In HDR mode the current compressor will always use HDR RGBA or HDR RGB + LDR A endpoints. This is functional, but tends to result in poor image quality in LDR-only regions, especially at lower bitrates.
We should investigate dynamically switching between LDR and HDR endpoints, based on the per-block or even per-partition data.
The most obvious way to do this would be toggle HDR endpoints only if the max component value in a block/partition is above the maximum LDR value. This keeps the representable dynamic range, but loses any increased precision for dark regions at low exposure levels (not sure how much increased precision there is in practice - need to investigate).
In HDR mode the current compressor will always use HDR RGBA or HDR RGB + LDR A endpoints. This is functional, but tends to result in poor image quality in LDR-only regions, especially at lower bitrates.
We should investigate dynamically switching between LDR and HDR endpoints, based on the per-block or even per-partition data.
The most obvious way to do this would be toggle HDR endpoints only if the max component value in a block/partition is above the maximum LDR value. This keeps the representable dynamic range, but loses any increased precision for dark regions at low exposure levels (not sure how much increased precision there is in practice - need to investigate).