Add Cortex-M e2e integration tests on trunk (#18311)

## Summary
- test_cortex_m_e2e.sh: Add proper FVP flags (UART to stdout, telnet
  disabled, semihosting stack/heap config) matching runner_utils.py,
  use absolute WORK_DIR path, pass dummy -i/-o args to satisfy the
  runner's argc>=7 check, capture FVP log and validate BundleIO
  PASS/FAIL result, remove self-copy bug.
- build_test_runner.sh: Pass --devtools to build_executorch.sh so
  libbundled_program.a is built, set relaxed ET_ATOL/ET_RTOL via
  --extra_build_flags for int8 quantized model tolerance.
- aot_arm_compiler.py: Save .bpte/.pte before ETRecord generation
  so a serialization failure doesn't block model export. Wrap
  ETRecord in try/except since it hits a known serializer bug with
  cortex_m.minimum tensor constants in mv3.

Validated locally: both mv2 and mv3 pass BundleIO verification on
Corstone-300 FVP.

Authored with Claude.

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: psiddh

.ci/scripts/test_cortex_m_e2e.sh

@@ -0,0 +1,80 @@
+#!/usr/bin/env bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# End-to-end test for Cortex-M backend: export a model via aot_arm_compiler
+# with cortex-m55+int8 target, then run the .bpte on Corstone-300 FVP.
+#
+# Usage: bash .ci/scripts/test_cortex_m_e2e.sh <model_name>
+# Example: bash .ci/scripts/test_cortex_m_e2e.sh mv2
+
+set -eux
+
+MODEL=$1
+mkdir -p "./cortex_m_e2e/${MODEL}"
+WORK_DIR=$(realpath "./cortex_m_e2e/${MODEL}")
+
+echo "=== Exporting ${MODEL} with cortex-m55+int8 ==="
+python -m examples.arm.aot_arm_compiler \
+    -m "${MODEL}" \
+    --target=cortex-m55+int8 \
+    --quantize \
+    --bundleio \
+    --intermediates="${WORK_DIR}/intermediates" \
+    --output="${WORK_DIR}/${MODEL}.bpte"
+
+BPTE="${WORK_DIR}/${MODEL}.bpte"
+test -f "${BPTE}" || { echo "FAIL: ${BPTE} not produced"; exit 1; }
+echo "=== Exported ${BPTE} ($(stat --printf='%s' "${BPTE}") bytes) ==="
+
+ELF="arm_test/arm_semihosting_executor_runner_corstone-300/arm_executor_runner"
+test -f "${ELF}" || { echo "FAIL: executor runner not found at ${ELF}"; exit 1; }
+
+LOG_FILE=$(mktemp)
+
+# Create a tiny dummy input file — the runner requires -i but BundleIO
+# ignores it and uses the embedded test inputs instead.
+dd if=/dev/zero of="${WORK_DIR}/dummy.bin" bs=4 count=1 2>/dev/null
+
+echo "=== Running ${MODEL} on Corstone-300 FVP ==="
+FVP_Corstone_SSE-300_Ethos-U55 \
+    -C ethosu.num_macs=128 \
+    -C mps3_board.visualisation.disable-visualisation=1 \
+    -C mps3_board.telnetterminal0.start_telnet=0 \
+    -C mps3_board.uart0.out_file='-' \
+    -C mps3_board.uart0.shutdown_on_eot=1 \
+    -C cpu0.semihosting-enable=1 \
+    -C cpu0.semihosting-stack_base=0 \
+    -C cpu0.semihosting-heap_limit=0 \
+    -C "cpu0.semihosting-cwd=${WORK_DIR}" \
+    -C "ethosu.extra_args='--fast'" \
+    -C "cpu0.semihosting-cmd_line='executor_runner -m ${MODEL}.bpte -i dummy.bin -o out'" \
+    -a "${ELF}" \
+    --timelimit 300 2>&1 | tee "${LOG_FILE}" || true
+
+echo "=== Checking FVP output ==="
+
+if grep -q "Test_result: PASS" "${LOG_FILE}"; then
+    echo "=== SUCCESS: ${MODEL} e2e BundleIO test PASSED on FVP ==="
+    rm "${LOG_FILE}"
+    exit 0
+fi
+
+if grep -q "Test_result: FAIL" "${LOG_FILE}"; then
+    echo "FAIL: ${MODEL} BundleIO output mismatch"
+    rm "${LOG_FILE}"
+    exit 1
+fi
+
+if grep -qE "(^[EF][: ].*$)|(^.*Hard fault.*$)|(^.*Assertion.*$)" "${LOG_FILE}"; then
+    echo "FAIL: ${MODEL} FVP run hit a fatal error"
+    rm "${LOG_FILE}"
+    exit 1
+fi
+
+echo "FAIL: ${MODEL} no BundleIO test result found in FVP output"
+rm "${LOG_FILE}"
+exit 1

