Raw dependency graph data (JSON)

GStreamer Plugin for Edge Impulse

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-09-08
Readiness: orange
Scope: RISC-V (riscv64/linux) support status for GStreamer Plugin for Edge Impulse
Audience: Technical leadership, resource allocation strategy
Verification policy: Every claim is cross-referenced to a primary upstream source. Items that could not be verified against a second source are marked [NEEDS VERIFICATION].

1. Project Overview

GStreamer Plugin for Edge Impulse (edgeimpulse/gst-plugins-edgeimpulse) is a Rust-based GStreamer plugin providing seven pipeline elements (audio inference, video inference, crop, overlay, OCR, sink, continue-if) that run Edge Impulse machine-learning models inside a GStreamer pipeline. It is a cdylib Cargo project (Cargo.toml/Cargo.lock, no CMake, no autotools) built on gstreamer-rs bindings, with a default feature set (ffi, ocr, presentation, ingestion) that pulls in the edge-impulse-runner FFI crate, cairo-rs/pango/pangocairo for overlay rendering, and ocrs/rten for self-contained OCR.

Governance: No foundation membership (not Linux Foundation, Apache, or CNCF affiliated). No MAINTAINERS, OWNERS, CODEOWNERS, GOVERNANCE.md, or CONTRIBUTING.md exist anywhere in the repository history. Licensed under Clear BSD License (BSD-3-Clause-Clear), Copyright (c) 2025 EdgeImpulse Inc..

Corporate sponsor: Single-vendor repository owned by Edge Impulse Inc. (acquired by Qualcomm in 2025, per publicly known corporate history [NEEDS VERIFICATION - not independently confirmed within this repo’s own materials]). Repository itself carries no documentation of that acquisition or any resulting platform-support commitments.

Maintainer concentration: Of 132 commits (2025-02-10 to 2026-07-27), Fernando Jimenez Moreno (personal Gmail address, ferjmoreno@gmail.com) authored ~96% (127 commits). Mateusz Majchrzycki (mateusz@edgeimpulse.com, corporate address) authored 2 commits. Raul James (raul@nobi.io, an unrelated external company) authored 1 commit. This is a de facto single-maintainer project with a token corporate commit, not a multi-company-governed one.

Community culture on new ports: No tier policy, no CONTRIBUTING.md, and no documented process exists for proposing a new platform port. Platform support is expressed only as ad hoc Cargo feature flags and Docker build variables (TARGET_LINUX_AARCH64, USE_ETHOS, USE_AKIDA, USE_MEMRYX, USE_QUALCOMM_QNN, TENSORRT_VERSION, etc.), not as a formal acceptance policy. There is no articulated stance toward RISC-V or any other new architecture - the question has never been raised in the project’s history.

2. Port History and Upstreaming Timeline

Date Event Source
N/A No RISC-V port, commit, PR, or issue exists at any point in the project’s history Full history search: git log --all --grep, pickaxe -S"riscv", and repo-wide grep, zero matches

No RISC-V-related contributor has ever touched this repository. There is no upstreaming timeline to report because no RISC-V work has ever been proposed, in this repo or in its issue/PR tracker. The project is not fully upstream on RISC-V in the sense of “already merged” - it is simply absent: zero commits, zero PRs, zero issues referencing RISC-V across the entire 132-commit history and all 54 pull requests (#1-#56) and both issues (#8, #49) enumerated.

3. Upstream Support Tier

No formal tier policy exists (see Section 1). Platform support tiers are inferred only from what CI actually exercises and what cross-compilation tooling exists.

