Remove extern "C" wrapping and fix format specifiers for ARM embedded builds

[Ninja91]

Summary:
Remove redundant extern "C" blocks wrapping CMSIS-NN header includes and fix -Wformat errors in format specifiers. These changes are required for the CC pipeline's FVP benchmark runner to compile on ARM embedded targets (Cortex-M55/M85 with MVE).

Context

The extern "C" wrapping around CMSIS-NN headers causes ARM embedded builds targeting MVE-capable processors to fail. CMSIS-NN's arm_nn_math_types.h temporarily closes its inner extern "C" before including arm_mve.h, but the outer extern "C" from the op files remains active, forcing arm_mve.h into C linkage where C++ function overloading is illegal.

Additionally, format specifiers (%hhd for ScalarType, %d/%ld for int64_t) cause -Wformat errors treated as build failures on ARM toolchains.

Changes:

1. Removed extern "C" wrapping from all 13 op .cpp files and cmsis_scratch_buffer_context.h
2. Consolidated CMSIS-NN includes in cortex_m_ops_common.h — added #include "arm_nnfunctions.h" (without extern "C") so op files get it transitively
3. Added #include <cinttypes> for PRIi64 macro
4. Fixed %hhd → %d with static_cast<int> for ScalarType values
5. Fixed %d/%ld → PRIi64 for int64_t values

Differential Revision: D95739935

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

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18000

• 📄 Preview Python docs built from this PR

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

⏳ No Failures, 279 Pending

As of commit e529b28 with merge base dda73d3 ():
💚 Looks good so far! There are no failures yet. 💚

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

[meta-codesync]

@Ninja91 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D95739935.

[github-actions]

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 Cortex-M backend CMSIS-NN integration to unblock ARM embedded (MVE) builds by removing problematic outer extern "C" wrapping around CMSIS-NN headers and by addressing -Wformat build failures from mismatched format specifiers.

Changes:

• Removed file-local extern "C" blocks around arm_nnfunctions.h includes across Cortex-M op sources and cmsis_scratch_buffer_context.h.
• Centralized CMSIS-NN includes in cortex_m_ops_common.h (including arm_nnfunctions.h) for transitive access from op files.
• Fixed several format specifiers in cortex_m_ops_common.h (ScalarType printing and int64_t via PRIi64), and added <cinttypes>.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
backends/cortex_m/ops/op_transpose.cpp	Removes local CMSIS-NN include wrapping (now relies on common header).
backends/cortex_m/ops/op_softmax.cpp	Removes local CMSIS-NN include wrapping (now relies on common header).
backends/cortex_m/ops/op_quantized_transpose_conv2d.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_mul.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_max_pool2d.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_linear.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_depthwise_conv2d.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_conv2d.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_avg_pool2d.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_quantized_add.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_pad.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_minimum.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/op_maximum.cpp	Removes local CMSIS-NN include wrapping.
backends/cortex_m/ops/cortex_m_ops_common.h	Adds <cinttypes>, centralizes CMSIS-NN includes, and fixes format specifiers.
backends/cortex_m/ops/cmsis_scratch_buffer_context.h	Removes extern "C" wrapping around CMSIS-NN include.

---

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

[Copilot]

In validate_single_quant_params, the zero_point parameter is unused. If this header is compiled with -Wall, this can trigger -Wunused-parameter (often fatal with -Werror on embedded toolchains). Consider removing the parameter, marking it ET_UNUSED, or implementing the intended zero-point range validation.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

---

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

[Copilot]

validate_single_quant_params takes zero_point but explicitly ignores it. Callers (e.g., validate_quantization_params / ops casting zero_point down to int32_t) therefore don't validate that zero points are within an expected/representable range, and the nearby doc comment claims that zero points are checked. Consider either validating zero_point (e.g., int8 or at least int32 range, consistent with CMSIS/AOT expectations) or removing the parameter/updating the contract to avoid misleading validation.

Suggested change

	(void)zero_point;
	ET_CHECK_MSG(
	ET_CHECK_MSG(
	zero_point >= std::numeric_limits<int8_t>::min() &&
	zero_point <= std::numeric_limits<int8_t>::max(),
	"%s zero_point must be in int8 range [-128, 127] [Value: %" PRIi64 "]",
	param_name,
	zero_point);
	ET_CHECK_MSG(

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

---

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

[Copilot]

Now that cortex_m_ops_common.h includes CMSIS-NN headers, this file’s direct #include "arm_nnfunctions.h" is redundant and also puts CMSIS includes before the common header (which may introduce ordering issues depending on CMSIS header dependencies). Prefer relying on cortex_m_ops_common.h alone, or include it first and drop the standalone CMSIS include.

Suggested change

#include "arm_nnfunctions.h"

[rascani]

As we discussed OOB, we typically follow the Google C++ style guide, which recommends against transitive includes.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (1)

backends/cortex_m/ops/cortex_m_ops_common.h:105

• The doc comment for validate_quantization_params claims it "Checks that zero points fit in int8 range", but validate_single_quant_params currently ignores zero_point entirely ((void)zero_point). Either add the zero_point range validation here or update the comment so it matches the actual checks being performed.

inline void validate_single_quant_params(
    const int64_t zero_point,
    const int64_t multiplier,
    const int64_t shift,
    const char* param_name) {
  (void)zero_point;
  ET_CHECK_MSG(
      multiplier >= std::numeric_limits<int32_t>::min() &&
          multiplier <= std::numeric_limits<int32_t>::max(),
      "%s multiplier must be in int32 range [Value: %" PRIi64 "]",
      param_name,
      multiplier);

  ET_CHECK_MSG(
      shift >= -31 && shift <= 31,
      "%s shift must be in range [-31, 31] [Value: %" PRIi64 "]",
      param_name,
      shift);
}

/**
 * Validate quantization parameters for inputs and output.
 *
 * Checks that zero points fit in int8 range, multipliers fit in int32 range,
 * and shifts are within a valid bit-shift range (0-31).
 *
 * Ensures parameters comply with Ahead-Of-Time (AOT) quantization requirements
 * and CMSIS-NN kernel expectations.
 *
 * Raises errors via ET_KERNEL_CHECK if any check fails.
 */

---

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

[Copilot]

PR description mentions consolidating CMSIS-NN includes by adding arm_nnfunctions.h to cortex_m_ops_common.h for transitive inclusion, but cortex_m_ops_common.h currently only includes arm_nn_types.h. Either update the PR description to reflect the current approach (explicit includes per op file) or actually add the include to match the described change.

[Copilot]

pad_out computes out_n/out_h/out_w/out_c from input_dims and pad values, then writes output_data using those computed extents without verifying that the provided 'out' tensor has matching sizes/numel. If out is incorrectly sized, this becomes an out-of-bounds write. Validate out sizes against the computed padded shape (or derive loop bounds from out.sizes()) before populating output_data.

[Copilot]

pad_out replaces arm_pad_s8 with a scalar nested-loop implementation. This is likely slower and increases code size on embedded targets; since the goal is to fix linkage issues (not replace kernels), restore the CMSIS-NN arm_pad_s8 call now that CMSIS headers are included without an outer extern "C" wrapper.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated no new comments.

---

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

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated no new comments.

---

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