Raw dependency graph data (JSON)

libfdk-aac

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-09-30
Readiness: Yellow (clean-distro-build)
Scope: RISC-V (riscv64/linux) support status for libfdk-aac
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

libfdk-aac is an open-source packaging of the Fraunhofer IIS FDK AAC codec library, maintained by Martin Storsjo (mstorsjo) as a personal GitHub project. The underlying codec code originates from Fraunhofer-Gesellschaft zur Forderung der angewandten Forschung e.V. (Fraunhofer IIS, Erlangen, Germany), copyright 1995-2018. The repository periodically merges aosp/main from Google’s Android tree, which is how ongoing AOSP contributions reach it.

The library provides AAC-LC, HE-AACv1, and HE-AACv2 encode and decode. It is consumed by FFmpeg (optional AAC encoder), GStreamer (gstreamer-plugins-bad fdk-aac plugin), and Android (AOSP media stack).

Governance: None in the sense of a foundation, steering committee, or published policy. There is no MAINTAINERS, CODEOWNERS, PLATFORMS.md, or SUPPORT.md file, and no docs/ directory. The only governance-flavored artifact is an OWNERS file inherited from the AOSP mirror relationship (jmtrivi@google.com, plus an AOSP janitors include), which reflects Android/Google’s internal review process for the mirror, not an active governance body for the standalone GitHub project. Martin Storsjo is the sole gatekeeper for the public repository. documentation/ contains only two codec-spec PDFs (aacDecoder.pdf, aacEncoder.pdf), not platform or build documentation.

Corporate involvement: Effectively none beyond Fraunhofer’s original authorship and the AOSP merge relationship. All recent commits (CMake fixes, CI expansion to aarch64/macOS/MSVC-arm64 through 2025-2026) are authored by Martin Storsjo; his employer is not stated in the repository.

Community culture on new ports: Architecture-specific contributions have been accepted historically: s390x SIMD support was added via PR #159 (merged December 2023, approximately 500 lines of vector intrinsics), SPARC architecture detection via PR #178 (merged February 2025), and an AArch64 assembly optimization was attempted via PR #47 (closed, not merged). No one has filed a RISC-V optimization request, and no contributor is currently working on one. The maintainer has been actively broadening CI/architecture coverage (aarch64, MSVC arm64) through 2025-2026, which suggests general openness to new-architecture work via the standard PR process, but nothing RISC-V-specific has ever been proposed or discussed.

License: Fraunhofer FDK AAC Codec Library license (BSD-style redistribution terms per the repository NOTICE file). Source must be made available alongside binaries, the “Fraunhofer FDK AAC” name must be altered in modified versions, and the license explicitly states no patent license is granted for encoder/decoder use (patent licensing must be obtained separately via Via Licensing). This restricts which distributions can ship it in default/free sections, independent of architecture. The AOSP-inherited METADATA file also flags license_type: BY_EXCEPTION_ONLY, an AOSP-internal constraint on Google’s own use, not one governing public GitHub consumers.

RISE Project involvement: None. Confirmed by direct enumeration of the full RISE blog post history (35 posts via the WordPress sitemap, 2024-05-15 through 2026-09-28) with no post mentioning libfdk-aac, fdk-aac, or audio/AAC codec work; by absence from the RISE Python wheel builder package list (85 packages, none audio-codec related, and not applicable in any case since libfdk-aac is not a Python package); and by a zero-result GitHub search for libfdk-aac repositories in the riseproject-dev org. libfdk-aac and Fraunhofer do not appear on the RISE members list (Premier: Alibaba Damo, Google, MediaTek, NVIDIA, Qualcomm, Red Hat, SiFive, Tenstorrent; General: Akeana, Andes, Beijing ESWIN, Institute of Software Chinese Academy of Sciences, Canonical, Douyin/ByteDance, Microchip, NextSilicon, Quintauris, SpacemiT, ZTE).

