Matlab mex cpp interface by robince · Pull Request #15 · robince/gcmi

robince · 2026-04-03T23:12:11Z

No description provided.

robince · 2026-04-03T23:12:26Z

Copilot

Pull request overview

Introduces an initial MATLAB-native C++ MEX “optimized runtime” (with build/test/package tooling) and wires it into repo setup/benchmark documentation, while adding planning docs for a separate Python-only optimized path.

Changes:

Add a new matlab/cpp_mex/ C++ MEX implementation (probe entrypoints + info_cc_slice_cpp / info_cd_slice_cpp) with MATLAB buildtool tasks and smoke tests.
Teach setup_gcmi (and docs) to auto-add matlab/cpp_mex/bin/<release>/<mexext>/ when present and update benchmark harness to support multiple optimized backends + label encodings.
Add Python optimization spec/backlog docs and remove older MATLAB build-cleanup plan documents.

Reviewed changes

Copilot reviewed 28 out of 31 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
setup_gcmi.m	Adds `IncludeNative` option and auto-path discovery for compiled C++ MEX binaries.
README.md	Documents auto-runtime discovery and how to build/smoke-test the new C++ MEX path.
plans/PYTHON_OPTIMIZED_IMPLEMENTATION_SPEC.md	New design spec for a Python-only optimized path (Numba-first).
plans/PYTHON_OPTIMIZED_IMPLEMENTATION_BACKLOG.md	New execution backlog/tasks for the Python optimization effort.
plans/MATLAB_BUILD_CLEANUP_SPEC.md	Removes prior MATLAB build-cleanup spec document.
plans/MATLAB_BUILD_CLEANUP_BACKLOG.md	Removes prior MATLAB build-cleanup backlog document.
matlab/cpp_mex/tooling/gcmi_cpp_mex_test.m	Adds compile-and-smoke-test runner for the new C++ MEX entrypoints.
matlab/cpp_mex/tooling/gcmi_cpp_mex_package.m	Adds packaging helper to zip compiled artifacts per release/arch.
matlab/cpp_mex/tooling/gcmi_cpp_mex_config.m	Adds compiler/OpenMP configuration discovery for building C++ MEX targets.
matlab/cpp_mex/tooling/gcmi_cpp_mex_compile.m	Adds multi-target MEX build driver + build metadata emission.
matlab/cpp_mex/tooling/gcmi_cpp_mex_clean.m	Adds cleanup helper for build/output directories.
matlab/cpp_mex/tests/run_gcmi_cpp_mex_smoke_tests.m	Adds a minimal test entrypoint for `matlab/cpp_mex` smoke tests.
matlab/cpp_mex/src/info_cc_slice_bridge.cpp	Adds classic-C API bridge for a C-API MEX wrapper target.
matlab/cpp_mex/src/gcmi_kernels.cpp	Implements the core C++ kernels (`info_cc_slice`, `info_cd_slice`) with OpenMP/BLAS.
matlab/cpp_mex/README.md	Documents the new C++ MEX rewrite directory, entrypoints, layouts, and label conventions.
matlab/cpp_mex/mex/info_cd_slice_cpp.cpp	Adds C++ MEX adapter entrypoint for `info_cd_slice_cpp`.
matlab/cpp_mex/mex/info_cc_slice_cpp.cpp	Adds C++ MEX adapter entrypoint for `info_cc_slice_cpp`.
matlab/cpp_mex/mex/info_cc_slice_cpp_capi.c	Adds a classic C-API MEX wrapper variant for `info_cc_slice_cpp`.
matlab/cpp_mex/mex/gcmi_cpp_runtime_probe.cpp	Adds OpenMP runtime probe MEX entrypoint.
matlab/cpp_mex/mex/gcmi_cpp_ping.cpp	Adds build/runtime identity probe MEX entrypoint.
matlab/cpp_mex/mex/gcmi_cpp_omp_probe.cpp	Adds OpenMP probe MEX entrypoint.
matlab/cpp_mex/mex/gcmi_cpp_blas_probe.cpp	Adds BLAS/LAPACK probe MEX entrypoint.
matlab/cpp_mex/include/gcmi_mex_utils.hpp	Adds shared BLAS/LAPACK + utility helpers for the C++ kernels.
matlab/cpp_mex/include/gcmi_mex_adapter_utils.hpp	Adds MATLAB C++ MEX adapter utilities (argument checking, label parsing).
matlab/cpp_mex/include/gcmi_kernels.hpp	Declares the C++ kernel APIs used by MEX entrypoints/bridges.
FUNCTIONALITY.md	Updates repo functionality inventory to include `matlab/cpp_mex` and label conventions.
buildfile.m	Adds MATLAB `buildtool` tasks to compile/test/bench/package the C++ MEX runtime.
benchmarks/run_matlab_benchmarks.m	Adds optional path injection, label encoding support, and optimized-data preparation for alternative backends.
benchmarks/README.md	Documents running benchmarks with the new C++ MEX backend path/label encoding options.
benchmarks/cpp_mex_default_function_map.m	Adds a default function-name map for the new C++ MEX backend.
.gitignore	Ignores `matlab/cpp_mex/bin/` build outputs and local scratch/settings artifacts.

