chore: release v0.17.0

Cargo.toml

@@ -22,7 +22,7 @@ members = [
 exclude = ["fuzz"]
 
 [workspace.package]
-version = "0.16.0"
+version = "0.17.0"
 edition = "2021"
 license = "BSD-2-Clause"
 repository = "https://github.com/remade-with-rust/rusty_h264"
@@ -48,9 +48,9 @@ wide = { version = "0.7", default-features = false }
 # scalar oracle arm would be dead while every gate claimed to test it. Members
 # that want defaults opt back IN (enabling is allowed; only disabling is
 # ignored).
-rusty_h264-common = { path = "crates/rusty_h264-common", version = "0.16.0", default-features = false }
-rusty_h264-encoder = { path = "crates/rusty_h264-encoder", version = "0.16.0", default-features = false }
-rusty_h264-decoder = { path = "crates/rusty_h264-decoder", version = "0.16.0", default-features = false }
+rusty_h264-common = { path = "crates/rusty_h264-common", version = "0.17.0", default-features = false }
+rusty_h264-encoder = { path = "crates/rusty_h264-encoder", version = "0.17.0", default-features = false }
+rusty_h264-decoder = { path = "crates/rusty_h264-decoder", version = "0.17.0", default-features = false }
 
 [profile.release]
 opt-level = 3

crates/rusty_h264-accel/CHANGELOG.md

@@ -0,0 +1,20 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.17.0](https://github.com/Remade-With-Rust/rusty_h264/compare/rusty_h264-accel-v0.16.0...rusty_h264-accel-v0.17.0) - 2026-10-25
+
+### Other
+
+- overflow-checks cost 22.5% of decode — moved to release-checked
+- audit pass 2 — 6% to 89% of in-scope gates, every unit >=85%
+- close H-03/04/05/07/13/14/15/16/31/39 + threat model
+- use-protection-please first pass — 41 gates x 8 units, depth=deep
+- an intra record is one boolean, and the filter stops touching a TLS
+- pack a uniform macroblock from one block, not sixteen
+- one motion test per macroblock edge when both sides are uniform

crates/rusty_h264-cli/Cargo.toml

@@ -18,7 +18,7 @@ path = "src/main.rs"
 [dependencies]
 # CLI default = asm. rusty_alloc comes from rusty_h264-common's default features
 # (pulled transitively); keep common's defaults on when disabling facade asm.
-rusty_h264 = { path = "../rusty_h264", version = "0.16.0", default-features = false, features = ["std", "global-alloc"] }
+rusty_h264 = { path = "../rusty_h264", version = "0.17.0", default-features = false, features = ["std", "global-alloc"] }
 rusty_h264-common = { workspace = true, features = ["std"] }
 
 

crates/rusty_h264-common/CHANGELOG.md

@@ -0,0 +1,36 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.17.0](https://github.com/Remade-With-Rust/rusty_h264/compare/rusty_h264-common-v0.16.0...rusty_h264-common-v0.17.0) - 2026-10-25
+
+### Other
+
+- bump rusty_alloc-api to =2.2.6 ([#19](https://github.com/Remade-With-Rust/rusty_h264/pull/19))
+- bump rusty_alloc-api to =2.2.5 ([#17](https://github.com/Remade-With-Rust/rusty_h264/pull/17))
+- 87% of every allocation was one Box::new that skipped its own pool
+- narrow the motion grids to (i16,i16) - 2.90 GB of pool clear removed
+- narrow the reference-index grids i32 -> i8 (-103 MB of memset)
+- overflow-checks cost 22.5% of decode — moved to release-checked
+- audit pass 2 — 6% to 89% of in-scope gates, every unit >=85%
+- close H-03/04/05/07/13/14/15/16/31/39 + threat model
+- use-protection-please first pass — 41 gates x 8 units, depth=deep
+- put the 8x8-transform flag in the intra table, so the filter stops asking
+- record one more refutation on the derivation's argument list
+- move the derivation's output pointer into a register argument
+- build the packed record where it lives (spills -72%)
+- add a spill column to the row-hook census, and record what it settled
+- outline the per-lane edge arms, now the minority case
+- the filter keeps boundary strengths at byte width
+- an intra record is one boolean, and the filter stops touching a TLS
+- coded flags a word at a time, and a provable kind row
+- pack a uniform macroblock from one block, not sixteen
+- one motion test per macroblock edge when both sides are uniform
+- take the uniformity answer from mb_kind instead of recomputing it
+- fold the paired motion tests, share the edge nnz mask, split the hot prefix
+- derive boundary strengths at u8, not i32 (-223 instrs on the decoder's derivation path)

crates/rusty_h264-common/Cargo.toml

@@ -20,7 +20,7 @@ wide = { workspace = true }
 bytemuck = "1"
 # Optional openh264 asm kernels (deblock, …). The unsafe FFI is quarantined inside
 # that crate; we call only its safe wrappers, so this crate stays forbid(unsafe).
-rusty_h264-accel = { path = "../rusty_h264-accel", version = "0.16.0", optional = true }
+rusty_h264-accel = { path = "../rusty_h264-accel", version = "0.17.0", optional = true }
 # Our own allocator, installed process-wide. Declared HERE because every crate and
 # every one of the 43 example/bench binaries links this crate, so one declaration
 # covers all routes. See the hazard note in src/lib.rs before making this non-optional.