2. Port History and Upstreaming Timeline

Date Event Source
2022-05-13 Single commit 3aabcb6abd adds RISC-V architecture detection to libFDK/include/FDK_archdef.h: 6 preprocessor #define lines selecting 16-bit lookup tables and 32x16 multiply preference (generic C path, no SIMD) commit 3aabcb6abd
2022-05-13 Commit authored by Zhang Ye (haocheng.zy@alibaba-inc.com), committed by Mao Han (han_mao@linux.alibaba.com), both at Alibaba commit 3aabcb6abd
2023-2026 Debian and Ubuntu package the codec as part of releases 2.0.2/2.0.3 and build riscv64 .deb packages using the generic C fallback path, current through Ubuntu 26.04 “resolute” Debian tracker, Ubuntu 26.04 resolute package page

No further RISC-V-related commit, issue, or PR exists anywhere in the repository history. Confirmed repo-wide by git grep -i riscv against a fresh clone (HEAD 2212850, 2026-09-26): exactly one match, the FDK_archdef.h line above. GitHub commit search (riscv repo:mstorsjo/fdk-aac) independently returns 0 results.

The port is fully upstream in the narrow sense that the 2022 architecture-detection commit is on master. It is not fully upstream in any engineering sense: there is no riscv64 CI, no SIMD optimization, and no RISC-V contributor actively maintaining the target. The 2022 commit is the entirety of RISC-V-specific work ever done on this project.

3. Upstream Support Tier

The project has no documented tier policy; no PLATFORMS.md, SUPPORT.md, or equivalent exists.

Evidence-based tier assignment:

Criterion amd64 arm64 riscv64
CI runner in upstream repo Yes (ubuntu-latest) Yes (ubuntu-24.04-arm, macOS universal, MSVC ARM64, llvm-mingw aarch64) No
SIMD/intrinsic optimization Yes (x86/ headers) Yes (arm/ headers and .cpp files) No (generic C fallback only)
Official GitHub release binaries No (no GitHub releases exist for any architecture; source-tag-only project) No No
Architecture characterization in FDK_archdef.h Yes (full) Yes (full) Yes (6 lines, 2022)
Distro binary packages Yes (all major distros) Yes (all major distros) Yes: Debian sid/trixie, Ubuntu 24.04 (noble) and 26.04 (resolute)

riscv64 compiles, links, and produces correct output via the generic C fallback path, and is packaged by at least two major distributions, but receives zero upstream CI coverage and zero SIMD acceleration. It sits below amd64/arm64 as a third-tier target: functionally supported through downstream build farms, not through any upstream engineering investment.

4. Technical Architecture and RISC-V-Specific Subsystems

libfdk-aac is a fixed-point AAC codec. Its performance-critical operations are: fixed-point multiply (32x16, 32x32, MLA), fixed-point multiply-accumulate for filter banks, complex multiply for the MDCT/FFT butterfly, arithmetic shift with saturation (scale), count-leading-zeros (clz) for normalization, absolute value (abs), and FFT scramble. Each has architecture-specific intrinsic implementations for ARM, MIPS, x86, and (partially) PowerPC. None has a RISC-V implementation.

Architecture-specific file inventory (verified by direct repository listing against HEAD 2212850):

Arch Header files (libFDK/include/<arch>/) Implementation files (libFDK/src/<arch>/) Total
x86 4 (abs_x86.h, clz_x86.h, fixmul_x86.h, fixpoint_math_x86.h) 0 4
arm 6 (clz_arm.h, scale_arm.h, fixmul_arm.h, fixmadd_arm.h, scramble_arm.h, cplx_mul_arm.h) 2 (scale_arm.cpp, fft_rad2_arm.cpp) 8
riscv 0 0 0

arm64 is the most complete port (intrinsic headers plus standalone .cpp implementations). x86 has intrinsic headers only. riscv64 has zero files of either kind, and no libFDK/include/riscv/ or libFDK/src/riscv/ directory exists.

Per-component status:

Component amd64 arm64 riscv64
abs intrinsics generic (no dedicated file) missing, generic C
clz intrinsics (__builtin_clz) intrinsics (__builtin_clz + ARM hint) missing, generic C loop
fixmul (32x16) intrinsics intrinsics missing, 64-bit C cast fallback
fixmadd generic (no dedicated file) intrinsics missing, generic C
fixpoint_math intrinsics generic (no dedicated file) missing, generic C
scale (shift+sat) generic (no dedicated file) intrinsics + dedicated .cpp missing, generic C
cplx_mul (FFT) generic (no dedicated file) intrinsics missing, generic C
FFT scramble generic (no dedicated file) intrinsics missing, generic C
Architecture tuning (FDK_archdef.h) full full scalar flags only (16-bit tables, MULT_32x16 preference); no SIMD/RVV

