ci: release-only builds, remove benchmarks from CI

.github/workflows/ci.yml

@@ -1,18 +1,29 @@
-name: C++ CI with Google Test
+name: CI
 
 on:
-  push:
-    branches: [ main ]
+  # Fast check on PRs — build + unit tests only (no benchmarks)
   pull_request:
     branches: [ main ]
 
+  # Release builds: triggered by version tags (v1.0.0, v2.3.1, etc.)
+  push:
+    tags:
+      - 'v*'
+
 jobs:
+  # ---------------------------------------------------------------
+  # Quick CI gate — runs on every PR to main
+  # Builds the library and runs only fast unit tests (<2 min total).
+  # Benchmarks and long-running tests are SKIPPED here because
+  # GitHub's free runners are too slow / unreliable for them.
+  # See TESTING.md for the full rationale.
+  # ---------------------------------------------------------------
   build-and-test:
     runs-on: ubuntu-latest
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4
 
     - name: Set up C++ environment
       run: |
@@ -26,14 +37,65 @@ jobs:
         sudo make
         sudo cp lib/*.a /usr/lib
 
-    - name: Create build directory
-      run: mkdir -p build
-
     - name: Configure CMake
-      run: cmake -S . -B build
+      run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
 
     - name: Build the project
-      run: cmake --build build
+      run: cmake --build build -j$(nproc)
+
+    - name: Run unit tests (benchmarks excluded)
+      run: |
+        cd build
+        ctest --output-on-failure --timeout 120 -L unit 2>/dev/null || \
+        ctest --output-on-failure --timeout 120
+        echo "See TESTING.md for info on skipped/disabled tests"
+
+  # ---------------------------------------------------------------
+  # Release job — only runs when a version tag is pushed.
+  # Builds an optimised binary and uploads it as a GitHub Release.
+  # ---------------------------------------------------------------
+  release:
+    if: startsWith(github.ref, 'refs/tags/v')
+    needs: build-and-test
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v4
+
+    - name: Set up C++ environment
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y build-essential cmake g++ libgtest-dev
+
+    - name: Build and install Google Test
+      run: |
+        cd /usr/src/gtest
+        sudo cmake .
+        sudo make
+        sudo cp lib/*.a /usr/lib
+
+    - name: Build release binaries
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
+        cmake --build build -j$(nproc)
+
+    - name: Package binaries
+      run: |
+        mkdir -p release
+        # Copy library
+        cp build/libTinyML.a release/
+        # Copy test/benchmark executables (if they exist)
+        find build/bin -type f -executable -exec cp {} release/ \; 2>/dev/null || true
+        find build -maxdepth 1 -type f -executable -exec cp {} release/ \; 2>/dev/null || true
+        # Create tarball
+        tar -czf tinyml-${{ github.ref_name }}-linux-x86_64.tar.gz -C release .
+        ls -lh tinyml-${{ github.ref_name }}-linux-x86_64.tar.gz
 
-    - name: Run tests
-      run: cd build && ctest --output-on-failure --timeout 120
+    - name: Create GitHub Release
+      uses: softprops/action-gh-release@v2
+      with:
+        files: tinyml-${{ github.ref_name }}-linux-x86_64.tar.gz
+        generate_release_notes: true

CMakeLists.txt

@@ -328,6 +328,15 @@ else()
     endif()
 endif()
 
+# =============================================================================
+# BENCHMARKS
+# These are built but NOT registered as CTest tests so they don't run in CI.
+# Free GitHub Actions runners are too slow/unreliable for benchmarks — results
+# are meaningless on shared hardware and long runs waste CI minutes.
+# Run benchmarks locally:  cd build && ./bin/SIMDBenchmark
+# See TESTING.md for full details.
+# =============================================================================
+
 # SIMD Benchmark executable (using XSIMD)
 add_executable(SIMDBenchmark benchmarks/benchmark_xsimd.cpp)
 target_link_libraries(SIMDBenchmark TinyML gtest gtest_main)
@@ -340,7 +349,7 @@ else()
         target_include_directories(SIMDBenchmark PRIVATE "${CMAKE_BINARY_DIR}/_deps/xsimd-src/include")
     endif()
 endif()
-add_test(NAME SIMDBenchmark COMMAND SIMDBenchmark)
+# Not registered as ctest — run locally only
 
