filament

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-08-14
Scope: RISC-V (riscv64/linux) support status for Google Filament
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

Google Filament is a physically based rendering (PBR) engine written in C++, targeting Android (arm64-v8a primary), Linux, macOS, iOS, Windows, and WebAssembly. Its primary use case is Android application rendering. The engine provides OpenGL ES, Vulkan, and Metal backends; a material compiler (matc/filamat); and supporting tools for mesh optimization, texture compression, and asset pipelines.

Filament is released under the Apache License 2.0. It is a Google-internal open-source project hosted at github.com/google/filament and explicitly described as “not an officially supported Google product.” Governance is informal and single-vendor: Google employees own all CI/CD configuration, the .github/ workflow directory is restricted to maintainers, and all contributors must sign the Google CLA. No MAINTAINERS, OWNERS, or CODEOWNERS file exists in the repository.

The five highest-commit contributors are all Google employees or ex-Google employees: pixelflinger/Mathias Agopian (2,621 commits, @google), prideout/Philip Rideout (1,754 commits), bejado (1,210 commits, @google), romainguy/Romain Guy (788 commits), and poweifeng/Powei Feng (783 commits). There is no evidence of non-Google corporate contributors with significant commit counts.

Filament is not a member of, tracked by, or funded by the RISE Project. The RISE working groups cover compilers, toolchains, system libraries, kernel, language runtimes, AI/ML, and distro integration; graphics rendering engines are outside RISE’s current scope. A scan of all 33 RISE blog posts (2024-05 through 2026-08) found zero references to Filament.

There is no documented tier policy for platform support and no stated criteria for accepting new architecture ports. Given the single-vendor governance model, a community-contributed RISC-V port would require active Google maintainer buy-in, a signed Google CLA, and ongoing CI coverage funded by Google infrastructure. There is no precedent for external contributors adding new architecture support to this project.

The project is actively developed: v1.75.0 was released 2026-08-04, with 10+ PRs merged per week as of August 2026.

2. Port History and Upstreaming Timeline

No RISC-V port has ever been started for google/filament.

Date Event Source
- Zero issues filed for riscv or riscv64 GitHub Issues API
- Zero PRs filed for riscv or riscv64 GitHub PR search
- Zero commits mentioning riscv or riscv64 GitHub Commits API
- No riscv64 toolchain cmake in build/ github.com/google/filament build/ tree
- No riscv64 CI job in any of 14 workflow files github.com/google/filament .github/workflows/

There are no key contributors to cite. There is no upstreaming timeline because no port effort has been initiated by any party.

3. Upstream Support Tier

Filament has no formal tier policy document. Support tiers are inferred from CI coverage, release binary availability, and BUILDING.md documentation.

Tier criterion amd64 (x86_64) arm64 (aarch64) riscv64
CI build job Yes (ubuntu-24.04) Yes (arm-ubuntu-24.04-16core) No
CI test job Yes Yes No
Release binary (linux) Yes (filament-v1.75.0-linux.tgz) Yes (filament-v1.75.0-arm-linux.tgz) No
Android ABI x86, x86_64 arm64-v8a, armeabi-v7a No
Documented in BUILDING.md Yes Yes (primary target) No
CMake toolchain file Yes (implicit native) Yes (android aarch64) No

BUILDING.md states: “our implementation is optimized first and foremost for arm64-v8a.” riscv64 is absent from every layer of the official support surface.

4. Technical Architecture and RISC-V-Specific Subsystems

Filament’s own source tree contains zero RISC-V-specific code. All three RISC-V files in the repository are in vendored third-party libraries.

4.1 Architecture-Specific Components

Color Grading SIMD (filament native)

ColorGrading.cpp has five #if defined(__ARM_NEON) guards. ColorGradingNeon.h provides 432 lines of hand-optimized NEON tone-mapping loops. No __riscv equivalent exists anywhere in the filament source tree. On riscv64, the scalar C fallback path runs.

bluegl OpenGL trampolines (filament native)

BlueGLCoreLinuxAArch64Impl.S and BlueGLCoreDarwinAArch64Impl.S exist for aarch64. BlueGLCoreLinuxImpl.S exists for x86. No riscv64 .S file or equivalent exists. On riscv64 Linux, bluegl falls back to the generic runtime-link path [NEEDS VERIFICATION - runtime fallback behavior not directly tested].

math/fast.h SIMD paths (filament native)

libs/math/include/math/fast.h uses #ifdef __ARM_NEON and #ifdef __aarch64__ exclusively. No #ifdef __riscv guards exist anywhere in filament’s math library.