ISA extensions used on riscv64: none. The 2022 commit selects ARCH_PREFER_MULT_32x16, SINETABLE_16BIT, POW2COEFF_16BIT, LDCOEFF_16BIT, and WINDOWTABLE_16BIT, compile-time flags that cause the generic C code to use 16-bit ROM tables (reducing memory bandwidth) and prefer 32x16 over 32x32 multiplies. No Zba, Zbb, Zbc, or RVV (Vector) extension is referenced anywhere in the repository; confirmed by GitHub code search across the terms riscv, riscv64, rvv, vfloat32m1_t, zbb, rv64, __riscv_vector, and .S-extension files, all returning either zero matches or only the single FDK_archdef.h hit.

Functional correctness: the codec compiles and runs on riscv64 via the generic C path. Debian and Ubuntu build it successfully as part of routine package builds. No known correctness bugs exist for riscv64 (Section 11).

Performance: no benchmark data exists. Multiple targeted searches (GitHub issue/PR search, general web search across 2024-2026, academic search for RVV-intrinsic benchmark suites, and the RISE Project blog) found zero cycles/frame, fps, or throughput figures for libfdk-aac on any riscv64 silicon. General RVV benchmark literature exists (for example arXiv:2511.18867, “VecIntrinBench”, showing 2.98x-78x speedups on generic kernels such as memcpy and matrix operations), but none of it covers libfdk-aac or any AAC codec. Because every hot path (fixmul, cplx_mul/FFT, scale, clz, abs) runs generic scalar C on riscv64, the absence of benchmark data is consistent with the absence of any comparative performance engineering on this target. Data not available: no published cycles/frame figures for libfdk-aac on any riscv64 silicon.

5. Build System, Cross-Compilation, and Toolchain

libfdk-aac supports two build systems, autoconf/automake (configure.ac, Makefile.am, autogen.sh) and CMake (CMakeLists.txt, minimum version 3.5.1, >= 3.15 recommended). Neither has any riscv64-specific logic; a direct read of CMakeLists.txt at HEAD 2212850 confirms only 4 build options exist in total, none architecture-related:

option(BUILD_SHARED_LIBS "Build shared library" ON)
option(BUILD_PROGRAMS "Build extra utilities" OFF)
option(FDK_AAC_INSTALL_CMAKE_CONFIG_MODULE "Install CMake package configuration file" ON)
option(FDK_AAC_INSTALL_PKGCONFIG_MODULE "Install pkg-config .pc file" ON)

No CMAKE_SYSTEM_PROCESSOR branch, toolchain file, or -DUSE_X=OFF pattern exists anywhere in the repository.

Cross-compilation with autoconf:

autoreconf -fi
./configure \
  --host=riscv64-linux-gnu \
  --prefix=/usr \
  CC=riscv64-linux-gnu-gcc \
  CXX=riscv64-linux-gnu-g++ \
  AR=riscv64-linux-gnu-ar
make -j$(nproc)

Cross-compilation with CMake:

cmake -S . -B build \
  -DCMAKE_SYSTEM_NAME=Linux \
  -DCMAKE_SYSTEM_PROCESSOR=riscv64 \
  -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc \
  -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++ \
  -DCMAKE_INSTALL_PREFIX=/usr \
  -DBUILD_PROGRAMS=OFF
cmake --build build -j$(nproc)

No riscv64-specific CMake toolchain file exists in the repository, and none is required because the codebase has no CMAKE_SYSTEM_PROCESSOR branches to select.

Toolchain requirements: C++98 minimum. No minimum GCC or Clang version is documented anywhere in the repository for any architecture, riscv64 included; CI simply uses whatever compiler ships on ubuntu-latest / ubuntu-24.04-arm / macos-latest. GCC riscv64 support has been complete since GCC 7. On Debian/Ubuntu, gcc-riscv64-linux-gnu and g++-riscv64-linux-gnu are the standard cross-toolchain packages.

QEMU: no QEMU setup exists in upstream CI; a repo-wide case-insensitive search for “qemu” returns zero matches. To run the test suite on a cross-compiled riscv64 binary, qemu-riscv64-static with binfmt_misc registration would need to be set up manually; no such script is provided by the project.

