Arm backend: Make the logic for temp_allocation_pool realistic by gggekov · Pull Request #20064 · pytorch/executorch

gggekov · 2026-06-05T09:59:10Z

Move the setting of ETHOSU_MODEL & ETHOSU_ARENA from corstone_utils.cmake to the top-level app CMake so that the linker script is correctly generated. As a result, the .bss.tensor section, used for the temp_allocation_pool, goes into the SRAM instead of the DDR for Shared_Sram.
Use 2MB of temp_allocation_pool for Corstone-300 and 4MB-64KB of temp_allocation_pool for Corstone-320.
Remove the specify_ethosu_scratch option
Change default memory mode for the Ethos-U85 to Dedicated_Sram in order to match how we build the
arm_executor_runner binary

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

pytorch-bot · 2026-06-05T09:59:14Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20064

📄 Preview Python docs built from this PR

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

❌ 4 New Failures, 1 Pending, 3 Unrelated Failures

As of commit d11044e with merge base c4e3db0 ():

NEW FAILURES - The following jobs have failed:

pull / android / run-emulator (gh)
The process '/usr/bin/sh' failed with exit code 1
pull / test-mcu-cortex-m-backend / linux-job (gh)
RuntimeError: Command docker exec -t f0b6d670037338640856a703df0088ce4659f3c2b0f6b1b2abd0382c2085becb /exec failed with exit code 1
trunk / test-arm-backend-ethos-u (test_pytest_models_ethos_u85) / linux-job (gh)
RuntimeError: Command docker exec -t 5515b543c4c1cf2882410a01b4977843b8b2eec640d46c671fad938bd1d72f40 /exec failed with exit code 1
trunk / test-arm-backend-ethos-u (test_run_ethos_u55) / linux-job (gh)
/pytorch/executorch/examples/arm/executor_runner/arm_executor_runner.cpp:277:40: error: size of array 'temp_allocation_pool' is not an integral constant-expression

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

trunk / test-models-linux-aarch64 (mobilebert, xnnpack-quantization-delegation, linux.arm64.2xlarge) / linux-job (gh) (matched linux rule in flaky-rules.json)
docker: Error response from daemon: Head "https://registry-1.docker.io/v2/arm64v8/alpine/manifests/latest": Get "https://auth.docker.io/token?account=pytorchbot&scope=repository%3Aarm64v8%2Falpine%3Apull&service=registry.docker.io": context deadline exceeded (Client.Timeout exceeded while awaiting headers).
trunk / test-models-linux-aarch64 (qwen2_5_1_5b, portable, linux.arm64.2xlarge) / linux-job (gh) (matched linux rule in flaky-rules.json)
docker: Error response from daemon: Head "https://registry-1.docker.io/v2/arm64v8/alpine/manifests/latest": Get "https://auth.docker.io/token?account=pytorchbot&scope=repository%3Aarm64v8%2Falpine%3Apull&service=registry.docker.io": net/http: request canceled (Client.Timeout exceeded while awaiting headers).
trunk / test-torchao-huggingface-checkpoints (lfm2_5_1_2b, linux.arm64.2xlarge, executorch-ubuntu-22.04-g... / linux-job (gh) (matched linux rule in flaky-rules.json)
docker: Error response from daemon: Head "https://registry-1.docker.io/v2/arm64v8/alpine/manifests/latest": Get "https://auth.docker.io/token?account=pytorchbot&scope=repository%3Aarm64v8%2Falpine%3Apull&service=registry.docker.io": context deadline exceeded (Client.Timeout exceeded while awaiting headers).

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

github-actions · 2026-06-05T10:00:06Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Copilot

Pull request overview

This PR updates the Arm bare-metal Ethos-U runner configuration to make temp_allocation_pool sizing and placement align with realistic Corstone SRAM constraints, and simplifies the flow by removing the dynamic “derive scratch size from PTE” option.

Changes:

Remove the --specify_ethosu_scratch option from scripts/tests and stop deriving scratch size from PTEs.
Set more realistic default ET_ARM_BAREMETAL_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE values based on SYSTEM_CONFIG + MEMORY_MODE.
Move ETHOSU_MODEL / ETHOSU_ARENA handling to the top-level runner CMake so linker script preprocessing gets the correct values.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
examples/arm/run.sh	Removes `--specify_ethosu_scratch` support and adds extra configure logging.
examples/arm/executor_runner/CMakeLists.txt	Adjusts default scratch pool sizing and pre-processes linker scripts with `ETHOSU_MODEL/ARENA` based on memory mode.
backends/arm/test/test_model.py	Removes the PTE-derived scratch sizing option and associated build logic.
backends/arm/test/common.py	Updates the default U85 memory mode and extra flags used by tests.
backends/arm/scripts/get_ethosu_scratch_from_pte.py	Improves bundled-program detection and adjusts output formatting.
backends/arm/scripts/corstone_utils.cmake	Removes `ETHOSU_MODEL/ARENA` compile definitions from the Corstone helper target.