ASTC texture compression RVV (third_party/tinyexr/deps/astcenc)

astcenc_vecmathlib_rvv_n.h (1,230 lines, Arm Limited / Olaf Bernstein, Apache-2.0, copyright 2026) implements N-wide 32-bit float/int/mask SIMD vector math for ASTC texture encoding using RVV. Dispatch guard: #elif defined(__riscv_v_fixed_vlen) && (__riscv_v_fixed_vlen == 256). This requires VLEN=256 fixed-length mode. Scalable-vector mode (the default on most current RISC-V hardware) is not supported. RVV intrinsics used include __riscv_vle32_v_f32m1, __riscv_vfmv_v_f_f32m1, __riscv_vsetvlmax_e32m1, vbool32_t, and many others. ISA extension required: V extension (RVV), VLEN=256 fixed-length.

libdeflate LZ77 matchfinder RVV (third_party/tinyexr/deps/libdeflate)

lib/riscv/matchfinder_impl.h (97 lines, Eric Biggers, MIT, 2024) implements matchfinder_init_rvv() and matchfinder_rebase_rvv() for bulk LZ77 hash table initialization using scalable RVV (m8 LMUL, no fixed-length constraint). Guard: defined(ARCH_RISCV) && defined(__riscv_vector). ISA extension required: V extension (RVV), scalable.

Abseil stack unwinder (third_party/abseil)

absl/debugging/internal/stacktrace_riscv-inl.inc (203 lines, Google Apache-2.0) implements frame-pointer-based stack walking for riscv64 Linux using ucontext_t / mcontext_t via uc_mcontext.__gregs[REG_PC] and REG_S0. No ISA extensions required; pure base RV64I. Complete implementation.

4.2 Component Comparison Table

Component Owner amd64 arm64 riscv64
ColorGrading SIMD tone-mapping filament native scalar hand-tuned NEON (432 lines) missing (scalar fallback)
bluegl OpenGL trampolines filament native .S asm .S asm (2 files) missing (generic fallback)
math/fast.h SIMD filament native scalar NEON + aarch64 intrinsics missing (scalar)
ASTC texture compression SIMD third_party/astcenc SSE-4 (1,351 lines), AVX2 (1,229 lines) NEON (1,130 lines), SVE (1,106 lines) RVV VLEN=256 only (1,230 lines)
libdeflate matchfinder third_party/libdeflate x86 RVV equiv. (123 lines) arm (79 lines) RVV scalable (97 lines)
Abseil stack unwinder third_party/abseil x86 (406 lines) aarch64 (289 lines) riscv (203 lines)
GPU backends (Vulkan, OpenGL, Metal) filament native generic C/GLSL generic C/GLSL no riscv64 path
Shader compiler (filamat/matc) filament native generic generic no riscv64 path
Job system filament native generic C++ atomics generic C++ atomics no riscv64 path

ISA extensions used by third-party RVV code: V extension only. No Zba, Zbb, Zbc, or other extensions referenced anywhere in the repository.

No hand-written .S assembly files exist for riscv64 in either filament or its vendored dependencies (filament-level; individual deps not exhaustively checked).

5. Build System, Cross-Compilation, and Toolchain

5.1 Requirements (from BUILDING.md)

  • CMake >= 3.22.1
  • Clang >= 17.0 (required; MSVC supported on Windows only; libc++ required on Linux/macOS, NOT libstdc++)
  • libc++-17-dev and libc++abi-17-dev
  • ninja >= 1.10

5.2 Native Linux Build (amd64 only, documented) 

mkdir out/cmake-release && cd out/cmake-release
CC=/usr/bin/clang CXX=/usr/bin/clang++ CXXFLAGS=-stdlib=libc++ \
  cmake -G Ninja -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX=../release/filament ../..
ninja

5.3 Cross-Compilation Model

Filament uses a two-phase cross-compilation model because host tools (matc, resgen, cmgen, filamesh) must run on the build host to generate shaders and resources during the target build:

  • Phase 1: Build host tools on x86_64, export via FILAMENT_EXPORT_PREBUILT_EXECUTABLES_DIR
  • Phase 2: Cross-compile target with -DCMAKE_TOOLCHAIN_FILE=<toolchain> -DIMPORT_EXECUTABLES_DIR=<host-tools-dir>

CMake key variables: IMPORT_EXECUTABLES_DIR, FILAMENT_EXPORT_PREBUILT_EXECUTABLES_DIR, FILAMENT_IMPORT_PREBUILT_EXECUTABLES_DIR.