Known build failures on riscv64: none reported. Zero issues and zero PRs in the tracker mention a riscv64 build failure.

6. Feature Coverage and Gap Analysis vs arm64 and amd64

Feature matrix:

Feature amd64 arm64 riscv64
AAC-LC decode Yes Yes Yes
HE-AACv1 decode (SBR) Yes Yes Yes
HE-AACv2 decode (PS) Yes Yes Yes
AAC-LC encode Yes Yes Yes
HE-AACv1 encode Yes Yes Yes
SIMD-accelerated fixed-point multiply Yes (x86 intrinsics) Yes (ARM intrinsics) No (generic C)
SIMD-accelerated FFT/MDCT Partial (complex mul) Partial (complex mul, scramble) No
SIMD-accelerated scale/shift No (generic C) Yes (ARM intrinsics) No
clz hardware acceleration Yes Yes No (generic C loop) [NEEDS VERIFICATION: GCC may still emit a riscv64 clz-class instruction from __builtin_clz without an explicit dispatch path; this was not independently tested against a Zbb-capable target]
16-bit ROM tables (memory optimization) Yes Yes Yes (set by the 2022 commit)
CI-validated correctness Yes Yes (macOS arm64, MSVC ARM64) No

Functional gaps: none. All codec features work on riscv64 via the generic C path.

Performance gaps: every SIMD-accelerated hot path falls back to generic C on riscv64. The magnitude of the regression relative to arm64/NEON or amd64 cannot be quantified because no benchmark data exists (Section 4).

Security hardening gaps: no riscv64-specific stack protection, shadow call stack, or CFI configuration exists in the build system. This is not a riscv64-specific regression: neither arm64 nor amd64 has project-level security hardening configuration either. Data not available: no published security audit of the riscv64 code path.

Floating-point / NaN semantics: the codec is fixed-point throughout; floating point is not used in the audio pipeline, so NaN-sensitivity issues structurally do not apply. Consistent with this, a GitHub issue search for “nan” in the repository returns zero results.

7. CI/CD Infrastructure

The upstream CI is a single file, .github/workflows/ci.yml (270 lines, triggered on: [push, pull_request] with no branch/label restriction). Verified by a fresh repository clone at HEAD 2212850 and a direct, complete read of the file, plus a case-insensitive grep -in riscv (zero matches) and a repository-wide git grep -in riscv (exactly one match, the FDK_archdef.h architecture-detection line, which is source code, not CI configuration). No .gitlab-ci.yml, Jenkinsfile, or .cirrus.yml exists anywhere in the repository.

The job matrix comprises 15 jobs: linux-autotools and linux-cmake (runners ubuntu-latest, ubuntu-24.04-arm), linux-sanitizers-gcc-autotools, linux-sanitizers-clang-cmake, macos-autotools, macos-cmake, mingw-cross-autotools, mingw-cross-cmake, msys-cmake, msvc-cmake, msvc-cmake-ninja-arm64, msvc-cmake-arm64, llvm-mingw (aarch64-w64-mingw32 target), linux-ffmpeg, check-archive, and check-install.

CI coverage by target:

Target CI present Runner Test depth
Linux x86_64 (autotools/cmake) Yes ubuntu-latest Build + sanitizers (ASan, UBSan)
Linux aarch64 Yes ubuntu-24.04-arm Build
macOS (universal) Yes macos-latest Build
Windows x86_64/ARM/ARM64 (MSVC) Yes windows-latest Build
MinGW x86_64 / aarch64 (llvm-mingw) Yes ubuntu-latest Cross-compile
FFmpeg integration (x86_64) Yes ubuntu-latest Build + functional
riscv64 (any form) No None None

RISE runners: not used. The RISE Project announced the RISE RISC-V Runners program on 2026-03-24 (free, native RISC-V CI on GitHub), with a six-weeks-in follow-up post on 2026-05-04. libfdk-aac does not use them; no evidence of any onboarding attempt exists.

QEMU emulation in CI: not present.

Adding riscv64 CI would require either a QEMU-based cross-compilation job (functional validation only) or onboarding to the RISE RISC-V Runners program (functional and, if native hardware is used, performance validation).

8. Distribution and Release Status

GitHub releases: none. Confirmed directly against https://github.com/mstorsjo/fdk-aac/releases (“There aren’t any releases here”). The project ships source only via git tags; no binary release assets of any kind exist for any architecture, so there is nothing to check for riscv64 release-asset naming.

