDR data.
232
233- - -
234
235_Copyright © 2019-2024, Arm Limited and contributors. All rights reserved._
236