Arm backend: Add baseline functionality for zero copy IO project.

[wwwind]

Add baseline functionality for the VGF zero-copy I/O project.

This PR measures the current delegate-boundary copy cost before introducing host-memory import or descriptor rebinding. It records:

• input/output copy bytes,
• input/output memcpy time,
• dispatch-and-wait time,
• total VGF execute time.

The baseline uses 20 warmup iterations followed by 200 measured iterations in a single process.

Initial host-emulation measurements show:

• MobileNetV2 FP32 224

  • input: 602,112 bytes
  • output: 4,000 bytes
  • input memcpy p50: ~0.014 ms
  • output memcpy p50: ~0.0004 ms
  • VGF execute p50: ~79.4 ms

• RIFE

  • input copies: 22,118,400 bytes
  • output copy: 11,059,200 bytes
  • total boundary copy: 33,177,600 bytes / frame (~31.64 MiB)
  • memcpy p50: ~0.772 ms
  • memcpy p90: ~0.937 ms

These measurements were collected using the emulation layer and are not native Mali-device performance results.

No zero-copy behaviour is introduced by this PR.

The goal is to establish a reproducible baseline against which later zero-copy host-memory import changes can be compared.

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/23309

• 📄 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 Pending

As of commit 306a22d with merge base fdd5140 ():

NEW FAILURES - The following jobs have failed:

• Apple / build-benchmark-app / macos-job (gh)
  RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 65
• trunk / test-samsung-models-linux / linux-job (gh)
  test_mv3_fp16

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

[zingo]

I see some fails

The following tests FAILED:
53 - vgf_execution_stats_enabled_test (Not Run)
54 - vgf_execution_stats_disabled_test (Not Run)
Errors while running CTest

Was this added by your new macros maybe?

[wwwind]

let me take a look.

[zingo]

Ok to merge when CI finish and all is OK.