Arm backend: Direct drive patches and README.md

[perheld]

cc @digantdesai @SS-JIA @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17735

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 3 New Failures, 1 Cancelled Job

As of commit c8c9a44 with merge base 331ac4a ():

NEW FAILURES - The following jobs have failed:

• Build Presets / apple (macos) / build (gh)
  The process '/opt/homebrew/bin/git' failed with exit code 128
• pull / test-samsung-models-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 0a3d5bddbe6e68586aadcd33198d89783c4bbf828ee7d0be9bbdd0a88f7bf8c1 /exec failed with exit code 1
• pull / test-samsung-quantmodels-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 2b3c6bb70091ff00e24bfff816b90a0275a491bfcd9dbff4c535cdf8f8da8ec8 /exec failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• trunk / test-models-macos-coreml (efficient_sam) / macos-job (gh)
  ##[error]The operation was canceled.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Copilot]

Pull request overview

Updates the Arm backend to better support the experimental “Direct Drive” Ethos-U85-on-Linux workflow by adjusting runtime output handling, documenting the workflow, and centralizing the CMake preset configuration into a preset file.

Changes:

• Add an arm_ethosu_linux.cmake preset file and switch the arm-ethosu-linux CMake preset to reference it.
• Update the Ethos-U Cortex-A (Linux) runtime path to handle output buffer size/layout mismatches via scratch buffering + layout adjustment instead of failing.
• Expand backends/arm/README.md with target-specific build/setup sections, including Direct Drive workflow instructions.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
tools/cmake/preset/arm_ethosu_linux.cmake	New build preset file for Ethos-U Linux, including Release-flag adjustment.
CMakePresets.json	Switch arm-ethosu-linux to load the new preset file instead of inline cache variables.
backends/arm/runtime/EthosUBackend_Cortex_A.cpp	Adjust Linux output handling to support IO/tensor size mismatch using scratch buffers + copy_with_layout_adjustment.
backends/arm/README.md	Document target workflows and add Direct Drive setup/build/run instructions.

Comments suppressed due to low confidence (2)

tools/cmake/preset/arm_ethosu_linux.cmake:19

• CMAKE_C_FLAGS_RELEASE / CMAKE_CXX_FLAGS_RELEASE are being set to just -UNDEBUG, which overwrites CMake’s default Release flags (typically includes -O* and -DNDEBUG). This can unintentionally strip optimizations and other Release settings. Prefer appending -UNDEBUG to the existing *_FLAGS_RELEASE (or adding it via a config-specific add_compile_options generator expression) instead of replacing the variable value outright.

set(CMAKE_C_FLAGS_RELEASE
    "-UNDEBUG"
    CACHE STRING "Avoid NDEBUG forward-decl mismatch in musl Release builds"
)
set(CMAKE_CXX_FLAGS_RELEASE
    "-UNDEBUG"
    CACHE STRING "Avoid NDEBUG forward-decl mismatch in musl Release builds"
)

backends/arm/README.md:110

• This section says each target flow has both a one-time setup step and a build command, but the VGF workflow currently documents only the setup step (no build invocation). Either add the corresponding build command for VGF or adjust the wording so it doesn’t promise a build step for every flow.

Pick one of the target flows below. Each flow has a one-time setup step and a build command.

### Baremetal (Ethos-U) workflow

Builds ExecuTorch runtime libraries for Cortex-M with Ethos-U acceleration.

Setup:

./examples/arm/setup.sh --i-agree-to-the-contained-eula

Build:

./backends/arm/scripts/build_executorch.sh

### VGF (Vulkan ML extensions) workflow

Setup:

./examples/arm/setup.sh --disable-ethos-u-deps --enable-mlsdk-deps

The current flow lowers to TOSA and converts to VGF for use in external projects,
so the `executor_runner` is not typically used here.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[zingo]

Hi @SS-JIA this adds a file and also updates files outside Arm folder (but in Arm parts)

[zingo]

OK to merge after @SS-JIA also gives OK :)