Arm backend: Move rescales from SUM visitor to pass

[martinlsm]

In the SUM node visitor, an INT8->INT32 RESCALE node is inserted prior to the SUM node; similarly, an INT32->INT8 RESCALE node is inserted after.

This patch moves the insertion to InsertRescaleInt32Pass. Since SUM is decomposed, insertion of RESCALE nodes should be carried out before DecomposeSumPass (which decomposes SUM into a chain of single dim SUMs). The ordering is important to avoid redundant INT8/INT32 RESCALE nodes being inserted between each SUM node in the chain after decomposition. Only one INT8->INT32 RESCALE is needed before the chain, and an INT32->INT8 after it; between the SUM nodes in the chain, the edges are already in the correct INT32 data type.

Test plan

Tests exercising the modified pass of this patch have been added to backends/arm/test/passes/test_insert_rescale_i32_pass.py.

cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15299

• 📄 Preview Python docs built from this PR

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

❌ 8 New Failures, 3 Unrelated Failures

As of commit 0cec77a with merge base f7ca57e ():

NEW FAILURES - The following jobs have failed:

• pull / test-samsung-models-linux / linux-job (gh)
  RuntimeError: Command docker exec -t d97976fc5a540a3fb7b1579b1a8a28ca7c3764ee3f2c28b949e775e09897d486 /exec failed with exit code 1
• pull / unittest-nxp-neutron / linux-job (gh)
  RuntimeError: Command docker exec -t 79f4f9cf89d309b7d2946255ff18fa58ccfaee125fc74302dd2d0af371e1ad2b /exec failed with exit code 1
• Test CUDA Builds / export-gemma3-cuda-non-quantized / linux-job (gh)
  RuntimeError: Command docker exec -t 24961937ee3a8952ac05a0154d2c3335b0b8c8431635910b7ad6f1ee4d9af9fa /exec failed with exit code 2
• Test CUDA Builds / export-gemma3-cuda-quantized-int4-tile-packed / linux-job (gh)
  RuntimeError: Command docker exec -t 1c6cb421645efbe8f119615476eb48a0a7226d86d15ecf39f6ce74c292e372f2 /exec failed with exit code 2
• Test CUDA Builds / export-voxtral-cuda-non-quantized / linux-job (gh)
  RuntimeError: Command docker exec -t 9b7391ee9a44a7d14cf920bca7a1539a9338100637de81bfaf41c794be5c2c40 /exec failed with exit code 2
• Test CUDA Builds / export-voxtral-cuda-quantized-int4-tile-packed / linux-job (gh)
  RuntimeError: Command docker exec -t 41f48ca06e0de19ed9c506978fe748c4034f789fc46b31113bb458bf21a085eb /exec failed with exit code 2
• Test CUDA Builds / export-voxtral-cuda-quantized-int4-weight-only / linux-job (gh)
  RuntimeError: Command docker exec -t 6966dd199e55e71b8ef3470a522f1b367b930e97d7e3a64b49cf1ffcbb93c855 /exec failed with exit code 2
• Test Metal Backend / export-voxtral-metal-artifact / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 2

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

• pull / unittest / linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / unittest-release / linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / unittest-editable / linux / linux-job (gh) (trunk failure)

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

[martinlsm]

@pytorchbot label ciflow/trunk

[martinlsm]

@pytorchbot label "partner: arm"

[Sebastian-Larsson]

Unrelated CI failures

[Sebastian-Larsson]

This is failing
trunk / test-arm-backend (test_pytest_ops_ethosu_fvp) / linux-job (gh)

[martinlsm]

This is failing trunk / test-arm-backend (test_pytest_ops_ethosu_fvp) / linux-job (gh)

This must be a flaky test. It has not been seen in our internal CI nor have I managed to reproduce it locally by looping this test.

[zingo]

It will probably work in that case if you retrigget the job?