Architecture CI builds CI tests Official binary release Cross-compile tooling
amd64 (x86_64) Yes - ci.yml, docs.yml, e2e.yml all run on ubuntu-latest Yes - cargo test for runner-free and eim feature sets in ci.yml No releases published for any architecture (“There aren’t any releases here.”) Native, implicit
arm64 (aarch64) No dedicated CI job; build-only via manual local Docker script (test-cross-compilation.sh, not invoked by any GitHub Actions workflow) No No releases published for any architecture aarch64-unknown-linux-gnu target, .cargo/config.toml linker entry, dedicated Dockerfile/docker-compose.yml aarch64-build service
riscv64 No No No releases published for any architecture; no PyPI, Ubuntu, or Arch RISC-V package exists under this name None - no target triple, no toolchain, no Dockerfile stage, no compose service

Evidence: .github/workflows/ci.yml, docs.yml, e2e.yml (all ubuntu-latest, no matrix), GitHub Releases page (“There aren’t any releases here”).

4. Technical Architecture and RISC-V-Specific Subsystems

The plugin itself is inference-pipeline glue code (GStreamer elements calling into TFLite/ONNX/EIM backends); the model runtime is out-of-tree via the edge-impulse-runner/edge-impulse-ffi-rs crates. There is no architecture-specific compute code in this repository at all, for any architecture.

  • src/ contains 29 pure-Rust files (audio/, crop/, ocr/, overlay/, filter/, sink/, tracker/kalman.rs, video/, common.rs, detection.rs, meta.rs, resize.rs, lib.rs). No file named *riscv*, *_x86*, *_arm*, or *_amd64* exists.
  • grep -n "target_arch" across the whole repository returns 0 matches. No #[cfg(target_arch = ...)] conditional compilation exists anywhere.
  • No #ifdef __riscv guards (confirmed both by local grep and GitHub code-search index).
  • The only conditional-compilation target in Cargo.toml is [target.'cfg(target_os = "macos")'.dev-dependencies] - an OS gate, not a CPU-architecture gate.
Component amd64 arm64 riscv64
Core plugin logic (Rust) Native, full Cross-compiles via Docker, generic Rust (no arch-specific code) Absent - no target triple registered
Inference backend (ffi/edge-impulse-ffi-rs) Native TFLite prebuilt libs Native aarch64 build path documented Not evaluated in-repo; see Section 9 for the transitive Edge Impulse Inferencing SDK (C++) gap, which has an unconditional non-aarch64/non-Apple fallback to x86_64 prebuilt libraries
OCR backend (rten/ocrs) AVX2/AVX-512 SIMD NEON SIMD Scalar fallback only - RVV not mentioned anywhere in rten’s documented SIMD backends (AVX2, AVX-512, Arm Neon, WebAssembly SIMD)
Overlay rendering (Cairo/Pango) Native Native Builds (per Section 9 dependency analysis) but compositing runs scalar-only pending an unmerged pixman-rvv.c upstream

Because this project is inference-glue code rather than a performance-differentiating library in its own right, it is not classified as optimization-purpose (see Section 13); this section documents the absence of arch-specific code for completeness, not as an optimization-gap analysis.

5. Build System, Cross-Compilation, and Toolchain

  • Build system: Cargo/Rust only. No CMakeLists.txt, no cmake/ directory, no BUILDING.md/INSTALL/docs/building.md exist anywhere in the repository.
  • Stated MSRV: rust-version = "1.75" in Cargo.toml; the Docker build image itself uses Rust 1.90 (rust:1.90-slim-bullseye).
  • The only documented cross-compilation target in the entire repository is aarch64-unknown-linux-gnu:
    • Toolchain: gcc-aarch64-linux-gnu/g++-aarch64-linux-gnu via dpkg --add-architecture arm64 + apt, rustup target add aarch64-unknown-linux-gnu.
    • Build command: cargo build --target aarch64-unknown-linux-gnu --features ffi,ocr,presentation,ingestion --release with env CC_aarch64_unknown_linux_gnu=aarch64-linux-gnu-gcc, CXX_aarch64_unknown_linux_gnu=aarch64-linux-gnu-g++, TARGET_LINUX_AARCH64=1, USE_FULL_TFLITE=1.
    • Orchestration: docker-compose.yml service aarch64-build, driven by test-cross-compilation.sh (a manual/local dev script, not invoked by any GitHub Actions workflow).
  • No QEMU usage exists anywhere in the repository.
  • riscv64 build: not documented, not implemented. There is no riscv64gc-unknown-linux-gnu target addition, no riscv toolchain package installation, no Dockerfile stage, and no docker-compose.yml service for riscv64. A user attempting a riscv64 build would need to add the Rust Tier-2 target, install a riscv64 cross-toolchain, and replicate the aarch64 Docker/CI plumbing from scratch - none of this exists today.
  • Known build failures on riscv64: none documented, because no riscv64 build has ever been attempted per the available evidence.