Existing toolchain cmake files in build/:

  • toolchain-aarch64-linux-android.cmake
  • toolchain-arm7-linux-android.cmake
  • toolchain-x86-linux-android.cmake
  • toolchain-x86_64-linux-android.cmake

No build/toolchain-riscv64-linux-gnu.cmake or any equivalent exists.

5.4 riscv64 Build Status

No riscv64 build has been attempted or documented. To attempt a riscv64 build, a contributor would need to:

  1. Author build/toolchain-riscv64-linux-gnu.cmake modeled after the aarch64-android toolchain
  2. Supply Clang >= 17 cross-compiler targeting riscv64-linux-gnu with -stdlib=libc++
  3. Build host tools phase for x86_64 and export matc, resgen, cmgen, filamesh
  4. Cross-compile with the new toolchain file and IMPORT_EXECUTABLES_DIR pointing to phase 1 output
  5. Add #ifdef __riscv guards or stub paths in CMakeLists.txt where arch-specific paths are currently amd64/arm64 only

No QEMU-based build or test is configured anywhere in the project. None of the above steps are upstreamed or tested.

5.5 Known Build Failures

No riscv64 build failures are documented because no riscv64 build has been attempted. The Abseil dependency requires -latomic with some cross-toolchains on riscv64 (see Abseil issue #1702).

6. Feature Coverage and Gap Analysis vs arm64 and amd64

6.1 Functional Gaps

Feature amd64 arm64 riscv64 Notes
OpenGL ES backend Yes Yes Unknown - untested bluegl trampoline missing
Vulkan backend Yes Yes Unknown - untested No riscv64 Vulkan driver known
Metal backend N/A macOS/iOS N/A Apple platforms only
ASTC texture compression Yes (SSE4/AVX2) Yes (NEON/SVE) Partial - VLEN=256 only Fixed-length RVV constraint
WebP texture decode Yes Yes Scalar only No RVV path in libwebp
zstd compression Yes Yes Yes (scalar) RVV optimization PRs stalled upstream
Android ABI target x86/x86_64 arm64-v8a (primary) No Not listed in BUILDING.md
filamat material compiler Yes Yes Untested Host tool; may work if built natively
ColorGrading SIMD scalar NEON hand-tuned Missing (scalar) Performance gap, not correctness gap
bluegl trampolines .S asm .S asm Missing (fallback) May affect OpenGL load latency

6.2 Performance Gaps

  • ColorGrading/tone mapping: Arm64 has 432 lines of hand-written NEON. riscv64 runs the scalar path. Throughput difference is unquantified (no published benchmarks on any RISC-V platform).
  • ASTC encoding: RVV path requires __riscv_v_fixed_vlen == 256. Hardware that does not set this fixed-length mode runs the scalar fallback. SpacemiT X60 (VLEN=256) would hit the RVV path; other hardware may not.
  • WebP decode: Scalar only on riscv64. Expected 3-5x throughput gap vs x86 SSE2/AVX2 [NEEDS VERIFICATION - figure from libwebp report, not from filament-specific benchmark].
  • zstd: RVV optimization PRs stalled; scalar-only in practice.
  • Basis Universal (basisu) texture transcoding: SSE-4.1 encoder/transcoder disabled on riscv64 at compile time; no RVV path exists upstream; scalar fallback only.

6.3 Floating-Point and NaN Semantics

Issue #10280 (open, 2026-08-09): NaN propagation in lit shading model from half-precision NoV underflow on Metal fastMathEnabled=YES. This is Metal/macOS-specific and not a riscv64 issue. No riscv64-specific floating-point correctness issues have been filed because riscv64 has never been tested.

6.4 Security Hardening

Data not available: no riscv64-specific security hardening analysis (shadow call stack, pointer authentication equivalent, CFI) was found in any upstream issue or document.

7. CI/CD Infrastructure

Direct inspection of all 14 workflow files confirms zero occurrences of “riscv” in any CI configuration file.