Distribution packages:

Distribution Package name riscv64 available Version Notes
Ubuntu 26.04 “resolute” libfdk-aac2, libfdk-aac-dev Yes 2.0.2-3~ubuntu5 Verified via the architecture-specific download page: real binary, 462.1 kB package size, 780.0 kB installed size, live download link. Also ships for amd64, arm64, armhf, i386, ppc64el, s390x
Ubuntu 24.04 “noble” libfdk-aac2, libfdk-aac-dev Yes 2.0.2-3~ubuntu4 460.4 kB download
Debian sid/trixie libfdk-aac2t64, libfdk-aac-dev Yes 2.0.3-1 non-free/misc section
openSUSE Tumbleweed libfdk-aac (RPM) Reported available 2.0.3 [NEEDS VERIFICATION: single source, not independently confirmed via a direct openSUSE package-page fetch in this research pass]
Arch Linux RISC-V port libfdk-aac No 2.0.3-2 (x86_64 only, in Arch extra) Not carried in the archriscv community port; checked directly, homepage and search return no match
PyPI n/a n/a n/a https://pypi.org/pypi/libfdk-aac/json returns HTTP 404; no Python wrapper exists (fdk-aac is a C library, not a Python package)
Fedora fdk-aac Unknown Unknown Data not available: Koji build system was not accessible during research

All riscv64 binaries listed above are built from unmodified upstream source via the generic C fallback path; no riscv64-specific packaging patches were identified in any distribution’s packaging metadata.

To get a working riscv64 binary:

  • Ubuntu 24.04/26.04: apt install libfdk-aac-dev (universe repo)
  • Debian sid/trixie: apt install libfdk-aac-dev (non-free repo must be enabled)
  • Any other distribution: build from source using the cross-compilation steps in Section 5

No native riscv64 hardware validation by the Debian or Ubuntu build farms was confirmed in this research; distro builds are compile-only unless the packaging metadata states otherwise [NEEDS VERIFICATION: whether Debian’s rv-manda-03 buildd or Ubuntu’s build infrastructure runs the fdk-aac test suite under emulation as part of the package build].

9. Dependencies

libfdk-aac is intentionally self-contained: all codec subsystems (libAACdec, libAACenc, libSBR, libMpegTP, libFDK, libSYS) are bundled within the repository, and there is no external codec/DSP library dependency. The only dependencies are the toolchain and runtime C/C++ libraries.

Dependency Role riscv64 status
GCC (build-dependency, critical) Primary C/C++ compiler used by upstream CI and by Debian/Ubuntu package builds Complete riscv64 support since GCC 7; ships as the standard cross/native toolchain on all riscv64-supporting distributions, no blocking issues
CMake (build-dependency, critical) One of the two supported build systems (minimum 3.5.1) Host-tool only, not cross-compiled; ships for riscv64 as a standard build dependency on Debian/Ubuntu, no blocking issues
autoconf (build-dependency, optional) Part of the alternative autotools build path (autogen.sh / configure.ac) Host-tool only; ships for riscv64 as a standard build dependency, no blocking issues
automake (build-dependency, optional) Part of the alternative autotools build path Host-tool only; ships for riscv64 as a standard build dependency, no blocking issues
GCC (runtime-dependency, critical) Supplies the libstdc++ C++ runtime that all fdk-aac sources link against (the codebase is entirely C++) libstdc++ ships in every riscv64 GCC toolchain and every riscv64 distribution runtime image; no blocking issues
glibc (runtime-dependency, critical) C library; also supplies libm, used only for sin() at fixed-point table-generation time (not in the real-time encode/decode path) glibc’s riscv64 port is first-class and ships as a standard package on Debian, Ubuntu, Arch RISC-V, and Fedora; see the dedicated glibc RISC-V readiness report for detail; no blocking issues

No JIT backend, external SIMD library, numerics library, crypto library, compression library, or external memory allocator exists anywhere in fdk-aac’s dependency chain. The per-architecture intrinsic headers under libFDK/include/{x86,arm,mips,ppc}/ are bundled, optional, in-tree source files, not external dependencies; none has a riscv64 equivalent, which is the project’s actual performance gap (Section 4), not a dependency-availability problem. Because every listed dependency (GCC, CMake, autoconf, automake, glibc) already has mature, first-class riscv64 support across major distributions, the dependency chain is not a blocker to libfdk-aac’s riscv64 readiness; the gap is entirely in upstream’s own lack of riscv64 CI and SIMD optimization.