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

Copilot · 2026-04-03T23:17:24Z

+    'lapack_library', cfg.LapackLibrary, ...
+    'targets', {requested}, ...
+    'generated_at', char(datetime('now', TimeZone='local', Format='yyyy-MM-dd''T''HH:mm:ssXXX')));
+fid = fopen(fullfile(cfg.OutputDir, 'build-info.json'), 'w');


fopen() can return -1 on failure (e.g., permissions, missing directory). As written, fclose(-1) in the onCleanup will error and jsonencode output will be lost. Please check fid >= 0 and raise a clear error if the build-info.json cannot be opened for writing.

Suggested change

fid = fopen(fullfile(cfg.OutputDir, 'build-info.json'), 'w');

buildInfoPath = fullfile(cfg.OutputDir, 'build-info.json');

[fid, msg] = fopen(buildInfoPath, 'w');

if fid < 0

error('gcmi_cpp_mex_compile:BuildInfoOpenFailed', ...

'Unable to open "%s" for writing: %s', buildInfoPath, msg);

end

robince · 2026-04-04T10:21:45Z

@codex review

Copilot

Pull request overview

Copilot reviewed 36 out of 39 changed files in this pull request and generated 5 comments.

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

Copilot · 2026-04-04T10:26:40Z

+    try {
+        gcmi::info_cc_slice(
+            x,
+            static_cast<gcmi::mwSize>(ntrl),
+            static_cast<gcmi::mwSize>(xdim),
+            static_cast<gcmi::mwSize>(npage),
+            y,
+            static_cast<gcmi::mwSize>(ydim),
+            static_cast<gcmi::mwSize>(threads),
+            output);
+        *error_message = nullptr;
+    } catch (const std::exception& ex) {
+        *error_message = ex.what();
+    } catch (...) {
+        *error_message = "unknown error";
+    }


gcmi_info_cc_slice_bridge sets *error_message = ex.what(), but the std::exception object is destroyed when leaving the catch block, so the returned pointer can dangle. The caller (info_cc_slice_cpp_capi.c) then uses this pointer in mexErrMsgIdAndTxt, which is undefined behavior and can crash or print garbage. Store the message in storage that outlives the function call (e.g., a static/thread_local std::string whose c_str() is returned, or copy into a caller-owned buffer) before returning the pointer.

Copilot · 2026-04-04T10:26:40Z

+        const auto nTrials = dims[0];
+        const auto nPages = dims[1];
+        auto out = factory_.createArray<double>({nTrials, nPages});
+        auto normalized = gcmi::copnorm_slice_kernel(raw_data(x), nTrials, nPages, threads);
+        if (!normalized.empty()) {
+            std::copy(normalized.begin(), normalized.end(), raw_data(out));
+        }


This MEX wrapper allocates a full std::vector<double> via copnorm_slice_kernel(...) and then copies it into a newly allocated MATLAB array. For large inputs this doubles peak memory and adds an extra pass over the data. Consider refactoring the kernel to write directly into the MATLAB output buffer (or add an overload that takes an output pointer) to avoid the extra allocation/copy.

Copilot · 2026-04-04T10:26:40Z

+function i_rmdir_if_exists(pathname)
+if exist(pathname, 'dir')
+    rmdir(pathname, 's');
+end
+end


i_rmdir_if_exists is defined but never called anywhere in this file, which adds dead code to a small regression harness. Either remove it, or add the intended call site so the helper is exercised (and consider adding a regression check around the behavior it was meant to support).

Copilot · 2026-04-04T10:26:41Z

 repoRoot = fileparts(fileparts(fileparts(mfilename('fullpath'))));
 addpath(repoRoot);
+addpath(fullfile(repoRoot, 'benchmarks'));
 setup_gcmi();



This test runner now unconditionally adds benchmarks/ to the MATLAB path, but the file doesn’t appear to call any benchmark helpers (it defines its own assertions locally). If this path isn’t required, consider removing it to avoid path pollution/shadowing and to keep the regression harness minimal.