6. Feature Coverage and Gap Analysis vs arm64 and amd64

Feature (Cargo flag) amd64 arm64 riscv64
ffi (default) - core ML inference via Edge Impulse Inferencing SDK Full Full (documented cross-compile path) Not buildable as-is: edge-impulse-ffi-rs’s CMake logic has an unconditional non-aarch64/non-Apple fallback to x86_64 prebuilt TFLite libraries, meaning a naive riscv64 build would silently link wrong-ISA binaries rather than fail cleanly (see Section 9)
ocr (default) - rten/ocrs text detection/recognition Full, AVX2/AVX-512 accelerated Full, NEON accelerated Would compile via the Rust Tier-2 riscv64gc-unknown-linux-gnu target (pure Rust, no C++ FFI) but runs entirely unaccelerated - scalar fallback only
presentation (default) - Cairo/Pango overlay text rendering Full Full Builds (all C dependencies - Cairo, Pango, HarfBuzz, FreeType - are packaged for Ubuntu 26.04 riscv64) but compositing is scalar-only pending pixman’s unmerged RVV backend
ingestion (default) Full Full Not independently verified; no known blocker found but also no riscv64 evidence of any kind
eim (non-default, external EIM process backend) Full Full Not independently verified

Functional gaps: The default feature set (ffi is in default = [...]) will not produce a working ffi/eim-backed inference element on riscv64 because of the unconditional x86_64 fallback in the transitive edge-impulse-ffi-rs build logic (see Section 9). This is the single hard functional blocker.

Performance gaps: OCR (rten/ocrs) and overlay compositing (Cairo via pixman) would both run correctly but unaccelerated on riscv64, versus SIMD-accelerated paths on amd64/arm64. Magnitude of that gap: Data not available - no riscv64 benchmark of any kind exists for this plugin or its dependencies (see Section 11).

Security hardening gaps: Data not available - no riscv64 build has been produced to assess hardening flags, ASLR behavior, or any other security posture on this architecture.

NaN / floating-point semantics issues: Data not available - no riscv64 build or test run exists to surface such issues; no issue report of any kind mentions floating-point or NaN behavior for this project on any architecture.

7. CI/CD Infrastructure

No riscv64 CI exists. Confirmed by reading all three GitHub Actions workflow files directly at HEAD (c763395b638eaef46b8ff1e8f58f8e2de057b692, 2026-07-27):

  • .github/workflows/ci.yml: triggers on push/PR to main; single job on runs-on: ubuntu-latest; no strategy: matrix: block; installs GStreamer/Rust deps, runs cargo fmt --check, several cargo build --no-default-features --features <set> combinations, a backend-guard negative-build check, and cargo test for two feature sets. No --target, no QEMU, no docker buildx.
  • .github/workflows/docs.yml: triggers on push to main and workflow_dispatch; both build and deploy jobs run on ubuntu-latest; builds cargo doc and publishes to GitHub Pages.
  • .github/workflows/e2e.yml: triggers on push/PR to main, gated to skip fork PRs; single job on runs-on: ubuntu-latest; runs docker compose -f docker-compose.test.yml up --build with EI_ENGINE: tflite.
  • .gitlab-ci.yml, Jenkinsfile, .cirrus.yml: none exist in the repository.
  • Repo-wide grep -rniE "riscv" (excluding .git) across the entire working tree: zero matches.
  • No RISE runner references (no riseproject-dev mentions, no RISE runner labels) anywhere in the CI configuration.
Architecture CI: build CI: test CI: release-blocking Runner
amd64 Yes Yes Yes (implicit, only CI target) GitHub-hosted ubuntu-latest (native)
arm64 No (manual local Docker script only, not in CI) No No N/A (not run in CI)
riscv64 No No No None