11. Known Bugs and Active Issues

RISC-V-specific issues: none. GitHub issue and PR searches for “riscv” and “riscv64” in repo:mstorsjo/fdk-aac return zero results, confirmed independently across multiple passes this session (GitHub MCP search tools and direct github.com web searches, the latter used as a substitute during periods when the MCP search tool was rate-limited). A search for “nan” also returns zero results, consistent with the codec being entirely fixed-point.

Open issue potentially relevant to any future riscv64 SIMD work:

ID Title Status Severity Notes
#166 Are CPU specific optimizations with -mcpu supported? Open (filed 2024-02-08) Medium Reports a performance regression on aarch64 caused by -mcpu flags disabling conditional SIMD paths. Not RISC-V-specific, but a cautionary precedent for any future riscv64 SIMD/RVV PR that gates code on compiler -march/-mcpu flags

Open PRs (none RISC-V related):

PR Title Status Notes
#168 Visual Studio / NuGet packaging Open (Feb 2024) Unrelated to riscv64
#139 Meson build system Open (Oct 2021) Would affect all targets if merged; stalled
#40 Whitespace cleanup Open (Jan 2016) Stale
#37 Typo fix Open (Jan 2016) Stale

No correctness bugs are open for riscv64, and no riscv64 build failure has ever been reported in the issue tracker.

12. Objections and Upstream Blockers

Technical blockers: none. The generic C path compiles and runs correctly on riscv64 today. A contributor could submit a PR adding libFDK/include/riscv/ headers with RVV intrinsics without any known upstream technical obstacle.

Stated objections: none on record. No maintainer has expressed opposition to riscv64 work in any issue, PR, or commit. The project’s acceptance of s390x SIMD (PR #159) and SPARC detection (PR #178) establishes precedent for accepting new-architecture contributions.

Organizational blockers: the project is a single-maintainer personal repository (Martin Storsjo). PR review throughput is low: two of the four currently open PRs date to January 2016 and remain unmerged nearly a decade later. A RISC-V optimization PR could sit in review for an extended period. The maintainer has expressed no RISC-V-specific interest anywhere in the available record.

License blocker: the Fraunhofer FDK AAC license’s no-patent-license term restricts which distributors can ship riscv64 (or any-architecture) binaries in default/free repository sections, but it does not block upstream code contribution or building from source, independent of architecture. For internal chip-company use, or for Debian’s non-free channel, the license is workable as-is.

Acceptance probability: high for a minimal PR (riscv64 CI job, QEMU-based). Medium-to-low for a full RVV optimization PR, given the low maintainer review bandwidth evidenced by the multi-year-stale open PRs and the absence of a second reviewer.

13. Readiness Assessment

  • Color: Yellow (clean-distro-build)
  • Release provider: distro

Justification: Upstream mstorsjo/fdk-aac has no riscv64 CI at all: its only CI file, .github/workflows/ci.yml, has zero riscv/riscv64 matches across its 15 jobs (confirmed by direct file read and repo-wide git grep -i riscv, which found only the unrelated #elif defined(__riscv) preprocessor stub in libFDK/include/FDK_archdef.h), and it publishes no GitHub Releases or binary assets of any kind. Per the distribution floor used by this grading methodology, this would default to orange, except that a Linux distribution builds it from unmodified upstream source: Ubuntu 26.04 “resolute” ships libfdk-aac2/libfdk-aac-dev 2.0.2-3~ubuntu5 for riscv64 as a real downloadable binary (packages.ubuntu.com/resolute/riscv64/libfdk-aac2, 462.1 kB, verified via direct fetch of the architecture-specific download page), built via the project’s generic C fallback path with no riscv64-specific packaging patches identified. This satisfies the “clean-distro-build” yellow floor rather than the patched/uncertain “downstream-only” orange tier. libfdk-aac is a fixed-point AAC codec whose value proposition is standards-compliant encode/decode correctness (it is the Fraunhofer reference implementation), not RISC-V-specific speed superiority over a simpler alternative, so this is not graded as an optimization-purpose project, notwithstanding that its only RISC-V-aware code (FDK_archdef.h lines 225-231, a 2022 Alibaba commit) is limited to generic table-format/multiply-preference flags with no RVV/SIMD implementation.

