[3/4] Optimize SILK LPC synthesis#116
Closed
zshang-oai wants to merge 1 commit into
Closed
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #116 +/- ##
==========================================
+ Coverage 82.75% 82.80% +0.04%
==========================================
Files 26 26
Lines 5626 5652 +26
==========================================
+ Hits 4656 4680 +24
- Misses 745 746 +1
- Partials 225 226 +1
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
RFC 6716 / 8251 conformationStatus: pass The action extracts the RFC 6716 reference implementation, applies the RFC 8251 decoder update patch, and then builds the patched reference tools. Legend: numeric cells are Inputs use the shared RFC 6716 / RFC 8251 bitstream corpus; accepted references follow RFC 8251 Section 11.
Run output |
zshang-oai
force-pushed
the
codex/reduce-silk-decode-hot-path
branch
from
f861bef to
7a0a95c
Compare
zshang-oai
force-pushed
the
codex/reduce-silk-lpc-synthesis
branch
from
3845541 to
7a0a95c
Compare
There was a problem hiding this comment.
Pull request overview
This PR further optimizes the SILK decoder LPC synthesis hot path by removing remaining per-sample branching in the first-subframe case and adding a specialized wideband (16-tap) steady-state implementation.
Changes:
- Replaces dual normalized/reversed LPC coefficient prep with a single reversed+normalized coefficient array.
- Reworks first-subframe LPC synthesis to reuse the steady-state path by constructing a contiguous history+output buffer.
- Adds an unrolled 16-tap steady-state LPC synthesis fast path for wideband frames while keeping the generic loop for 10-tap cases.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
zshang-oai
force-pushed
the
codex/reduce-silk-lpc-synthesis
branch
from
02c9e11 to
a19859e
Compare
zshang-oai
changed the title
zshang-oai
changed the base branch from
codex/reduce-silk-decode-hot-path
to
main
zshang-oai
force-pushed
the
codex/reduce-silk-lpc-synthesis
branch
from
02c9e11 to
a19859e
Compare
Contributor
Author
|
Abandoning this draft; the measured gain on the production-like 48 kHz stereo conformance decode workload is too small to carry forward. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Reduce the remaining low-risk LPC synthesis overhead in the SILK decode hot path. This now sits directly on
mainafter #115 merged.Major changes
96as an implicit assumption.Why
After #114 and #115 remove the larger allocation and copy costs, the remaining SILK decode profile still spends visible time in the branchy first-subframe LPC reconstruction and in the generic 16-tap loop used by wideband frames.
This keeps the same floating-point decoder behavior, but removes per-tap history branching on the first subframe and lets the common 16-tap case run without the generic inner loop.
Validation
Ran:
GOCACHE=/private/tmp/opus-go-build-pr116-rebase GOLANGCI_LINT_CACHE=/private/tmp/opus-golangci-lint-pr116-rebase golangci-lint run GOCACHE=/private/tmp/opus-go-build-pr116-rebase go test ./...End-to-end stress harness only, no focused microbenchmark.