8. Distribution and Release Status

  • GitHub Releases: Zero releases exist for the repository, for any architecture - “There aren’t any releases here.” (releases page). This means there are no release assets of any kind, not merely an absence of riscv64 assets.
  • PyPI: Package does not exist under the name gstreamer-plugin-for-edge-impulse - GET https://pypi.org/pypi/gstreamer-plugin-for-edge-impulse/json returns HTTP 404. This is not project-appropriate anyway (it is a Rust cdylib, not a Python package), included here only because it was checked as part of the standard distribution sweep.
  • RISE Python wheel builder: https://gitlab.com/api/v4/projects/56254198/packages/pypi/simple/gstreamer-plugin-for-edge-impulse/ redirects to the same PyPI 404.
  • Ubuntu 26.04 (resolute): Project graph SPARQL query for deb#BinaryPackage under plausible names (gstreamer plugin for edge impulse, python3-gstreamer-plugin-for-edge-impulse, libgstreamer-plugin-for-edge-impulse) in suite resolute, architecture riscv64: zero bindings. Live confirmation via packages.ubuntu.com: “Sorry, your search gave no results” for any architecture.
  • Arch Linux RISC-V (archriscv.felixc.at): no matching package listed.
  • Maven, npm, OCI: Not applicable to this project’s ecosystem (Rust Cargo cdylib); not checked.

What a user must do to get a working binary: There is no path to a working riscv64 binary today through any official or community channel. A user would need to: (1) build a riscv64 toolchain and Cargo cross-compilation plumbing analogous to the existing aarch64 path (none of which exists in this repo, see Section 5), (2) resolve the transitive edge-impulse-ffi-rs/Edge Impulse Inferencing SDK (C++) x86_64-fallback blocker (see Section 9), and (3) self-host the resulting binary, since upstream publishes no release binaries for any architecture whatsoever.

