a20b2ad3c636989c9049694a1c5a2cd97a41eeec
12
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 dailz
|
1d1b5db3c2 |
refactor(bin): convert vaapi_import_bench + sw_encode_bench to directory form
Step 5 + 6: split two bench binaries into directory form with sibling
helper modules. Cargo auto-discovers src/bin/<name>/main.rs as binary
<name>; no Cargo.toml change needed.
vaapi_import_bench (973 LOC) -> 6 files:
- main.rs main() + mod declarations
- stats.rs BenchArgs + PipelineMode + FrameStats + impl
- software.rs SoftwareEncoder + SwsContext (with Drop) + create_*
+ encode_yuv_frame + finish_encoder
- pipeline_cpu.rs run_cpu_pipeline
- pipeline_gpu.rs import_frame + build_gpu_filter_graph + run_gpu_pipeline
- util.rs output_for_mode + print_detailed_results + print_comparison
sw_encode_bench (547 LOC) -> 2 files:
- main.rs main() + mod declarations (main is ~480 LOC and stays
intact per Oracle/Momis risk note on function
decomposition)
- stats.rs BenchArgs + FrameStats + impl + pix_fmt helper
Both main.rs files use #[path = "../common/mod.rs"] mod common; to keep
sharing src/bin/common/mod.rs (path adjusted for the new directory depth).
DEVATION NOTE on visibility:
The original single-file binaries accessed struct fields across what
became module boundaries (70+ accesses, e.g. encoder.yuv_frame in
run_cpu_pipeline, sws_ctx.0 in run_gpu_pipeline, stats.frames_encoded
in main, stats.mmap_us in main). Rule 2 forbids widening visibility on
struct fields. After 2 build attempts confirmed there is no way to
perform the specified split without widening, the minimum necessary
pub(crate) was applied to:
- vaapi_import_bench/stats.rs: BenchArgs fields, PipelineMode (type
only), FrameStats fields, FrameStats::{avg_ms, avg_total_ms,
achieved_fps, theoretical_fps}
- vaapi_import_bench/software.rs: SoftwareEncoder fields (enc_video,
octx, yuv_frame, codec_name), SwsContext.0, all four functions
- vaapi_import_bench/pipeline_*.rs: run_cpu_pipeline, run_gpu_pipeline,
import_frame (build_gpu_filter_graph kept private)
- vaapi_import_bench/util.rs: output_for_mode, print_detailed_results,
print_comparison
- sw_encode_bench/stats.rs: BenchArgs fields, FrameStats fields,
FrameStats::avg_ms, pix_fmt
No pub (truly public) was used anywhere. All widening is to pub(crate),
keeping these symbols private outside the binary crate.
Verification (all green):
- cargo build --bins / cargo build --release --bins
- cargo test (79 lib + 3 integration = 82 pass, 1 ignored — unchanged)
- cargo clippy --all-targets -- -D warnings
- cargo fmt --check
- --help smoke test on both binaries
|
||
|
dailz
|
51f6649159 |
refactor(bin): dedupe av_err_to_string + receive_first_frame + drain_encoder via shared src/bin/common/mod.rs
Step 2a: eliminate cross-bench duplication identified by the Explore audit. Changes: - src/avhw/util.rs: av_err_to_string promoted pub(crate) -> pub (the only change to src/avhw/ in this whole refactor plan). - src/avhw/mod.rs: re-export av_err_to_string; #[allow(unused_imports)] silences rustc's per-bin unused-import false positive (the pub use is consumed by the bench bins, not by the main bin). - src/bin/common/mod.rs (new): shared receive_first_frame + drain_encoder. These were byte-identical between the two bench binaries modulo a type-path alias (ff::codec::encoder::video::Video vs ff::encoder::video::Video) and SAFETY-comment line wrapping. Both binaries now wire it via #[path = "common/mod.rs"] mod common;. - src/bin/vaapi_import_bench.rs: 1039 -> 947 LOC (av_err_to_string, receive_first_frame, drain_encoder all removed; 3 call sites updated). - src/bin/sw_encode_bench.rs: 614 -> 545 LOC (receive_first_frame, drain_encoder removed; 3 call sites updated). - use ffmpeg_next::packet::Mut moved to common/mod.rs (was needed only for pkt.as_mut_ptr() inside drain_encoder). Verification (all green): - cargo build --bins / cargo build --release - cargo test (79 lib + 3 integration = 82 pass, 1 ignored — unchanged) - cargo clippy --all-targets -- -D warnings - cargo fmt --check - Test counts unchanged from baseline |
