Release tracker - 3.3.0

[ScottTodd]

Overview

Our previous stable release was published on 2025-02-10: https://github.com/iree-org/iree/releases/tag/v3.2.0. We aim for roughly one stable release every 6 weeks, which will be around 2025-03-24. This issue tracks the release process for the next release.

Release process docs:

• https://iree.dev/developers/general/release-management/
• https://github.com/iree-org/iree/blob/main/RELEASING.md

Release checklist:

• Watch for major/breaking changes and decide to either batch them with this release or defer them until the next release
• Choose a release candidate from nightly releases to promote. The candidate should contain no major regressions and should include all packages, even those marked experimental in .github/workflows/build_package.yml like macOS and Windows packages.
• Compile release notes
• Promote to stable (on GitHub) and push packages to PyPI
• Roll forward downstream packages like iree-turbine and push similar package updates to PyPI
• Update release process docs for next time

Release notes

Candidate: TBD

Commit range

• Previous release to main: v3.2.0...main
• Previous release to release candidate: TBD

(Contributors can edit this section as they land patches)

---

Announcements

• A presentation on IREE was given at Vulkanized 2025 titled The Long Tail of AI: SPIR-V in IREE and MLIR. See the presentations and talks section of the repository README for other talks.

Notable changes

Framework support

• Expanded support for ONNX operators and models with changes like [VM] Add some missing arith-to-vm patterns #20079.

Compiler

• Added support for AMD Radeon RDNA4 GPUs with ROCm/HIP and Vulkan. See the documentation at https://iree.dev/guides/deployment-configurations/gpu-rocm/#choosing-hip-targets and a demo application with shark-ai's SDXL implementation.
• Add umbrella optimization flags #20047, [docs] Update Optimization Options #20216: A new --iree-opt-level=[O0,O1,O2,O3] umbrella flag was added to the compiler. See the documentation at https://iree.dev/reference/optimization-options/.
• Refresh compile/run examples in deployment configuration guides. #19920, Expand on device and driver documentation, notably for ROCm. #20112: The https://iree.dev/guides/deployment-configurations/ pages have been updated with simpler sample code and more comprehensive sections for CPU and GPU target and feature selection.
• (PENDING) Substantial progress has been made on GPU data-tiling.
• Performance improvements for ROCm: Do not approximate erf on rocm. #19969, [AMDGPU] Do not rewrite or approximate math functions on ROCm #19970, [AMDGPU] Do not rewrite or approximate math functions on ROCm #20222.
• (PENDING) Removed the iree_input dialect. Users should lower into the hal and flow dialects as required: Deleting iree_input dialect. #20256.

Runtime

• HIP runtime bug fixes and performance improvements: [hip] Fix async read offsets. #19999, [hip] Post the correct error to the semaphores. #20000, [hip] Move alloc operations off the main stream #20001, [hip] Move calls to hipFree and similar to the cleanup thread. #20020.
• The HIP runtime is now supported on Windows: [hip] Handle when devices cannot access peers, such as on Windows. #20114.

Development tools

• [Tool] --show-dialects, --list-passes needed to take input #20003: The iree-opt --show-dialects and iree-opt --list-passes commands are now functional.
• E2E tests for math ops. #20169: New tests for 28+ math operations.

Changelog

Full list of changes: v3.2.0...v3.3.0

---

[ScottTodd]

Time to pick release candidates. I'll nominate 3.3.0rc20250317 based on some feedback from testing and CI results, but we could also push for 3.3.0rc20250318 or 3.3.0rc20250319.

This installs the 0317 release for all packages:

pip install \
  --find-links https://github.com/iree-org/iree/releases/expanded_assets/iree-3.3.0rc20250317 \
  iree-base-compiler==3.3.0rc20250317 \
  iree-base-runtime==3.3.0rc20250317 \
  iree-tools-tf==20250317.1203 \
  iree-tools-tflite==20250317.1203 \
  --find-links https://github.com/iree-org/iree-turbine/releases/expanded_assets/dev-wheels \
  iree-turbine==3.3.0rc20250317 \
  --find-links https://github.com/nod-ai/shark-ai/releases/expanded_assets/dev-wheels \
  shortfin==3.3.0rc20250317 \
  sharktank==3.3.0rc20250317

Looking at changes to https://github.com/nod-ai/shark-ai/commits/main/ recently, it's probably safe and beneficial to bump that to 0319 instead of 0317 as there were several cleanup, documentation, and API surface area changes.

[ScottTodd]

We're running some QA on last night's build, 3.3.0rc20250320. Still not fully decided which release candidates we will promote.

[ScottTodd]

We received a number of requests to bump forward to include additional changes in this release, so the new versions scheduled for promotion on Monday are 3.3.0rc20250320 for IREE packages and 3.3.0rc20250321 for shark-ai.

Install instructions for testing:

pip install \
  --find-links https://iree.dev/pip-release-links.html \
  iree-base-compiler==3.3.0rc20250320 \
  iree-base-runtime==3.3.0rc20250320 \
  iree-tools-tf==20250320.1206 \
  iree-tools-tflite==20250320.1206 \
  --find-links https://iree.dev/pip-release-links.html \
  iree-turbine==3.3.0rc20250320 \
  --find-links https://github.com/nod-ai/shark-ai/releases/expanded_assets/dev-wheels \
  shortfin==3.3.0rc20250321 \
  sharktank==3.3.0rc20250321

[ScottTodd]

Releases are published.