9. Dependencies

Dependency Role riscv64 build riscv64 test riscv64 release Community
GStreamer (core + audio/base/app/video bindings) Pipeline framework Packaged Ubuntu 26.04 riscv64 (resolute): libgstreamer1.0-0, -dev, gstreamer1.0-plugins-base/good No upstream riscv64 CI job Debian sid/testing, Ubuntu Noble, Alpine edge ship riscv64; no official upstream Linux binaries for any arch See project-reports/gstreamer.md
liborc (ORC) - GStreamer’s JIT SIMD compiler JIT backend for DSP (colorspace/audio conversion, compositing) Packaged Ubuntu 26.04 riscv64: liborc-0.4-0t64, -dev. Native RVV 1.0 backend exists (~2,700 lines, Samsung-authored, landed May 2025, first released in ORC 0.4.42, Jan 2026) No ORC riscv64 CI Released in ORC 0.4.42, packaged Debian/Alpine RVV API gated behind ORC_ENABLE_UNSTABLE_API; RV32 explicitly rejected; two open GStreamer core MRs (!11768, !11773) unmerged as of Jun 2026. See project-reports/gstreamer.md
Cairo (cairo-rs) Overlay-element rendering surface Packaged Ubuntu 26.04 riscv64: libcairo2, -dev; pure portable C, builds trivially No upstream CI (Anubis-blocked GitLab); Debian buildd confirms clean riscv64 builds Debian sid/Ubuntu/Arch RISC-V ship it Compositing engine pixman has no merged RVV backend (pixman-rvv.c, 3,272 lines, unmerged as of Jun 2026) -> scalar fallback, ~4-5x throughput deficit on vector-dominated compositing per project-reports/cairo.md
Pango / pangocairo Text shaping/layout Packaged Ubuntu 26.04 riscv64: libpango-1.0-0, -dev, libpangocairo-1.0-0; pure C, no arch-specific code found Not independently verified Packaged Ubuntu/Debian riscv64 No dedicated report
HarfBuzz (via Pango) OpenType/complex-script shaping Packaged Ubuntu 26.04 riscv64: libharfbuzz0b, -dev No upstream riscv64 CI job Debian/Ubuntu package it; no official upstream Linux binary for any arch See project-reports/harfbuzz.md
FreeType (via Pango/Cairo) Glyph rendering Packaged Ubuntu 26.04 riscv64: libfreetype6, -dev/libfreetype6-dev No upstream riscv64 CI Debian buildd/Ubuntu launchpad ship it See project-reports/freetype.md
rten/rten-tensor/rten-imageproc Tensor/ML inference engine backing the OCR feature Not a Debian/Ubuntu package (crates.io source only); compiles as generic scalar Rust via Tier-2 riscv64gc-unknown-linux-gnu target; no RVV kernel path No riscv64-specific CI/test evidence; GitHub issue search for “riscv” on robertknight/rten returned zero results Published to crates.io (source only, no arch-gated release) Documented SIMD backends: “AVX2, AVX-512, Arm Neon and WebAssembly SIMD” - RISC-V/RVV not mentioned anywhere
ocrs OCR detection/recognition models built on rten Not a Debian/Ubuntu package (crates.io only) No riscv64-specific issues found crates.io only Inherits rten’s scalar-fallback status entirely
edge-impulse-runner (Rust) (ffi/eim/ingestion features) FFI bridge to native ML inference Not a Debian/Ubuntu package; the ffi backend’s CMake logic has an unconditional non-aarch64/non-Apple fallback to x86_64 prebuilt TFLite libraries - a riscv64 build would silently misroute to wrong-ISA binaries rather than fail cleanly No CI at all for riscv64 (both GitHub Actions workflows run only on ubuntu-latest) Zero GitHub releases published for any architecture Critical structural gap. See project-reports/edge-impulse-runner-(rust).md (readiness: orange). Repository’s own issue tracker (5 issues total) has zero RISC-V mentions
Edge Impulse Inferencing SDK (C++) (vendored, pulled transitively via edge-impulse-ffi-rs) Core NN inference + DSP, used whenever ffi feature is built (i.e. the default build) Not packaged on any channel; source-vendored only No CI of any kind exists in the repo for any architecture Zero GitHub releases for any architecture No porting/riscv* target among 26 named ports; falls through to untested generic porting/posix path. Transitive deps (TFLite Micro, ruy, gemmlowp) individually are packaged for Ubuntu 26.04 riscv64, but the SDK itself has never been built/tested on riscv64 by anyone traceable. See project-reports/edge-impulse-inferencing-sdk-(c++).md (readiness: orange)
image (Rust crate) JPEG/PNG/GIF decode for video/image frame handling Not a Debian/Ubuntu package (crates.io only) Not independently researched crates.io Pure/mostly-pure-Rust codecs; expected to build via Tier-2 riscv64gc target with no known blocker, but not verified against upstream issue tracker
tokio Async runtime Not a Debian/Ubuntu package (crates.io only) Not independently researched crates.io Low risk - no SIMD/crypto/allocator surface exercised by this plugin’s feature set
regex (transitive) Pattern matching with SIMD-accelerated memchr internals Not a Debian/Ubuntu package (crates.io only) Not researched crates.io Not independently verified for this task

Deep-dive - critical dependency (edge-impulse-ffi-rs -> Edge Impulse Inferencing SDK C++): This is the single hard blocker in the whole dependency chain. The default feature set (ffi is in default = [...]) makes this the default build path for the plugin. It has an unconditional CMake fallback that would silently link x86_64 prebuilt TFLite binaries on a riscv64 build (rather than failing cleanly), has zero CI on any architecture, and has never been built or tested on riscv64 by anyone traceable in the available evidence.