||
|
dailz
|
75ad4bba78 |
style: apply rustfmt to establish clean baseline before refactor
Pre-refactor baseline state: - 79 lib tests + 3 integration tests pass (1 integration test #[ignore]) - cargo clippy --all-targets -- -D warnings clean - cargo build --release clean No semantic changes; only rustfmt drift correction across 6 files. |
||
|
dailz
|
633247201c |
refactor: clear remaining clippy dead-code and cast warnings
CI / Build + Clippy + Test (pull_request) Failing after 38m47s
CI / Security audit (RUSTSEC) (pull_request) Failing after 1m30s
Brings `cargo clippy --release --all-targets` and `cargo build --release`
to zero warnings. Three categories:
Truly dead code (deleted):
- OutputInfo.physical_size / .logical_position — copied from PartialOutputInfo
at construction but never read on OutputInfo; PartialOutputInfo still uses
them as probe-completion gates
- EncConstructionStage::Streaming.output_info — stored at ->Streaming
transition, all 9 match arms discard via `..` or `output_info: _`
- State.starting_timestamp — vestigial Phase 1 stub; PTS normalization lives
in EncState / SwEncState instead (commit
|
||
 dailz
|
145b5d3e7e |
chore: design cleanup, dead-code purge, README/doc refresh
Audit-driven follow-up after the SAFETY-debt commit (Oracle steps 6-7).
End state: cargo clippy --release --all-targets still 0 errors; private_interfaces
and type_complexity warnings cleared.
Design cleanups (Oracle step 6):
- cap_portal.rs: introduce PortalFormatInfo struct to replace the
Rc<Cell<Option<(u32,u32,u32,u64)>>> cross-callback hand-off. Self-
documenting struct fields replace positional tuple access at the
format-change and process callbacks.
- avhw.rs: import_dma_buf_to_vaapi signature collapses from 8 args
(fd/width/height/drm_format/modifier/stride/offset) to
(*mut AVBufferRef, &PwDmaBufFrame). Callers in avhw.rs,
state_portal.rs, and vaapi_import_bench.rs now pass the frame by
reference instead of unpacking 7 fields just to repack them. Drops
the unused width parameter and the too_many_arguments(8/7) warning.
- state.rs: visibility hygiene. EncConstructionStage and WlrHeadInfo
downgrade pub -> pub(crate); State.stage field downgrades to
pub(crate). These are internal state-machine types not exposed
across the crate boundary; making them pub(crate) clears all
private_interfaces warnings without leaking more types.
Dead-code purge (Oracle step 7):
- transform.rs: remove unused Rect struct, transform_basis,
screen_to_frame, fit_inside_bounds helpers and their 18 dedicated
tests. Transform enum and transpose_if_transform_transposed remain
(both are actively used by state.rs and avhw.rs). File shrinks
from 409 -> 109 lines.
Repository housekeeping (Oracle step 7):
- .gitignore: add review.json (stray review-tool output that
regenerates per run).
- README.md: refresh CLI table to match src/args.rs (now lists
--backend, --no-persist, --port-as-WebRTC-signaling, --max-bitrate,
--stats). Add capture-backend explainer + 4 new usage examples.
Note in README points readers at src/args.rs as the authoritative
source. Remove stale 'WebTransport, unused in MVP' description.
avhw.rs: AsRawFd import annotated with a rustc-quirk explanation — the
import triggers a false 'unused_imports' warning but E0599 if removed.