CI dimension amd64 arm64 riscv64
Build job Yes (ubuntu-24.04) Yes (arm-ubuntu-24.04-16core) No
Test job Yes Yes No
QEMU emulation No No No
RISE runners No No No
Hardware runner GitHub-hosted GitHub-hosted None
Release blocking Yes Yes N/A

Workflow files inspected: auto-close-restricted-paths.yml, cocopods-deploy.yml, npm-deploy.yml, postsubmit-main.yml, postsubmit.yml, presubmit.yml, release.yml, status-android.yml, status-ios.yml, status-linux.yml, status-macos.yml, status-web.yml, status-windows.yml, verify-release-notes.yml. None contain “riscv”.

The .github/ directory is explicitly restricted to Google maintainers. External contributors cannot add CI jobs. Adding riscv64 CI requires a Google maintainer to configure and fund the runner infrastructure.

8. Distribution and Release Status

8.1 Official Release Binaries (v1.75.0, 2026-08-04)

Release assets (12 total) target: android, android-native, arm-linux, ios, linux, mac, web, windows. No asset filename contains “riscv64” or “riscv”.

Platform Binary available riscv64 variant
Linux (x86_64) Yes (filament-v1.75.0-linux.tgz) No
Linux (arm64) Yes (filament-v1.75.0-arm-linux.tgz) No
Android arm64-v8a Yes No riscv64 ABI
macOS Yes N/A
Windows Yes N/A
WebAssembly Yes N/A

8.2 Package Managers and Distributions

Channel Status Notes
PyPI pypi.org/project/filament Not google/filament filament-0.0.2.tar.gz is an unrelated stub package; source-only, no wheels
Debian Different project Debian “filament” package (v1.9.25+dfsg3) is the KDE thumbnail plugin, not google/filament
Ubuntu noble Not packaged packages.ubuntu.com search returned no results
Arch Linux RISC-V (archriscv.felixc.at) Not listed Not present in Arch RISC-V port
RISE wheel builder Not listed google/filament not among the ~80 packages built

8.3 Getting a Working Binary on riscv64

A user wanting to run google/filament on riscv64 must build from source with no upstream guidance. Required steps: author a riscv64 CMake toolchain file, perform the two-phase cross-compilation, and resolve any build failures independently. No documentation, no prebuilt toolchain, no CI results exist to reference.

9. Dependencies

9.1 Summary Table

Dependency Role riscv64 build riscv64 test riscv64 release Blocking issues
SPIRV-Tools SPIR-V optimizer/validator (Vulkan/filamat) Pass (pure C++) No riscv64 CI Source only None found
glslang GLSL/HLSL-to-SPIR-V compiler (filamat) Pass (pure C++) No riscv64 CI Source only None found
SPIRV-Cross SPIR-V decompiler (Vulkan backend) Pass (pure C++) No riscv64 CI Source only None found
SPIRV-Headers SPIR-V grammar headers (header-only) N/A N/A N/A None
abseil-cpp Base library (containers, strings, synchronization) Partial (-latomic workaround required) Partial (QEMU/RISE); 2 SEGFAULTs open Debian/Ubuntu package ships Issues #1702, #2002 open; see project-reports/abseil-cpp.md
zstd Compressed material archives, mesh data Pass (QEMU CI since 2025) QEMU CI on PR branch only Debian/Ubuntu packages ship RVV PRs stalled (#4546, #4471, #4557, #4622, #4668); see project-reports/zstd.md
Dawn WebGPU backend (optional) No riscv64 CI found Unknown Source only Effectively untested on riscv64
Vulkan Memory Allocator GPU heap allocator for Vulkan (header-only) Pass No riscv64 CI N/A None found
Basis Universal (basisu) GPU texture transcoding (KTX2) Partial (BASISU_SSE=OFF required) No riscv64 CI Source only SSE-4.1 paths disabled; scalar fallback only; no RVV path
draco Mesh geometry compression (glTF loader) Pass (pure C++) No upstream riscv64 CI; Debian passes Debian/Ubuntu ship riscv64 None blocking; see project-reports/draco.md
libwebp WebP texture decode (optional) Pass (scalar) No riscv64 CI Debian/Ubuntu ship (libwebp7) Scalar only; ~3-5x throughput gap; see project-reports/libwebp.md
meshoptimizer Mesh optimization (tools) Pass (SSE2/NEON compiled out) No riscv64 CI Source only None found
smol-v SPIR-V size optimizer (material bundles) Pass (pure C++) No riscv64 CI N/A None
imgui Debug UI overlay (filagui) Pass (pure C++) No riscv64 CI N/A None found
google/benchmark Microbenchmarking Pass No upstream riscv64 CI Source only CPU frequency estimation uses busy-loop fallback; see project-reports/benchmark.md
libpng PNG decode (imageio/tools) Pass (scalar; NEON skipped) No riscv64 CI Debian/Ubuntu ship None found; no RVV path exists
SDL2 (vendored 2.0.8) Window/input for desktop samples (optional) Stale - predates riscv64 maturity Not tested N/A SDL 2.0.8 (2018) has no riscv64 joystick/audio drivers; use system SDL2 >= 2.26
perfetto CPU/GPU tracing (dev-mode) Pass (issue #936 fixed) N/A N/A Issue #936 (closed): riscv64 stack-unwind bug fixed
tinyexr EXR image read (tools) Pass (header-only) N/A N/A None
cgltf glTF 2.0 parser (header-only) Pass N/A N/A None
stb Image I/O (header-only) Pass N/A N/A None
libassimp 3D asset importer (tools) Pass No riscv64 CI Debian/Ubuntu ship None found
googletest Unit test framework Pass (test #3756 cosmetic fail) Issue #3756 open (riscv64 not officially supported) Debian/Ubuntu ship Cosmetic test failure; no functional impact