OS-level C library dependencies (GStreamer, ORC, Cairo, Pango, HarfBuzz, FreeType): all present in Ubuntu 26.04 riscv64 (resolute) and do not block a native riscv64 build via apt.

Rust-crate dependencies (rten, ocrs, edge-impulse-runner, image, tokio, regex): not Ubuntu binary packages - they build from crates.io source via cargo/rustc, both of which are available in Ubuntu 26.04 riscv64. Their riscv64 viability is governed by Rust’s Tier-2 riscv64gc-unknown-linux-gnu target, not Debian packaging.

11. Known Bugs and Active Issues

ID Title Status Severity Notes
N/A No riscv64-related issue or PR exists N/A N/A The repository currently has 0 open issues and 3 open PRs (#27, #28, #48), none RISC-V related. The only 2 closed issues in the repo’s entire history (#8 “Make text overlay font size relative to output resolution”, closed 2025-07-29; #49 “Planned element: edgeimpulsedynamiccrop”, closed 2026-03-25) are unrelated to RISC-V, performance, or correctness on any architecture

Correctness bugs: None found, on any architecture. No issue or PR in this repository mentions NaN, floating-point problems, or riscv64 in any form.

Performance benchmarks: None found. No RISC-V (riscv64) benchmark data for this project exists in any searched source. The project’s README contains no performance numbers at all, RISC-V or otherwise - only a generic “Slow Inference” troubleshooting note suggesting input-resolution reduction, with no metrics.

12. Objections and Upstream Blockers

Stated objections: None found. No maintainer, issue, or PR has ever raised or rejected a RISC-V port proposal, because no such proposal has ever been made.

Technical blockers:

  1. No riscv64 Rust cross-compilation toolchain plumbing exists in the repo’s Docker/CI infrastructure (only aarch64 is documented; see Section 5).
  2. The default ffi feature’s transitive dependency chain (edge-impulse-runner -> edge-impulse-ffi-rs -> Edge Impulse Inferencing SDK C++) has an unconditional x86_64 fallback in its CMake build logic and zero riscv64 CI/testing of any kind (see Section 9).

Organizational blockers:

  1. Single-maintainer project (~96% of commits by one individual) with no formal contribution/governance process for proposing new platform ports (see Section 1).
  2. Not a RISE member and no RISE blog post, RFP, or funded work references this project or its immediate GStreamer/Edge Impulse dependency chain.
  3. Zero GitHub releases exist for any architecture, meaning there is no established release-engineering process this project could extend to riscv64 even if the technical blockers were resolved.

Acceptance probability: Data not available for a quantified estimate. Qualitatively: the absence of any tier policy, contribution process, or prior architecture-port precedent (only aarch64 exists, and even that has no dedicated CI) means a riscv64 port would be entering a project with no established process for accepting new platform work of any kind.

13. Readiness Assessment

  • Color: orange (plain “no upstream CI” base case - no upstream riscv64 CI exists, and no distribution channel of any kind (Ubuntu, PyPI, Arch RISC-V, or a GitHub release) ships this project for any architecture, so the distribution floor described in the color model does not apply - there is nothing to floor to)
  • Release provider: none
  • Justification: No upstream CI builds or tests riscv64 - all three GitHub Actions workflows run exclusively on ubuntu-latest (x86_64) with no matrix, no QEMU, and no cross-compile target (ci.yml). No release of any kind exists for any architecture (“There aren’t any releases here” - releases page), and no distribution channel (Ubuntu 26.04 riscv64, PyPI, Arch Linux RISC-V) packages this project under any plausible name, so the distribution floor cannot upgrade the grade to yellow.
  • Pending work that could change the grade: None identified. There are no open PRs, issues, or RISE involvement of any kind related to RISC-V for this project. The 3 currently open PRs (#27, #28, #48) are unrelated to RISC-V. A grade change would require new work to be initiated from zero: no in-flight activity exists to track.