 # Physics-Informed Neural Networks Benchmark executable
 add_executable(PhysicsBenchmark EXCLUDE_FROM_ALL benchmarks/benchmark_physics.cpp)
@@ -365,12 +374,12 @@ add_test(NAME LightweightAttentionTest COMMAND LightweightAttentionTest)
 add_executable(AttentionBenchmark benchmarks/benchmark_attention.cpp)
 target_link_libraries(AttentionBenchmark TinyML gtest gtest_main)
 target_include_directories(AttentionBenchmark PUBLIC ${PROJECT_SOURCE_DIR}/include)
-add_test(NAME AttentionBenchmark COMMAND AttentionBenchmark)
+# Not registered as ctest — run locally only
 
 # Simple Attention Benchmark executable (standalone)
 add_executable(SimpleAttentionBenchmark benchmarks/benchmark_simple_attention.cpp)
 target_include_directories(SimpleAttentionBenchmark PUBLIC ${PROJECT_SOURCE_DIR}/include)
-add_test(NAME SimpleAttentionBenchmark COMMAND SimpleAttentionBenchmark)
+# Not registered as ctest — run locally only
 
 # Advanced Optimizations Benchmark executable - disabled until
 # AdvancedOptimizations.cpp is fixed (missing member fields, headers)
@@ -418,7 +427,7 @@ else()
         target_include_directories(ReinforcementLearningBenchmark PRIVATE "${CMAKE_BINARY_DIR}/_deps/xsimd-src/include")
     endif()
 endif()
-add_test(NAME ReinforcementLearningBenchmark COMMAND ReinforcementLearningBenchmark)
+# Not registered as ctest — segfaults on CI, run locally only
 
 # Phase 11 Generative Models Test executable
 if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/tests/test_phase11_generative.cpp)
@@ -449,7 +458,7 @@ if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/benchmarks/benchmark_generative.cpp)
             target_include_directories(GenerativeModelsBenchmark PRIVATE "${CMAKE_BINARY_DIR}/_deps/xsimd-src/include")
         endif()
     endif()
-    add_test(NAME GenerativeModelsBenchmark COMMAND GenerativeModelsBenchmark)
+    # Not registered as ctest — run locally only
 endif()
 
 # Debugging: Print the include directories that will be passed to the compiler

TESTING.md

@@ -0,0 +1,72 @@
+# Testing & CI Policy
+
+## What runs in CI
+
+CI runs on **pull requests to `main`** and **release tags** (`v*`).
+Only fast unit tests execute in CI — the full suite finishes in under 2 minutes.
+
+```
+ctest --output-on-failure --timeout 120
+```
+
+## What does NOT run in CI (and why)
+
+### Benchmarks (removed from ctest)
+
+| Benchmark | Why skipped |
+|---|---|
+| `SIMDBenchmark` | Benchmark results on shared CI runners are meaningless — hardware varies per run |
+| `AttentionBenchmark` | Same reason — timing-sensitive, needs dedicated hardware |
+| `SimpleAttentionBenchmark` | Same reason |
+| `ReinforcementLearningBenchmark` | Segfaults on Linux CI runners (works locally on macOS) |
+| `GenerativeModelsBenchmark` | Benchmark — not a correctness test |
+
+**Benchmarks are still built** so compilation is verified. They just aren't registered
+with `add_test()` so `ctest` won't run them. Run them locally:
+
+```bash
+cd build
+./bin/SIMDBenchmark
+./bin/AttentionBenchmark
+./bin/SimpleAttentionBenchmark
+./bin/ReinforcementLearningBenchmark
+./bin/GenerativeModelsBenchmark
+```
+
+### Disabled tests (registered but skipped)
+
+| Test | Why disabled |
+|---|---|
+| `Phase8PhysicsInformedTest` | Takes **16+ minutes** on CI runners — too slow for free GitHub Actions |
+| `Phase8PhysicsComprehensiveTest` | Non-deterministic convergence — `loss_ratio` swings from 0.04 to 21+ across runs |
+| `Phase12ReinforcementTest` | Segfaults on Linux CI runners |
+
+Run these locally if you need them:
+
+```bash
+cd build
+ctest -R Phase8PhysicsInformedTest --force-new-ctest-process
+ctest -R Phase12ReinforcementTest --force-new-ctest-process
+```
+
+## Releases
+
+Pushing a version tag (e.g. `git tag v1.0.0 && git push --tags`) triggers:
+
+1. Full build + unit tests
+2. Release binary packaging (`libTinyML.a` + executables)
+3. Upload to GitHub Releases
+
+## Running the full suite locally
+
+```bash
+mkdir -p build && cd build
+cmake .. -DCMAKE_BUILD_TYPE=Release
+cmake --build . -j$(nproc)
+
+# Fast unit tests only (what CI runs)
+ctest --output-on-failure --timeout 120
+
+# Everything including disabled tests
+ctest --output-on-failure --timeout 1200 --force-new-ctest-process
+```