9.2 Critical Dependency Deep-Dive

abseil-cpp: Two open riscv64 issues. Issue #1702: -latomic link failure with some cross-toolchains; workaround is to add -latomic explicitly. Issue #2002: absl_hashtablez_sampler_test and absl_cordz_sample_token_test SEGFAULT on Debian riscv64 with GCC 15; no maintainer response. CRC32C hardware acceleration blocked pending Google riscv64 hardware (PR #1986). See project-reports/abseil-cpp.md.

Basis Universal (basisu): No riscv64 issues filed upstream. The SSE-4.1 encoder paths (basisu_kernels_sse.cpp, basisu_bc15_spmd_sse.cpp) are compiled out on riscv64 when built with BASISU_SSE=OFF. No NEON or RVV acceleration path exists. Texture transcoding throughput is scalar-only on riscv64. This affects all KTX2 texture workflows in filament.

zstd: Builds and runs on riscv64 with QEMU CI. Five RVV optimization PRs are stalled: unaligned access (#4546), RVV XXH3 (#4471), fast decode path (#4557), Huffman 4-way (#4622), prefetch (#4668). Scalar performance is functional. See project-reports/zstd.md.

libwebp: No RVV DSP path. All DSP paths (loop filter, transform, predictor, encode) are scalar on riscv64. See project-reports/libwebp.md.

11. Known Bugs and Active Issues

No riscv64-specific bugs exist in google/filament because riscv64 has never been tested. The following issues are relevant to correctness and performance on the platforms that are tested:

ID Title Status Severity Notes
#10280 lit shading model produces black faces on Metal (NaN propagation from half-precision NoV) Open Medium Metal/macOS only; fastMathEnabled=YES + float16 underflow; not riscv64
#10332 JobSystem: lock-free redesign and optimizations Open Perf macOS: 3.71ms->0.10ms (36x); Android ARM64: 4.02ms->0.63ms (6.4x); no riscv64 data
#9831 Move some presubmit tests to arm linux build Merged 2026-03-24 Low Added arm-linux CI; no riscv64 runner added

No correctness bugs specific to riscv64 are documented. The absence of reported bugs reflects the absence of any riscv64 testing, not the absence of bugs.

12. Objections and Upstream Blockers

Organizational blockers:

  • Single-vendor governance. CI and release infrastructure is restricted to Google maintainers. External contributors cannot add CI jobs or release targets without Google approval.
  • No stated interest from Google in riscv64 support. No issue, PR, comment, or blog post from any Google employee mentions RISC-V in the context of filament.
  • Google CLA required for all contributions. This is a standard barrier but must be cleared by any contributing organization.
  • No RISE involvement. Filament is outside RISE’s current working group scope; no funded effort exists.