14. Investment Analysis

RISE has not funded or performed any work on this project - no blog post, RFP, GitHub org (riseproject-dev, 25 repos checked), or wiki reference ties RISE to “GStreamer Plugin for Edge Impulse” in any way. All work items below are therefore fully unaddressed and not already covered by RISE.

14.1 Functional Enablement

  • Add riscv64gc-unknown-linux-gnu Rust target and toolchain plumbing to the Docker/CI infrastructure, mirroring the existing aarch64-unknown-linux-gnu path (Dockerfile stage, docker-compose.yml service, .cargo/config.toml linker entry, rustup target add).
  • Resolve the edge-impulse-ffi-rs/Edge Impulse Inferencing SDK (C++) unconditional x86_64 CMake fallback so a riscv64 build either produces a correct native build or fails cleanly instead of silently linking wrong-ISA binaries. This is a prerequisite for the plugin’s default ffi feature to function on riscv64 at all, and depends on upstream work in edge-impulse-runner and the Edge Impulse Inferencing SDK (C++) tracked separately (see project-reports/edge-impulse-runner-(rust).md and project-reports/edge-impulse-inferencing-sdk-(c++).md).
  • Verify the pure-Rust ocr feature path (rten/ocrs) builds cleanly under the Tier-2 riscv64gc target; no known blocker but unverified in practice.

14.2 Performance Optimization

  • Not in scope as a primary deliverable for this plugin itself (it is glue code, not an optimization-purpose library; see Section 13). Any performance gap traces to dependencies: rten/ocrs scalar-only OCR inference (no RVV backend documented anywhere in rten), and Cairo/pixman scalar-only compositing (unmerged pixman-rvv.c). Closing these requires upstream work in rten and pixman respectively, outside this project’s own control.

14.3 CI/CD Infrastructure

  • Add a riscv64 job to ci.yml (build + cargo test, matching the existing amd64 job’s feature-combination matrix). Given no native riscv64 GitHub-hosted runner exists in this project’s current CI, this would require either QEMU emulation or a RISE-provided riscv64 runner - neither is currently referenced anywhere in the repo’s workflows.

14.4 Ecosystem Enablement

Not applicable - this project does not have a significant dependent package ecosystem (npm, Maven, Python packages depending on it) that would require separate riscv64 enablement; Section 10 is omitted per the report’s scoping rule.

14.5 Summary Table

Area Work Item Effort (person-weeks) Owner Priority
Functional Add riscv64 Rust target + toolchain to Docker/CI, mirroring existing aarch64 path 1-2 Project maintainer or contributor High
Functional Fix edge-impulse-ffi-rs CMake x86_64 fallback for riscv64 (upstream dependency work, tracked separately) Data not available - depends on edge-impulse-runner/Edge Impulse Inferencing SDK (C++) scope, see those reports Edge Impulse Inc. / dependency maintainers Critical (blocks default feature set)
Functional Verify ocr (pure-Rust) feature builds on riscv64gc target 0.5-1 Project maintainer or contributor Medium
CI/CD Add riscv64 CI job (build + test) to ci.yml 1-2 (plus runner provisioning, unscoped) Project maintainer or contributor High
Performance Track upstream rten RVV backend and pixman-rvv.c merge status (no work owned by this project) N/A - upstream dependency work rten/pixman maintainers Low (not this project’s responsibility)

15. Updates

(No updates yet - initial report dated 2026-09-08.)

16. References