Copilot · 2026-04-04T10:26:41Z

+p.addParameter('Compile', true, @(x) islogical(x) && isscalar(x));
+p.addParameter('LegacyPath', fullfile(fileparts(fileparts(fileparts(fileparts(mfilename('fullpath'))))), 'extern', 'gcmi_mex'), @(x) ischar(x) || isstring(x));
+p.parse(varargin{:});


LegacyPath default goes up four directories from matlab/cpp_mex/tooling/gcmi_cpp_mex_test.m, which appears to land above the repository root. Given the repo layout used elsewhere (gcmi_cpp_mex_config goes up three levels to reach RepoRoot), this default likely points to a non-existent extern/gcmi_mex and silently disables legacy cross-checks. Consider deriving the legacy path from cfg.RepoRoot (or only walking up three levels) so the default matches the actual repo root.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9c2db91abb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-04T10:27:05Z

+
+        const auto xDims = x.getDimensions();
+        const auto yDims = y.getDimensions();
+        if (xDims.size() != 3 || xDims[0] != ntrl || xDims[1] != xdim) {


Accept single-page inputs in info_cc_slice_cpp

The shape guard hard-requires xDims.size() == 3, but MATLAB collapses trailing singleton dimensions, so valid one-page input [Ntrl, Xdim, 1] is commonly presented as a 2-D array [Ntrl, Xdim]. In that case this now throws info_cc_slice_cpp:shape instead of computing a single output page, which breaks single-slice workflows. Please treat 2-D input as Npage = 1 here.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-04T10:27:05Z

+        const auto threads = std::max<gcmi::mwSize>(1, scalar_to_size(inputs[5], "Nthread"));
+
+        const auto xDims = x.getDimensions();
+        if (xDims.size() != 3 || xDims[0] != xdim || xDims[1] != ntrl) {


Accept single-page inputs in info_cd_slice_cpp

This check similarly requires xDims.size() == 3, so a valid single-page array [Xdim, Ntrl, 1] that MATLAB represents as 2-D [Xdim, Ntrl] is rejected with info_cd_slice_cpp:shape. That makes the new discrete MEX path unusable for Npage == 1 calls even though the underlying kernel can operate on one page.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-04T10:27:05Z

+            output);
+        *error_message = nullptr;
+    } catch (const std::exception& ex) {
+        *error_message = ex.what();


Preserve exception message lifetime in C bridge

The bridge writes ex.what() into *error_message and returns, but that pointer is owned by the exception object and becomes invalid when the catch block exits. The C API wrapper then reads that dangling pointer in mexErrMsgIdAndTxt, which can produce corrupted messages or undefined behavior on runtime-error paths. Store a durable copy of the message (or emit the MATLAB error before returning) instead.

Useful? React with 👍 / 👎.

robince added 2 commits April 4, 2026 00:05

spec files

8ab55ef

Port fortran gcmi mex to cpp

8ce56c1

robince requested a review from Copilot April 3, 2026 23:12

Copilot started reviewing on behalf of robince April 3, 2026 23:12

View session

fix build failure

934a12f

Copilot AI reviewed Apr 3, 2026

View reviewed changes

robince added 7 commits April 4, 2026 09:22

add a github actions build

52f9f65

update readme and examples

0c5fc81

address review comments

0b73f4a

add linux release

eff1fbd

add numba test

155aca1

remove compiled mex from regression tests

45c60ab

add cpp mex openmp copnorm

3a0f240

robince force-pushed the matlab-mex-cpp branch from d01c06c to 3a0f240 Compare April 4, 2026 09:30

robince added 6 commits April 4, 2026 10:36

add branch release to check linux build

c01bab1

Remove macOS MATLAB release job and use Ubuntu 22.04

b52e8c0

try to get matlab-release working

5d547a7

Fix MATLAB release artifact path and drop macOS job

61c6014

Restore macOS MATLAB release job and fix artifact globs

cba8dc0

Update MATLAB release workflow action versions

9c2db91

robince requested a review from Copilot April 4, 2026 10:21

Copilot started reviewing on behalf of robince April 4, 2026 10:22

View session

Copilot AI reviewed Apr 4, 2026

View reviewed changes

chatgpt-codex-connector Bot reviewed Apr 4, 2026

View reviewed changes

robince added 4 commits April 4, 2026 11:34

Add small-kernel Numba path for info_cc_slice

553c6cc

address review comments

8760fe3

Write MEX kernel outputs directly into MATLAB buffers

c02e4d2

Merge branch 'numba-small-n-kernels' into matlab-mex-cpp

40c91e8

robince added 2 commits April 4, 2026 12:09

only build release on tag

fe19bd2

Fix legacy benchmark shims and drop unused Numba work

7b1ba1f

robince merged commit 8713b1d into master Apr 4, 2026
8 checks passed

-fid = fopen(fullfile(cfg.OutputDir, 'build-info.json'), 'w');
+buildInfoPath = fullfile(cfg.OutputDir, 'build-info.json');
+[fid, msg] = fopen(buildInfoPath, 'w');
+if fid < 0
+    error('gcmi_cpp_mex_compile:BuildInfoOpenFailed', ...
+        'Unable to open "%s" for writing: %s', buildInfoPath, msg);
+end

Conversation

robince commented Apr 3, 2026

Uh oh!

robince commented Apr 3, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI Apr 3, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

robince commented Apr 4, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants