Codex/release 0.4.0 validation

.bazelignore

@@ -0,0 +1,12 @@
+# Local experiments and unused vendor trees; excluded from //... discovery.
+share/google_benchmark
+share/benchmarks
+third_party/cblas
+third_party/csparse
+third_party/hdf5
+third_party/jburkardt
+third_party/matmul
+third_party/mesch12b
+third_party/mm_offloading
+third_party/spmatrix
+third_party/spblas

.bazelrc

@@ -135,10 +135,7 @@ build:openmp_only --define=MATRIX_BACKEND=openmp
 # macOS defaults: OpenMP + Accelerate. MPS stays available as an optional
 # runtime backend via sm_set_backend(SM_BACKEND_MPS).
 build:macos --config=accelerate
-build:macos --repo_env=SUITESPARSE_ROOT=/opt/homebrew/opt/suite-sparse
 
 # Linux defaults: OpenMP + OpenBLAS.
 build:linux --config=openblas
-build:linux --repo_env=OPENBLAS_ROOT=/home/linuxbrew/.linuxbrew/opt/openblas
-build:linux --repo_env=SUITESPARSE_ROOT=/home/linuxbrew/.linuxbrew/opt/suite-sparse
 

.bazelversion

@@ -0,0 +1 @@
+9.3.0

.clangd

@@ -1,23 +1,5 @@
 CompileFlags:
   Add:
-    # matio Include paths
-    - "-Ithird_party/matio"
-    - "-Ibazel-out/darwin_arm64-fastbuild/bin/external/+http_archive+matio"
-    - "-Ibazel-out/darwin_arm64-fastbuild/bin/external/+http_archive+matio/src"
-    - "-Iexternal/+http_archive+matio"
-    - "-Iexternal/+http_archive+matio/src"
-    # matrix public headers
     - "-Iapp/matrix/include"
-    # third-party benchmark dependencies used by share/google_benchmark
-    - "-Ithird_party/matmul/include"
-    - "-Ithird_party/mm_offloading"
-    # bazel external repo symlink for google benchmark headers
-    - "-Iexternal/google_benchmark+/include"
-
----
-If:
-  PathMatch: ^share/google_benchmark/benchmarks/legacy/.*
-Diagnostics:
-  Suppress:
-    - pp_file_not_found
-    - unused-includes
+    - "-Iapp/vector/include"
+    - "-Iapp/tensor/include"

.github/workflows/ci.yml

@@ -2,23 +2,50 @@ name: BUILD
 
 on:
   push:
-    branches:
-      - master
+    branches: [master]
   pull_request:
-    branches:
-      - master
+    branches: [master]
+  workflow_dispatch:
 
-jobs:
-    test:
-     # virtual environments: https://github.com/actions/virtual-environments
-      runs-on: macos-latest
-
-      steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-
-      - name: Test the code (unit tests)
-        run: bazel test //tests/...
+permissions:
+  contents: read
 
-      - name: Upload coverage reports to Codecov
-        uses: codecov/codecov-action@v3
+jobs:
+  test:
+    name: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [macos-latest, ubuntu-24.04]
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 45
+    env:
+      OPENBLAS_NUM_THREADS: "1"
+      OMP_NUM_THREADS: "2"
+    steps:
+      - uses: actions/checkout@v7
+      - uses: bazel-contrib/setup-bazel@0.19.1
+        with:
+          bazelisk-cache: true
+          disk-cache: true
+          repository-cache: true
+      - name: Prepare system OpenBLAS and LAPACKE
+        if: runner.os == 'Linux'
+        shell: bash
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libopenblas-pthread-dev liblapacke-dev gfortran ruby
+          prefix="$RUNNER_TEMP/mmatrix-openblas"
+          multiarch="$(gcc -print-multiarch)"
+          mkdir -p "$prefix/include" "$prefix/lib"
+          for header in /usr/include/"$multiarch"/openblas-pthread/*.h /usr/include/lapack*.h; do
+            ln -s "$header" "$prefix/include/$(basename "$header")"
+          done
+          ln -s /usr/lib/"$multiarch"/openblas-pthread/libopenblas.a "$prefix/lib/libopenblas.a"
+          ln -s /usr/lib/"$multiarch"/liblapacke.a "$prefix/lib/liblapacke.a"
+          echo "OPENBLAS_ROOT=$prefix" >> "$GITHUB_ENV"
+      - name: Build and test all published targets
+        run: bazel test //... --test_output=errors --color=no --curses=no
+      - name: Verify installed libraries
+        shell: bash
+        run: bash tools/ci/test_install.sh

.gitignore

@@ -79,10 +79,7 @@ Temporary Items
 /bazel-*
 /bazel.iml
 /output/
-.bazelversion
-.bazelignore
 user.bazelrc
-MODULE.bazel.lock
 
 #-------------------------------------------------------------------------------
 # Temporary files
@@ -108,6 +105,7 @@ MODULE.bazel.lock
 /share/simple_benchmark/*
 !/share/simple_benchmark/BUILD
 !/share/simple_benchmark/bench_sm_mps.c
+!/share/simple_benchmark/simple_bench.h
 /lib/*
 /_archive/*
 /archive/*
@@ -129,19 +127,21 @@ MODULE.bazel.lock
 /third_party/cblas
 
 #-------------------------------------------------------------------------------
-*.in
 .clang-tidy
-.clangd
 .envrc
 compile_commands.json
 *.png
 StyleGuide.md
 
-MODULE.bazel.lock
 profile.gz
 
 /.venv-bench/
 .github/skills/api-return-conventions/SKILL.md
 TENSOR_ARCHITECTURE_ANALYSIS.md
 TENSOR_PATTERNS_QUICK_REF.md
 app/tensor/docs/mps-performance.md
+
+# Generated benchmark and documentation output
+*.profraw
+*.profdata
+/site/

BUILD

@@ -4,24 +4,22 @@ load("@mkdocs//:defs.bzl", "mkdocs_build", "mkdocs_config", "mkdocs_serve")
 load("@rules_docs//docgen:defs.bzl", "docs")
 load("//tools/install:def.bzl", "installer")
 
-exports_files(["compile_commands.json"])
-
 installer(
     name = "matrix_installer",
     data = [
-        "//:LICENSE.txt",  # if available
-        "//app/matrix",  # the target to be installed
-        "//app/matrix:matrix_header",  # must be collected in a filegroup
+        "//:LICENSE.txt",
+        "//app/matrix",
+        "//app/matrix:matrix_header",
         "//app/vector",
         "//app/vector:vector_header",
-        "@log",
-        "@matio",
-        "@openblas//:install_files",
-        "@openmp//:libomp",
-        "@pcg",
-        "@suitesparse//:install_files",
-        "@zlib",
-    ],
+    ] + select({
+        "//app/matrix:macos": [
+            "//app/matrix:sm_mps",
+            "//app/tensor:st",
+            "//app/tensor:tensor_header",
+        ],
+        "//conditions:default": [],
+    }),
     system_integration = False,  # set "True" symlink to /usr/local/lib and /usr/local/include
 )
 
@@ -43,6 +41,7 @@ alias(
 docs(
     name = "docs_src",
     srcs = glob(["doc/**/*.md"]) + [
+        "CHANGELOG.md",
         "//doc/examples:tensor_basic.md",
     ],
     readme_content = "MMatrix Dokumentation.",