Pending work that could change the grade: no open PRs or issues target riscv64 (the repository’s 4 open PRs, #37, #40, #139, #168, and its issues are all unrelated; the one semi-relevant open issue, #166, on -mcpu flags disabling conditional SIMD on aarch64, is a cautionary precedent for any future riscv64 SIMD work, not active riscv64 work itself). No RISE Project involvement was found (absent from the RISE blog, the wheel builder list, and riseproject-dev org repos/issues). No upstream riscv64 CI job or tracking issue exists to escalate this beyond yellow; adding a QEMU-based riscv64 CI job to ci.yml, or onboarding to RISE RISC-V Runners, would be the concrete next step toward blue.

14. Investment Analysis

RISE has done nothing for libfdk-aac (Section 1). All work scoped below is net-new; none of it is already covered by RISE funding or infrastructure.

14.1 Functional Enablement

Functional correctness on riscv64 is already complete via the 2022 Alibaba commit and is already distributed by Ubuntu and Debian. No additional functional-enablement work is required.

14.2 Performance Optimization

The performance gap relative to arm64 is the primary technical deficiency, and it is unquantified because no benchmark exists. Priority order for hot-path optimization, based on the codec’s operation profile (Section 4):

  1. fixmul.h, 32x16 and 32x32 fixed-point multiply, called in every filter bank and spectral-coefficient operation. An RVV widening multiply (vwmul) would directly accelerate this.
  2. cplx_mul.h, complex multiply for MDCT/FFT butterflies. RVV interleaved vector arithmetic.
  3. scale.h, arithmetic right shift with saturation. Zbb extension (sra + max) or RVV.
  4. clz.h, count leading zeros. Zbb’s single-instruction clz.
  5. abs.h, absolute value. Zbb’s single-instruction abs.
  6. FFT index scramble (arm-equivalent scramble_arm.h). Lower priority.

The analogous s390x SIMD PR (#159) was approximately 500 lines. A complete RVV optimization covering items 1-5 is estimated at 600-900 lines across 5 header files, with no prior art in the upstream repository to build from.

14.3 CI/CD Infrastructure

A QEMU-based riscv64 CI job can be added to .github/workflows/ci.yml with a qemu-user-static cross-compile step; this validates correctness only, not performance. A native riscv64 CI job using the RISE RISC-V Runners program (announced 2026-03-24, six-weeks-in update posted 2026-05-04) would provide both correctness and performance signal, but libfdk-aac has not onboarded to it and no evidence of an attempt exists.

14.4 Ecosystem Enablement

libfdk-aac has no language-level package ecosystem: no PyPI package (confirmed 404), no npm package, no Maven artifact. Its only distribution surface is system .deb/.rpm packages, which Debian and Ubuntu already provide for riscv64 (Section 8). No ecosystem-enablement work is required beyond what distributions already do.

14.5 Summary Table

Area Work Item Effort (person-weeks) Owner Priority
Functional None; generic C path is correct and already ships in Debian/Ubuntu riscv64 0 n/a n/a
Performance Implement libFDK/include/riscv/ headers with RVV intrinsics for fixmul, cplx_mul, scale, clz, abs 4-6 Compiler/codec engineer with RVV experience High: required for real-time use on embedded or mobile riscv64
Performance Benchmark vs arm64 NEON to quantify the gap before and after optimization 1 Performance engineer High: justifies and validates the optimization work
CI/CD Add a riscv64 QEMU cross-compile job to upstream ci.yml and submit as a PR 0.5 Any engineer Medium
CI/CD Onboard to RISE RISC-V Runners for native-hardware CI 1 DevOps / RISE liaison Medium
Distro packaging Request Arch Linux RISC-V (archriscv) package addition 0.25 Any contributor Low
Upstream PR Submit the performance-optimization PR and shepherd it through review given low maintainer bandwidth 1-2 (review cycles) Senior engineer High: blocks upstream acceptance

Total estimated effort: 7.75-10.75 person-weeks for a complete, upstreamed, CI-validated RVV optimization.

15. References