Keep DS-CNN example inputs portable-safe

[JacobSzwejbka]

Summary

PR #21485 moved the shared DS-CNN example input to channels-last to satisfy the Cortex-M backend. That regressed the portable model test because portable convolution expects the input's standard NCHW dim order.

DS-CNN's input shape is (1, 1, 49, 10). With a singleton channel, its original NCHW tensor reports both contiguous and channels-last contiguous even though its strides remain (490, 490, 10, 1). The Cortex-M compiler's conditional conversion therefore did not materialize channels-last strides.

• Restore the shared DS-CNN example input to standard contiguous NCHW layout.
• Always request channels-last for 4D tensors inside the Cortex-M compiler.
• Propagate the converted example inputs for downstream bundle/reference generation.

This ports the release/1.4 fix from #21563 to main.

Test plan

• python3 -m py_compile examples/models/mlperf_tiny/ds_cnn.py backends/arm/scripts/aot_arm_compiler.py
• git diff --check
• Loaded the wrapper with PyTorch 2.13 and verified:
  • portable input shape (1, 1, 49, 10)
  • portable stride (490, 490, 10, 1)
  • Cortex-M channels-last stride (490, 1, 10, 1)
  • converted values are unchanged
• Portable and Cortex-M end-to-end coverage delegated to CI.
• The identical change is already running on release/1.4 and passed its full 20-wheel matrix.

Authored with Codex.

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

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21619

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 2 Cancelled Jobs, 26 Pending, 24 Unrelated Failures, 1 Unclassified Failure

As of commit aab62df with merge base 85ade9f ():

NEW FAILURES - The following jobs have failed:

• Build Presets / linux (pybind, linux.2xlarge, executorch-ubuntu-22.04-clang12) / build (gh)
  RuntimeError: Command docker exec -t 4b5a41eb837004a4eec64a5d1575dfae5f83d1d35b9ca2e3c29d448c059d91cc /exec failed with exit code 1
• pull / android / run-emulator (gh)
  adb: device offline

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• Check Labels / Check labels (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)

CANCELLED JOBS - The following jobs were cancelled. Please retry:

• Apple / build-benchmark-app / macos-job (gh)
• Build documentation / build (buck2) / Build doc (gh)

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

• Test CoreML Backend / Get changed files / get-changed-files (gh) (similar failure)
• Test RISC-V Backend / test-riscv (add, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (add, true, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (add, true, true) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (llama2, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (llama2, true, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (llama2, true, true) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mobilebert, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mobilebert, true, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mobilebert, true, true) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mv2, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mv2, true, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (mv2, true, true) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (resnet18, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (resnet18, true, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (resnet18, true, true) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (yolo26, false, false) / run / linux-job (gh) (similar failure)
• Test RISC-V Backend / test-riscv (yolo26, true, false) / run / linux-job (gh) (similar failure)

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

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

• Apple / build-demo-ios / macos-job (gh) (trunk failure)
• Apple / set-version (gh) (trunk failure)
• Test ARM Backend / test-arm / test-backend-linux (arm_tosa_fp, models) / linux-job (gh) (trunk failure)
• Test ARM Backend / test-arm / test-backend-linux (arm_tosa_fp, operators) / linux-job (gh) (trunk failure)
• Test ARM Backend / test-arm / test-backend-linux (arm_vgf_fp, models) / linux-job (gh) (trunk failure)
• Test ARM Backend / test-arm / test-backend-linux (arm_vgf_fp, operators) / linux-job (gh) (trunk failure)

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

[shoumikhin]

does release/1.4 need this too? #21485 was cherry-picked there as #21498, so the ds_cnn portable break probably rode along with it.

[digantdesai]

this is based on the implicit, always-do-to-nhwc call for cortex-m, I guess? should we add a test for that?

[shoumikhin]

merged to main.

two things are still open and would be lost here, so noting them:

1. digant's question above about adding a test for the cortex-m nhwc conversion.
2. does release/1.4 need this too? Fix DS-CNN Cortex-M input memory format #21485 went there as Fix DS-CNN Cortex-M input memory format #21498.

the ds_cnn portable job runs from the nightly workflow, not from pull or trunk, so the next scheduled run is the first place this shows up as green.