Add set of format-aware heuristics for compressing RGBM-encoded
textures (see issue #236 for details). Due to how ASTC compresses,
there are some recommendations for the input RGBM encoding, arbitrary
data with a naive RGBM encoding will likely compress badly.
The user must pre-encode the RGBM data; the API/CLI expects to have
the data passed in RGBM format. During encoding the user must keep
the resulting M value above a minimum threshold (32 recommended), in
order to avoid quantization during ASTC compression rounding M to
zero and resulting in black/white pixels. This thresholding reduces
the precision of dark values, but the result is still at least 8-bit
precision for the LDR range, and the HDR range is not impacted by the
clamping. This is not enforced, but you may get severe block artifacts
without a clamp. A standalone sample utility for applying RGBM
encoding and decoding to an uncompressed image, including support for
thresholding, can be found in ./Utils/astc_rgbm_codec.cpp.
To compress an RGBM-encoded image using the command line use the new
-rgbm <multiplier> command line option, where the multiplier is the
scale value used during reconstruction:
vec3 hdr_val = tex.rgb * tex.a * <multiplier>
This is used to correctly assess candidate encodings.
To compress using the API create the config using the new
ASTCENC_FLG_MAP_RGBM flag. This will configure the codec for an RGBM
scale of 5 by default. If your RGBM scale is not 5, then you must
change config.rgbm_m_scale to your scale factor and
config.cw_a_weight to double your scale factor before creating the
context.
The decompress using the API or the command line no special action is
required; the additional heuristics only impact the compression path.
屑屑屑