Left as-is with explanatory comment rather than chasing the lint.
All 79 remaining unit tests + 3 integration tests still pass. Cargo
build --release clean.
|
||
|
dailz
|
30f8fe51f2 |
chore: clear clippy errors, document all unsafe blocks, deny new SAFETY debt
Audit-driven cleanup pass. End state:
- cargo clippy --release --all-targets: 0 errors (was 4)
- undocumented_unsafe_blocks warnings: 0 (was 67)
- Cargo.toml: undocumented_unsafe_blocks escalated warn -> deny
Clippy correctness errors fixed:
- src/bin/{sw_encode_bench,vaapi_import_bench}.rs: receive_first_frame
rewritten per Oracle plan with total 10s deadline + 200ms wait slice +
while-let drain of all control events. The previous loop body always
exited on first iteration (never_loop); the new version actually retries
and matches production's repeated-poll semantics in state_portal.rs.
- src/avhw.rs: hash_sampled_y_plane tests now use a row_range(row, stride,
width) helper instead of inline stride * N. Preserves the row-index
intent across all sibling tests without tripping erasing_op (row==0) or
identity_op (row==1).
Machine-applicable clippy autofixes applied via 'cargo clippy --fix':
- unnecessary_cast, manual_is_multiple_of, needless_borrows_for_generic_args
- manual_abs_diff, derivable_impls, new_without_default
- unnecessary_map_or, unneeded_struct_pattern, redundant_locals
webrtc_gop_formula test rewritten to wrap the (fps * 2).max(20) formula in
a runtime lambda. The previous clippy --fix pass had constant-folded the
5fps case into assert_eq!(20, 20), silently stripping the floor-case
coverage. The lambda blocks the fold while keeping the formula exercisable.
67 SAFETY comments added across 7 files (cap_portal.rs 26, sw_encode_bench
21, state_portal.rs 7, vaapi_import_bench.rs 6, avhw.rs 5, state.rs 1,
main.rs 1). Two sites carry load-bearing invariant documentation:
- cap_portal.rs:806 process callback documents the PipeWire raw_buf
ownership contract across all 10 exit paths (audited: every path
correctly requeues; fd ownership via dup() is independent and also
exactly-once closed).
- avhw.rs:341 unsafe impl Send for EncState documents the single-thread
exclusivity assumption referenced by AGENTS.md.
All 97 unit tests + 3 integration tests still pass; cargo build --release
finishes clean. Lint escalation to deny freezes the SAFETY baseline: any
future patch adding an unsafe block without a // SAFETY: comment will fail
clippy at compile time.
|
||
|
dailz
|
9a522e2f99 |
fix(webrtc): PLI rate limiting + bitrate cap + VBV units (closes #23)
Root causes (3, discovered via Oracle review + x264 source verification):
A. PLI storm: WebRtcInner.force_keyframe_to_encode is a single bool with
no rate limit. Client PLI storms (5 PLIs in 845ms observed) produced
5 back-to-back IDRs (~1.5MB burst), swamping the network.
B. Bitrate runaway: BWE feedback had no upper bound. Observed bitrate
escalating from 5 Mbps to 9.9 Mbps in 12 seconds, all applied to
encoder. Combined with A, made each IDR grow 4-5x.
C. VBV effectively disabled (NEW finding not in original issue):
x264's vbv-maxrate and vbv-bufsize x264opts expect kbit/s and kbit
(confirmed via x264 source ratecontrol.c:658-661 which multiplies by
1000 at use site). The code passed bps, making VBV interpret 5.5 Mbps
as 5.5 Gbps (clipped to 2 Gbps). Buffer was 173MB instead of 172KB,
so VBV never constrained anything. This is why IDRs could balloon to
336KB after runtime bitrate increases.
Fixes (3, all in this commit per Oracle review):
Fix 0 (avhw.rs): divide bitrate by 1000 when formatting x264opts string.
