Struct EntropyMulTable 

#[non_exhaustive]
pub struct EntropyMulTable {
    pub dct8: f32,
    pub dct4x4: f32,
    pub dct4x8: f32,
    pub identity: f32,
    pub dct2x2: f32,
    pub afv: f32,
    pub dct16x8: f32,
    pub dct16x16: f32,
    pub dct16x32: f32,
    pub dct32x32: f32,
    pub dct64x32: f32,
    pub dct64x64: f32,
}

Per-strategy raw entropy multipliers for the AC strategy cost model.

These control the relative preference for each transform type in AC strategy selection. Higher values penalize a strategy (making it less likely to be chosen); lower values favor it. The 8x8-class values are normalized by DCT8’s value before use, so DCT8 always evaluates at 1.0. Larger transforms use raw values directly.

Default values match libjxl enc_ac_strategy.cc:584 (kTransforms8x8[i].entropy_mul). Experimental values from libjxl PR #4506 (Jon Sneyers, VarDCT cost tuning).

#[non_exhaustive] so future libjxl-side strategy additions can land without a breaking change. Construct via Self::reference or Self::experimental and mutate fields as needed.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

dct8: f32

DCT8 base value. All 8x8-class transforms are normalized by this. Reference: 0.8 (libjxl enc_ac_strategy.cc:357, kTransforms8x8[0].entropy_mul).

dct4x4: f32

DCT4x4 (four 4x4 sub-blocks per 8x8 block). Reference: 1.08. Experimental: 0.88 (PR #4506, ~19% reduction). Lowering favors DCT4x4 for textured/detailed regions (screenshots, text).

dct4x8: f32

DCT4x8 / DCT8x4 (half-block transforms for edges/detail). Reference: 0.859316 (libjxl enc_ac_strategy.cc).

identity: f32

Identity (pixel copy, no transform). Reference: 1.0428. Experimental: 0.88 (PR #4506, ~16% reduction). Lowering favors identity blocks for flat/noisy regions.

dct2x2: f32

DCT2x2 (2x2 Hadamard-like transform). Reference: 0.95 (libjxl enc_ac_strategy.cc).

afv: f32

AFV (Adaptive Frequency Variable, corner DCT). Reference: 0.818. Experimental: 0.75 (PR #4506, ~8% reduction). Lowering favors AFV for edge blocks with mixed content.

dct16x8: f32

DCT16x8 / DCT8x16 (larger transforms use raw values, not normalized by DCT8). Reference: 1.21 (libjxl enc_ac_strategy.cc).

dct16x16: f32

DCT16x16. Reference: 1.34 (libjxl enc_ac_strategy.cc).

dct16x32: f32

DCT16x32 / DCT32x16. Reference: 1.49 (libjxl enc_ac_strategy.cc).

dct32x32: f32

DCT32x32. Reference: 1.48 (libjxl enc_ac_strategy.cc).

dct64x32: f32

DCT64x32 / DCT32x64. Reference: 2.25 (libjxl enc_ac_strategy.cc).

dct64x64: f32

DCT64x64. Reference: 2.25 (libjxl enc_ac_strategy.cc).

Implementations

impl EntropyMulTable

pub fn reference() -> Self

Default values matching libjxl enc_ac_strategy.cc:584.

pub fn experimental() -> Self

Experimental values from libjxl PR #4506 (Jon Sneyers, VarDCT cost tuning).

Changes vs reference:

• dct4x4: 1.08 → 0.88 (~19% reduction) — favor detail-preserving 4x4 sub-blocks
• identity: 1.0428 → 0.88 (~16% reduction) — favor pixel-copy for flat regions
• afv: 0.818 → 0.75 (~8% reduction) — favor corner DCT for edge blocks

Trait Implementations

impl Clone for EntropyMulTable

fn clone(&self) -> EntropyMulTable

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for EntropyMulTable

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.