cargo: bump rusty_h264 from 0.12.0 to 0.15.0

[dependabot]

Bumps rusty_h264 from 0.12.0 to 0.15.0.

Changelog

Sourced from rusty_h264's changelog.

[0.15.0] - 2026-09-05

A decoder speed release. Every change is byte-identical: the 68-stream ffmpeg identity gate and the x264 corpus hash gate pass on every commit, and each round landed behind a pinned CPU-time ABBA clock. No public signature moved except the removals listed below.

Changed — decoder speed (rusty_h264-decoder, rusty_h264-common, rusty_h264-accel)

• Entropy decode. The CAVLC residual decoder was reshaped (ten instruction cuts), the CABAC engine is register-resident, the CABAC residual parser is one call per macroblock over a significance bitmask with bypass-coded signs, and the significance map walks a reversed bitmask with tzcnt/blsr. The CAVLC syntax layer parses straight into the pooled per-macroblock job with an N-generic residual output (cavlc::decode_residual_block_into). Clocked per round on the dense CABAC stream: +5% CAVLC, +10-11% CABAC bins.
• SIMD reachability. A post-LTO census of the whole decode binary found ten functions still running scalar behind a kernel that existed: the 4×4 IDCT+add, 4-wide and 2-wide motion compensation, intra 4×4 prediction, the chroma MC rounding, and two kernels with no caller. All are now reached (accel::idct4x4_add, w4_via_w8 composition, mc_chroma_w4 for the 2-wide case, an intra 4×4 rewrite on an edge run). The batched multi-block IDCT that the first fix introduced measured as a loss (its gather/scatter was scalar) and was replaced by the per-block kernel.
• Kernel round. Ten instruction cuts inside the kernels themselves: pmulhrsw rounding in luma and chroma MC, pmaddubsw chroma MC with row reuse, pabsw/pblendvb in the deblock filter, register-resident vertical-edge filtering through an in-register 8×8 transpose, and the in-register 4×4 IDCT transpose.
• Routing round. Ten reroutes of content gates, priced with a scatter cost model (37 + 6·L + 9·nnz vs a flat ~60 for the dense AVX2 dequant): the nnz <= 6 sparse-dequant gate had routed the wrong way, so every coded block now takes the dense dequant; I16 and chroma reconstruction take coded masks from the parse side; intra 4×4 availability is a z-order table; the intra boundary-strength table is a constant. Clock vs the entropy rounds: crowd CABAC +7.4% (13/13, z=3.61), all-intra +2.4%, shields CAVLC +6.9%.
• Dense-over-scatter round. A fused scan-order dequant + IDCT + add kernel (accel::idct4x4_deq_add) on every 4×4 route, driven by a precomputed transform::DequantQp (flat and weighted) and predict::reconstruct_4x4_scan_into; a fused I16 luma DC path (accel::luma_dc_from_scan); prebuilt 8×8 dequant constants (Dequant8Qp, DQ8_FLAT); the nnz raster as lane shuffles (nnz_raster_from_z). The 42-instruction dequantize pass and the scan/scatter helpers have no caller left on the decode path (they stay exported). Clock on top of the routing round: crowd +1.8% (11/13), all-intra +1.1%.
• The decoder's static text shrank from 174,748 to 169,427 instructions across the last two rounds (post-LTO census of the decode binary).

Added

... (truncated)

Commits

• 425d76c release: 0.15.0
• e95024b Merge origin/main (0.14.0: no_std common/decoder, chip API, legacy-knob const...
• 7b9c10c docs: dense-over-scatter round — ten wins, dequant pass retired, clocks
• 02fd997 codec: dense-over-scatter round E — fused I16 luma DC, prebuilt 8x8 dequant c...
• 1fca392 docs: routing round clock — +9.0% vs kernel round, +7.4%/+2.4%/+6.9% vs round...
• df7c91c codec: dense-over-scatter round D — fused scan-order dequant+IDCT+add kernel ...
• a667159 docs: routing round — ten reroutes with the DQROUTE model
• 2ec4652 codec: routing round C — all coded blocks take the dense dequant, AVX2 twin o...
• 9bd44c0 codec: routing round B — coded-mask routes, parse-side chroma counts, z-order...
• 63523a2 codec: routing round A — build-time routing for env knobs, stats taps out of ...
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)