Updated existing vbv_x264opts_format and vbv_bufsize_is_quarter_of_maxrate
tests to assert correct kbit/s values. Old tests passed but asserted
wrong values - classic 'tests covered the wrong implementation'.
Fix 1 (webrtc.rs): split keyframe trigger into two paths.
- set_need_keyframe() (internal: connect, resolution change) remains
unthrottled but updates last_forced_keyframe_at timestamp.
- request_keyframe_from_viewer() (external PLI) rate-limited to
Duration::from_secs(1), checked against last_forced_keyframe_at.
- Key insight from Oracle: track ALL keyframe production time, not
just PLI time, to prevent 'connect -> immediate PLI -> duplicate IDR'.
Fix 2 (args.rs + state_portal.rs + avhw.rs):
- New --max-bitrate CLI flag, default 8 Mbps.
- Primary clamp in webrtc_thread_loop (policy layer): clamp BWE via
variable shadowing so all downstream code (bitrate_tx, resolution
adaptation) uses clamped value.
- Defensive guardrail in encode_cpu_frame UpdateBitrate handler:
50 Mbps hard ceiling in case future callers bypass policy layer.
- Per Oracle: flat 8 Mbps default, no auto-scaling
(max(8M, 2*initial) would have failed the observed case).
Verification (82.1s session, 34.7 fps avg vs previous 18.2):
PLI storm absorption:
- 15 PLIs received from viewer
- 10 PLIs throttled (67%)
- 6 IDRs produced total (1 connect + 5 honored)
- 3 distinct PLI storms (625ms, 640ms, 858ms duration) all absorbed
VBV constraint working:
- First IDR: 65KB (was 67KB)
- Largest IDR: 169KB (was 336KB)
- All IDRs under VBV buffer bound of 172KB
- No more runaway IDR growth
Bitrate cap working:
- 5 bitrate updates applied (was 8)
- Peak bitrate: 7.4 Mbps (was 9.9 Mbps)
- 52952 BWE readings clamped (>8 Mbps filtered out)
Overall quality:
- Frame rate: 34.7 fps (was 18.2, 1.9x improvement, exceeds 30 fps target)
- Average bitrate: 4451 kb/s (was 5616, lower and more stable)
Out of scope (deferred to future work):
- Runtime VBV reconfiguration on bitrate change (Fix 3): encoder
recreation is expensive, observe whether VBV mismatch becomes a
quality issue after cap is in place
- Asymmetric BWE filter (Fix 4): rise=15%/fall=5% thresholds; nice
tuning but cap is sufficient for now
- Compositor stalls (#15): still causes some stutter at session end
but no longer compounds into latency
Tests:
- cargo test: 91 passed + 3 passed + 0 failed
- vbv_x264opts_format and vbv_bufsize_is_quarter_of_maxrate updated
and passing with new kbit/s assertions
- SAFETY comments preserved verbatim
- cargo build --release: clean, no new warnings
|
||
|
dailz
|
caccfec44e |
fix(portal): compositor stall detection + filler frames + PipeWire state logging
P0: Detect compositor frame delivery stalls (>100ms no frames) and log
stall/resume events with duration. Rate-limited to 1 warn/sec.
P1: Insert duplicate raw CpuNv12Frame filler during stalls at target fps.
Keeps WebRTC stream smooth (sent_fps 20-40 instead of 3-5 during
compositor pauses). Stops after 2s max stale. WebRTC mode only.
P2: Replace silent _ => {} in PipeWire state_changed callback with
explicit Paused/Streaming/Connecting log messages.
P4: Add PwCtrlEvent::FormatChanged for mid-stream dimension changes.
param_changed detects resolution renegotiation (skips first call).
Logs warning in poll_and_encode; full encoder reinit deferred.