CHANGELOG.md

@@ -0,0 +1,36 @@
+# Changelog
+
+## 0.4.0 (unreleased)
+
+Changes since `v0.3.1`.
+
+### Added
+
+- Vector modules for float/double, strided views, sparse vectors, geometry, and vector/tensor bridges.
+- Flat-buffer SGD, momentum, RMSProp, Adam, and gradient norm clipping kernels.
+- Shared NN activation/backward kernels, fused GEMM + bias + activation, and row/loss kernels.
+- Tensor storage abstraction, Metal buffer support, and f32/bf16 conversion utilities.
+- Interpreter kernels: broadcast binary arithmetic, LayerNorm, ReduceMax, batched matmul, gather, and materializing permute.
+- IEEE binary16 scalar/bulk conversion helpers.
+- GPU-resident matrix GEMM streams/plans and reproducible benchmark sources.
+
+### Changed / migration
+
+- SuiteSparse is no longer a library or build dependency. Sparse matrices continue to use COO with a lazy CSR row cache.
+- Removed `dms_to_cs`, `dms_from_cs`, and `cs_to_dms`, and the CSparse cache fields in `DoubleSparseMatrix`. Code accessing those APIs/fields must migrate; recompile consumers.
+- Matrix/tensor conversions now live in `st_convert.h` and `//app/tensor:st`, rather than `m_convert.h`. `st_from_sm` creates a rank-2 `[rows, cols]` tensor; `sm_from_st` requires a rank-2 f32 tensor.
+- Tensor implementation files are grouped into core, backend, runtime, and operation directories. Depend on public Bazel targets rather than source paths.
+- Bazel version and module lockfile are included for reproducible dependency resolution. Linux OpenBLAS remains a system installation.
+
+### Fixed during release preparation
+
+- Produce ordinary Linux object archives rather than forcing LLVM LTO; check system-linker consumption and choose the platform archive bundler.
+- Link separately packaged LAPACKE on Linux and preserve NN-kernel NaN/thread semantics under optimized builds.
+
+- Exercise actual OpenMP branches in interpreter thread-comparison tests and remove an unmatched floating-point pragma.
+- Reject size overflows before gather/permute byte copies; validate permutation indices before shape access; preserve arbitrary unaligned byte payloads.
+- Remove stale SuiteSparse and local HDF5 wiring, obsolete vendor build wrappers, and benchmark targets whose sources were not published.
+- Include the MPS benchmark helper header in the source release and exclude local experiments from Bazel discovery.
+- Install transitive library archives and required external headers; include tensor artifacts on macOS; use portable macOS/Linux install flags.
+- Add macOS/Ubuntu CI coverage with standard/bundle installation consumer checks.
+- Repair CI test discovery and document the platform boundaries and new API entrypoints.

MODULE.bazel

@@ -13,7 +13,7 @@ For more details, please check https://github.com/bazelbuild/bazel/issues/18958
 
 module(
     name = "mmatrix",
-    version = "0.1.0",
+    version = "0.4.0",
 )
 
 ######################################################################
@@ -22,10 +22,9 @@ module(
 
 bazel_dep(name = "platforms", version = "1.0.0")
 bazel_dep(name = "zlib", version = "1.3.1.bcr.5")
-bazel_dep(name = "apple_support", version = "1.24.2")
+bazel_dep(name = "apple_support", version = "2.5.2")
 bazel_dep(name = "bazel_skylib", version = "1.8.2")
 bazel_dep(name = "rules_shell", version = "0.6.1")
-bazel_dep(name = "google_benchmark", version = "1.9.2")
 bazel_dep(name = "rules_cc", version = "0.2.17")
 bazel_dep(name = "rules_python", version = "1.7.0")
 bazel_dep(name = "rules_docs", version = "0.3.0")
@@ -44,7 +43,7 @@ pip = use_extension("@rules_python//python/extensions:pip.bzl", "pip")
 pip.parse(
     hub_name = "pypi",
     python_version = "3.11",
-    requirements_lock = "//doc:requirements.txt",
+    requirements_lock = "//:doc/requirements.txt",
 )
 use_repo(pip, "pypi")
 
@@ -91,25 +90,7 @@ bazel_dep(name = "wolfd_bazel_compile_commands", version = "0.5.2", dev_dependen
 
 host_local_library = use_repo_rule("//tools:host_local_library.bzl", "host_local_library")
 
-#############################
-# Hombrew Local Libraries (MacOS)
-#############################
-
-# define new_local_repository rule for local installed non-Bazel libraries:
-new_local_repository = use_repo_rule("@bazel_tools//tools/build_defs/repo:local.bzl", "new_local_repository")
-
-# due to license issues in private lib folder.
-new_local_repository(
-    name = "hdf5",
-    path = "third_party/hdf5",
-)
-
-new_local_repository(
-    name = "log",
-    build_file = "@//:third_party/log.c/log.c.BUILD",
-    path = "third_party/log.c",
-)
-
+# System OpenBLAS (only fetched when the OpenBLAS backend is selected).
 # openblas is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
 host_local_library(
     name = "openblas",
@@ -122,16 +103,3 @@ host_local_library(
         "lib",
     ],
 )
-
-# Optional Homebrew SuiteSparse wrapper for the GraphBLAS benchmark only.
-host_local_library(
-    name = "suitesparse_graphblas_host",
-    build_file = "@//:third_party/suitesparse.BUILD",
-    env_var = "SUITESPARSE_ROOT",
-    linux_path = "/home/linuxbrew/.linuxbrew/opt/suite-sparse",
-    macos_path = "/opt/homebrew/opt/suite-sparse",
-    required_subdirs = [
-        "include",
-        "lib",
-    ],
-)