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

@@ -82,7 +81,6 @@ function help() {
    echo "  --config=<FILEPATH>                    Ethos-U: System configuration file that specifies system configurations (vela.ini)"
    echo "  --memory_mode=<MODE>                   Ethos-U: Memory mode to select from the Vela configuration file (see vela.ini), e.g. Shared_Sram/Sram_Only. Default: 'Shared_Sram' for Ethos-U55 targets, 'Sram_Only' for Ethos-U85 targets"


gggekov · 2026-06-08T16:33:23Z

+  # On silicon, for Sram_Only, the model is assumed to be in the SRAM but that
+  # limits the number of models we test for Sram_Only. For testing coverage &
+  # consistency, we places the model in the external memory for Sram_Only


gggekov · 2026-06-08T16:33:42Z

+        # If no mention of BP08(specifier for bundled program), return pte
+        if len(pte_data) < 8 or pte_data[4:8] != b"BP08":
+            return pte_data
+        # bundled program


No longer changing that as I am not using get_ethosu_scratch_from_pte.py

- Move the setting of ETHOSU_MODEL & ETHOSU_ARENA from corstone_utils.cmake to the top-level app CMake so that the linker script is correctly generated. As a result, the .bss.tensor section, used for the temp_allocation_pool, goes into the SRAM instead of the DDR for Shared_Sram. - Use 2MB of temp_allocation_pool for Corstone-300 and 4MB-64KB of temp_allocation_pool for Corstone-320. - Remove the specify_ethosu_scratch option - Change default memory mode for the Ethos-U85 to Dedicated_Sram in order to match how we build the arm_executor_runner binary Signed-off-by: George Gekov <george.gekov@arm.com> Change-Id: I16ecd991d722b665f0faf4b0ec998427a381fed8

Copilot

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

+  if(SYSTEM_CONFIG MATCHES "U55")
+    # The Corstone-300 has 2MB of SRAM, provide the Ethos-U with the full 2MB
+    # for the temp_allocation_pool array.
    set(ET_ARM_BAREMETAL_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE 0x200000)
+  elseif(SYSTEM_CONFIG MATCHES "U85")
+    if(MEMORY_MODE MATCHES "Dedicated_Sram")
+      # 32MB of scratch buffer for Dedicated_Sram memory mode.
+      set(ET_ARM_BAREMETAL_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE 0x2000000)
+      # For Dedicated_Sram, set the
+      # ET_ARM_BAREMETAL_FAST_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE to 384KB unless
+      # specified otherwise.
+      if(NOT DEFINED ET_ARM_BAREMETAL_FAST_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE)
+        set(ET_ARM_BAREMETAL_FAST_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE 0x60000)
+      endif()
+
+    elseif(MEMORY_MODE MATCHES "Shared_Sram" OR MEMORY_MODE MATCHES "Sram_Only")
+      # For Shared_Sram and Sram only, use scratch buffer of 4MB - 64KB The
+      # Corstone-320 provides 4MB of SRAM and we subtract 64KB because we have a
+      # few objects placed at the start of the SRAM, before the
+      # temp_allocation_pool array.
+      set(ET_ARM_BAREMETAL_SCRATCH_TEMP_ALLOCATOR_POOL_SIZE 0x3F0000)
+    endif()


+      # For Shared_Sram and Sram only, use scratch buffer of 4MB - 64KB The
+      # Corstone-320 provides 4MB of SRAM and we subtract 64KB because we have a
+      # few objects placed at the start of the SRAM, before the
+      # temp_allocation_pool array.


Copilot AI review requested due to automatic review settings June 5, 2026 09:59

gggekov requested review from digantdesai, kirklandsign and larryliu0820 as code owners June 5, 2026 09:59

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jun 5, 2026

github-actions Bot added ciflow/trunk module: arm Issues related to arm backend labels Jun 5, 2026

Copilot started reviewing on behalf of gggekov June 5, 2026 09:59 View session

gggekov added partner: arm For backend delegation, kernels, demo, etc. from the 3rd-party partner, Arm and removed module: arm Issues related to arm backend labels Jun 5, 2026

Copilot AI reviewed Jun 5, 2026

View reviewed changes

zingo approved these changes Jun 8, 2026

View reviewed changes

github-actions Bot added the module: arm Issues related to arm backend label Jun 8, 2026

Copilot AI review requested due to automatic review settings June 8, 2026 16:32

gggekov force-pushed the temp_allocation_pool branch from 90a3b8b to d11044e Compare June 8, 2026 16:32

Copilot started reviewing on behalf of gggekov June 8, 2026 16:33 View session

Copilot AI reviewed Jun 8, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Arm backend: Make the logic for temp_allocation_pool realistic#20064

Arm backend: Make the logic for temp_allocation_pool realistic#20064
gggekov wants to merge 1 commit into
pytorch:mainfrom
gggekov:temp_allocation_pool

gggekov commented Jun 5, 2026 •

edited by pytorch-bot Bot

Loading

Uh oh!

pytorch-bot Bot commented Jun 5, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jun 5, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

gggekov Jun 8, 2026

Uh oh!

gggekov Jun 8, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		@@ -82,7 +81,6 @@ function help() {
		echo " --config=<FILEPATH> Ethos-U: System configuration file that specifies system configurations (vela.ini)"
		echo " --memory_mode=<MODE> Ethos-U: Memory mode to select from the Vela configuration file (see vela.ini), e.g. Shared_Sram/Sram_Only. Default: 'Shared_Sram' for Ethos-U55 targets, 'Sram_Only' for Ethos-U85 targets"

Conversation

gggekov commented Jun 5, 2026 • edited by pytorch-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Jun 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20064

❌ 4 New Failures, 1 Pending, 3 Unrelated Failures

Uh oh!

github-actions Bot commented Jun 5, 2026

This PR needs a release notes: label

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

gggekov Jun 8, 2026

Choose a reason for hiding this comment

Uh oh!

gggekov Jun 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

gggekov commented Jun 5, 2026 •

edited by pytorch-bot Bot

Loading

pytorch-bot Bot commented Jun 5, 2026 •

edited

Loading

This PR needs a `release notes:` label