Technical blockers:

  • Two-phase cross-compilation requires authoring a riscv64 CMake toolchain file (no template exists).
  • Abseil-cpp requires -latomic workaround and has two open riscv64 test failures.
  • ASTC RVV path requires __riscv_v_fixed_vlen == 256; hardware not setting this fixed-length mode falls back to scalar. Current SpacemiT X60 uses VLEN=256 and would activate the RVV path [NEEDS VERIFICATION - hardware behavior not directly confirmed in filament context].
  • No Vulkan driver for riscv64 Linux is known to exist; the Vulkan backend cannot be tested.
  • Basisu has no RVV path; texture transcoding will be scalar-only regardless of hardware.

Acceptance probability: Low in the near term. The project has zero community governance, no external precedent for architecture additions, and no stated roadmap for RISC-V. A port landing upstream requires sustained Google maintainer engagement. The Android riscv64 ABI trajectory (Google announced riscv64 Android support in 2023 for API level 35+) is the most plausible forcing function for Google to add riscv64 to filament’s Android build, but no public timeline exists for this.

13. Investment Analysis

No prior RISE funding or work has been done on filament. All items below are net-new.

13.1 Functional Enablement

The minimum viable work to build filament on riscv64 Linux:

  • Author build/toolchain-riscv64-linux-gnu.cmake (2 person-days; copy aarch64-android toolchain and adjust)
  • Perform two-phase cross-compilation, debug link failures (abseil -latomic, any additional failures) (3-5 person-days)
  • Verify OpenGL ES backend via generic bluegl fallback path (2-3 person-days testing with a RISC-V board that has OpenGL ES support)
  • Fix or stub any #ifdef __aarch64__ paths that have no __riscv equivalent in filament native code (1-2 person-days; primarily math/fast.h and ColorGrading)

Total functional enablement estimate: 2-3 person-weeks.

Blocker: requires Google maintainer to review and merge the toolchain file and any CMake changes. Without maintainer engagement, this work cannot land upstream.

13.2 Performance Optimization

  • ColorGrading RVV SIMD: Port 432-line NEON tone-mapping to RVV intrinsics (3-5 person-weeks; requires shader/pipeline knowledge and RISC-V SIMD expertise)
  • bluegl riscv64 trampolines: Author BlueGLCoreLinuxRiscV64Impl.S (1-2 person-weeks; requires ABI knowledge; aarch64 version is the reference)
  • basisu RVV texture transcoding: No upstream RVV path exists; contributing to basisu upstream is a prerequisite (5-10 person-weeks; basisu encoder is complex)
  • libwebp RVV DSP: Outside filament scope; tracked separately
  • zstd RVV: Outside filament scope; tracked separately

Total performance optimization estimate: 10-18 person-weeks for filament-native work only.

13.3 CI/CD Infrastructure

  • Add riscv64 QEMU-based CI job to presubmit.yml and postsubmit.yml (1-2 person-days authoring; requires Google maintainer approval and runner provisioning)
  • Procure or access RISC-V hardware runners (if QEMU performance is insufficient for GPU/render testing) - organizational effort, timeline not estimable

Note: CI access is restricted to Google maintainers. An external contributor can draft the YAML but cannot enable the runner. This is an organizational dependency, not a technical one.

13.4 Ecosystem Enablement

Not applicable. google/filament has no dependent package ecosystem requiring separate enablement (it is a native C++ rendering engine, not a language runtime or package manager host).

13.5 Summary Table

Area Work Item Effort (person-weeks) Owner Priority
Functional riscv64 CMake toolchain file + two-phase cross-build 0.5 Contributor + Google maintainer Critical
Functional Resolve link failures (abseil -latomic, others) 0.5 Contributor Critical
Functional Stub/fix aarch64-only paths in CMakeLists and native code 0.5 Contributor High
Functional Verify OpenGL ES backend on riscv64 hardware 0.5 Contributor High
Performance ColorGrading RVV SIMD port 4 Contributor (RISC-V SIMD expert) Medium
Performance bluegl riscv64 trampolines (.S) 1.5 Contributor (ABI expert) Medium
Performance basisu RVV texture transcoding (upstream first) 8 Contributor + basisu maintainer Low
CI/CD riscv64 QEMU presubmit/postsubmit job 0.5 (authoring) Contributor + Google maintainer High
CI/CD riscv64 hardware runner provisioning organizational Google Low

Pre-condition for all items: Google maintainer engagement. Without a committed Google reviewer, no upstream work can land. The investment estimate above assumes that engagement has been secured.

14. Updates

No updates yet – initial report dated 2026-08-14.

15. References