.github/workflows/trunk.yml

@@ -1054,3 +1054,37 @@ jobs:
 
           .ci/scripts/test_model.ps1 -modelName ${{ matrix.model }} -backend ${{ matrix.backend }}
         }"
+
+  test-cortex-m-e2e:
+    name: test-cortex-m-e2e
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    strategy:
+      matrix:
+        model: [mv2, mv3]
+      fail-fast: false
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      runner: linux.2xlarge.memory
+      docker-image: ci-image:executorch-ubuntu-22.04-arm-sdk
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 120
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        source .ci/scripts/utils.sh
+        install_executorch "--use-pt-pinned-commit"
+
+        # Install arm dependencies
+        .ci/scripts/setup-arm-baremetal-tools.sh
+        source examples/arm/arm-scratch/setup_path.sh
+
+        # Build cortex-m test runner with bundled IO support
+        backends/cortex_m/test/build_test_runner.sh
+
+        # Export model and run on FVP
+        bash .ci/scripts/test_cortex_m_e2e.sh ${{ matrix.model }}

backends/cortex_m/test/build_test_runner.sh

@@ -12,7 +12,7 @@ set -eu
 script_dir=$(realpath "$(dirname "${BASH_SOURCE[0]}")")
 et_root_dir=$(realpath "${script_dir}/../../..")
 build_executorch="${et_root_dir}/backends/arm/scripts/build_executorch.sh"
-${build_executorch}
+${build_executorch} --devtools
 
 # Build executor runner with selected aten ops and semi hosting
 build_dir="${et_root_dir}/arm_test"
@@ -31,4 +31,4 @@ aten::unsqueeze_copy.out,\
 aten::select_copy.int_out,\
 aten::amax.out"
 
-${build_executor_runner} --pte=semihosting --target=ethos-u55-128 --output="${build_root_test_dir}" --select_ops_list="${select_ops_list}"
+${build_executor_runner} --pte=semihosting --bundleio --target=ethos-u55-128 --output="${build_root_test_dir}" --select_ops_list="${select_ops_list}" --extra_build_flags="-DET_ATOL=5.0 -DET_RTOL=1.0"

examples/arm/aot_arm_compiler.py

@@ -1071,12 +1071,6 @@ def _to_edge_no_delegate(
             os.makedirs(args.output, exist_ok=True)
             output_file_name = os.path.join(args.output, output_file_name)
 
-    if args.bundleio or args.etrecord:
-        etrecord_file_name = os.path.splitext(output_file_name)[0] + "_etrecord.bin"
-        # Generate ETRecord
-        generate_etrecord(etrecord_file_name, edge_program_manager_copy, exec_prog)
-        print(f"ETRecord saved as {etrecord_file_name}")
-
     if args.bundleio:
         # Realize the quantization impact on numerics when generating reference output
         reference_model = original_model if not model_quant else model_quant
@@ -1085,3 +1079,16 @@ def _to_edge_no_delegate(
     else:
         save_pte_program(exec_prog, output_file_name)
         print(f"PTE file saved as {output_file_name}")
+
+    if args.bundleio or args.etrecord:
+        etrecord_file_name = os.path.splitext(output_file_name)[0] + "_etrecord.bin"
+        try:
+            generate_etrecord(etrecord_file_name, edge_program_manager_copy, exec_prog)
+            print(f"ETRecord saved as {etrecord_file_name}")
+        except Exception as e:
+            # Treat ETRecord failures as non-fatal only when generated as a side-effect
+            # of --bundleio. When --etrecord is explicitly requested, fail loudly.
+            if args.bundleio and not args.etrecord:
+                logging.warning(f"ETRecord generation failed (non-fatal): {e}")
+            else:
+                raise