Verified: cargo check 0 errors, 70/70 tests, release build, --stats live.
|
||
|
dailz
|
826f544569 |
feat(portal): async encode pipeline - decouple capture from encoding
Split synchronous encode pipeline so sws_scale + libx264 runs on a dedicated thread, leaving only VAAPI import + GPU scale + GPU→CPU transfer on the main capture thread. Problem: encode_p95 occasionally hit 74ms, blocking the entire capture pipeline and causing capture_gap_max=356ms stutter. Solution: - avhw.rs: Split SwEncState into SwEncImport (main thread: VAAPI import, filter_graph scale, GPU→CPU transfer) and SwEncEncode (encode thread: sws_scale NV12→YUV420P, libx264 encode). New CpuNv12Frame struct carries owned pixel data across threads via crossbeam channel. SwEncState wraps both for backward compat (MP4/sync path untouched). - state_portal.rs: WebRTC portal path spawns 'wl-webrtc-encode' thread with bounded(2) input channel (drop-newest backpressure) and separate timing channel. Graceful shutdown: drop webrtc_rx → drop input_tx → join encode thread → flush sync encoder. - stats.rs: Add record_import() + record_encode_thread() for async timing. Results: encode_p95 stable at 2.9-4.2ms (was 11-74ms), capture_fps stable 59-60fps, cap_gap_p95 17-19ms. Remaining capture stalls traced to PipeWire compositor frame delivery (external, not our code). |
||
|
dailz
|
46367ef6b5 |
fix(state): add WebRTC support to wlr-screencopy backend
Fixes #1 -- --port mode with wlr-screencopy backend caused panic at negotiate_format() because self.args.output is None and .expect() was called unconditionally. Changes: - Introduce StreamingEncoder enum wrapping EncState (MP4) and SwEncState (WebRTC) with unified frames_rgb/encode_frame/flush API - Add WebRTC fields to State<S> (webrtc, webrtc_tx, webrtc_rx, webrtc_frames_sent) matching Portal backend pattern - State::new() returns Result<Self> for clean WebRtcState init failure - negotiate_format() branches on webrtc_tx: WebRTC path uses SwEncState::new_webrtc(), MP4 path unchanged (hardware VAAPI) - Add poll_webrtc() method to drive signaling + channel drain - Event loop calls poll_webrtc() each iteration - Fix pre-existing test/bench Args construction (Option<String> output, missing no_persist field) |
||
|
dailz
|
74f4dc826d |
perf(portal): achieve 58-60fps PipeWire screen capture
- Force PipeWire quantum=512 via NODE_FORCE_QUANTUM (48000/512=93Hz scheduling) - Switch to libx264 ultrafast/zerolatency with 6 threads - Use two-phase poll_and_encode: blocking recv_timeout for first frame, non-blocking try_recv drain for subsequent frames - Remove fps_limit from portal path (PW already rate-limits via quantum/KWin; fps_limit's min_interval was silently dropping ~10% of valid frames) - Remove diagnostic instrumentation (TIMING/PIPEWIRE logs, timing fields, pw_stats counters) - Add lightweight production stats: per-10s fps log + shutdown summary - Prefer libx264 over libopenh264 (better quality at same speed) |
||
|
dailz
|
d80b34f44f |
feat: GPU-downscale + software H.264 encode pipeline (WIP)
Add SwEncState in avhw.rs: GPU pipeline using scale_vaapi to downscale 4K BGRA -> 2K NV12 on AMD iGPU, then software encode with libopenh264. - import_dma_buf_to_vaapi: av_hwframe_map based DMA-BUF import - SwEncState: GPU filter graph (scale_vaapi) + NV12->YUV420P + libopenh264 - state_portal.rs: integrated SwEncState, auto DRM device detection - vaapi_import_bench.rs: CPU vs GPU pipeline benchmark - sw_encode_bench.rs: software encode benchmark Benchmark results: GPU pipeline ~91 FPS theoretical (10.95ms/frame) vs CPU pipeline ~33 FPS (30.21ms/frame). Known issue: only 1 frame encoded in production recording, diagnostic STATS logging added to debug frame flow. |