crates/rusty_h264-decoder/CHANGELOG.md

@@ -0,0 +1,41 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.17.0](https://github.com/Remade-With-Rust/rusty_h264/compare/rusty_h264-decoder-v0.16.0...rusty_h264-decoder-v0.17.0) - 2026-10-25
+
+### Fixed
+
+- the profile census compiles again, and the decoder builds no_std ([#20](https://github.com/Remade-With-Rust/rusty_h264/pull/20))
+
+### Other
+
+- 87% of every allocation was one Box::new that skipped its own pool
+- the plane recycle pool was never fed - 4.34 GB of wasted zeroing
+- narrow the motion grids to (i16,i16) - 2.90 GB of pool clear removed
+- the List-1 grids are B-only, so stop clearing them without B (-36%)
+- four more, from the encoder and the PPS-gated grids
+- the mvd cache stores what the context can read, not what was parsed
+- second pass of the copy hunt — the census's own two blind spots
+- carry the POC lookup table, not its per-block expansion (-37 MB)
+- narrow the reference-index grids i32 -> i8 (-103 MB of memset)
+- overflow-checks cost 22.5% of decode — moved to release-checked
+- audit pass 2 — 6% to 89% of in-scope gates, every unit >=85%
+- close H-03/04/05/07/13/14/15/16/31/39 + threat model
+- use-protection-please first pass — 41 gates x 8 units, depth=deep
+- no production panic sites — 47 to 0 (H-18)
+- put the 8x8-transform flag in the intra table, so the filter stops asking
+- move the derivation's output pointer into a register argument
+- build the packed record where it lives (spills -72%)
+- mark single-partition B macroblocks uniform (mb_uniform now -86..-96%)
+- coded flags a word at a time, and a provable kind row
+- pack a uniform macroblock from one block, not sixteen
+- mark the B classes whose motion the decoder already knows is uniform
+- one motion test per macroblock edge when both sides are uniform
+- take the uniformity answer from mb_kind instead of recomputing it
+- derive boundary strengths at u8, not i32 (-223 instrs on the decoder's derivation path)

crates/rusty_h264-encoder/CHANGELOG.md

@@ -0,0 +1,24 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.17.0](https://github.com/Remade-With-Rust/rusty_h264/compare/rusty_h264-encoder-v0.16.0...rusty_h264-encoder-v0.17.0) - 2026-10-25
+
+### Fixed
+
+- the profile census compiles again, and the decoder builds no_std ([#20](https://github.com/Remade-With-Rust/rusty_h264/pull/20))
+
+### Other
+
+- narrow the motion grids to (i16,i16) - 2.90 GB of pool clear removed
+- four more, from the encoder and the PPS-gated grids
+- narrow the reference-index grids i32 -> i8 (-103 MB of memset)
+- overflow-checks cost 22.5% of decode — moved to release-checked
+- audit pass 2 — 6% to 89% of in-scope gates, every unit >=85%
+- close H-03/04/05/07/13/14/15/16/31/39 + threat model
+- use-protection-please first pass — 41 gates x 8 units, depth=deep

crates/rusty_h264-encoder/Cargo.toml

@@ -19,7 +19,7 @@ rusty_h264-common = { workspace = true, default-features = false }
 # Optional hand-tuned asm acceleration (openh264 BSD-2 kernels). The unsafe FFI is
 # fully quarantined inside that crate; this crate calls only its safe wrappers, so
 # it stays forbid(unsafe) even with `--features asm`.
-rusty_h264-accel = { path = "../rusty_h264-accel", version = "0.16.0", optional = true }
+rusty_h264-accel = { path = "../rusty_h264-accel", version = "0.17.0", optional = true }
 
 [dev-dependencies]
 # The decoder is our round-trip oracle: encoder tests decode their own output and

crates/rusty_h264/CHANGELOG.md

@@ -0,0 +1,17 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.17.0](https://github.com/Remade-With-Rust/rusty_h264/compare/rusty_h264-v0.16.0...rusty_h264-v0.17.0) - 2026-10-25
+
+### Other
+
+- overflow-checks cost 22.5% of decode — moved to release-checked
+- audit pass 2 — 6% to 89% of in-scope gates, every unit >=85%
+- close H-03/04/05/07/13/14/15/16/31/39 + threat model
+- use-protection-please first pass — 41 gates x 8 units, depth=deep