name: CI

on:
  push:
    branches: [main, master]
    paths:
      - '**/*.cpp'
      - '**/*.h'
      - '**/*.hpp'
      - '**/CMakeLists.txt'
      - '.github/workflows/**'
  pull_request:
    types: [opened, synchronize, reopened]
  workflow_dispatch:

concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

env:
  # HuggingFace model info
  HF_MODEL_REPO: GaboxR67/MelBandRoformers
  HF_CHECKPOINT_PATH: melbandroformers/vocals/voc_fv6.ckpt
  HF_CONFIG_PATH: melbandroformers/vocals/voc_gabox.yaml
  # Music-Source-Separation-Training repo
  MSST_REPO: https://github.com/ZFTurbo/Music-Source-Separation-Training.git

jobs:
  # ===========================================================================
  # Prepare: Generate test data (runs once, shared via artifacts)
  # ===========================================================================
  prepare-test-data:
    runs-on: ubuntu-latest
    
    steps:
      - name: Checkout
        uses: actions/checkout@v4
      
      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
          
      - name: Clone MSST Repository
        run: git clone --depth 1 ${{ env.MSST_REPO }} msst
        
      - name: Install Dependencies
        run: |
          pip install torch torchaudio --index-url https://download.pytorch.org/whl/cpu
          pip install huggingface_hub scipy soundfile gguf librosa ml_collections einops pyyaml numpy tqdm beartype rotary_embedding_torch
          
      - name: Download Model from HuggingFace
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          python -c "
          from huggingface_hub import hf_hub_download
          import os
          token = os.environ.get('HF_TOKEN') or None
          hf_hub_download('${{ env.HF_MODEL_REPO }}', '${{ env.HF_CHECKPOINT_PATH }}', 
                          local_dir='./model', token=token)
          hf_hub_download('${{ env.HF_MODEL_REPO }}', '${{ env.HF_CONFIG_PATH }}',
                          local_dir='./model', token=token)
          "
          
      - name: Generate Test Audio
        run: |
          python scripts/generate_test_audio.py --output test_audio.wav --duration 5.0 --sample-rate 44100
          
      - name: Generate Test Data
        run: |
          python scripts/generate_test_data.py \
            --model-repo msst \
            --audio test_audio.wav \
            --checkpoint model/${{ env.HF_CHECKPOINT_PATH }} \
            --config model/${{ env.HF_CONFIG_PATH }} \
            --output test_data
            
      - name: Convert Model to GGUF
        run: |
          python scripts/convert_to_gguf.py \
            --ckpt model/${{ env.HF_CHECKPOINT_PATH }} \
            --config model/${{ env.HF_CONFIG_PATH }} \
            --out model.gguf \
            --dtype fp32
            
      - name: Upload Test Data Artifact
        uses: actions/upload-artifact@v4
        with:
          name: test-data
          path: |
            test_data/
            model.gguf
            test_audio.wav
          retention-days: 1

  # ===========================================================================
  # Build Matrix: Core Platforms + Vulkan
  # ===========================================================================
  build:
    needs: prepare-test-data
    strategy:
      fail-fast: false
      matrix:
        include:
          # Tier 1: Core Platforms (CPU)
          - { name: linux-x64-cpu,    os: ubuntu-22.04,     backend: cpu,    test: true  }
          - { name: linux-arm64-cpu,  os: ubuntu-22.04-arm, backend: cpu,    test: true  }
          - { name: macos-arm64,      os: macos-latest,     backend: cpu,    test: true  }
          - { name: macos-x64,        os: macos-15-intel,   backend: cpu,    test: true  }
          - { name: windows-x64-msvc, os: windows-2025,     backend: cpu,    test: true  }
          # Tier 2: Vulkan Backend
          - { name: linux-vulkan,     os: ubuntu-24.04,     backend: vulkan, test: true  }
          - { name: windows-vulkan,   os: windows-2025,     backend: vulkan, test: true  }
          
    runs-on: ${{ matrix.os }}
    
    steps:
      - name: Checkout
        uses: actions/checkout@v4
      
      - name: Clone GGML
        run: git clone --depth 1 https://github.com/ggerganov/ggml.git ggml
      
      - name: Download Test Data
        uses: actions/download-artifact@v4
        with:
          name: test-data
          
      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
          
      - name: Setup MSVC
        if: runner.os == 'Windows'
        uses: ilammy/msvc-dev-cmd@v1

      - name: Install Python Dependencies
        run: pip install numpy scipy
          
      # ----- Linux Dependencies -----
      - name: Install Dependencies (Linux)
        if: runner.os == 'Linux'
        run: |
          sudo apt-get update
          sudo apt-get install -y build-essential cmake
          
      - name: Install Vulkan SDK (Linux)
        if: matrix.backend == 'vulkan' && runner.os == 'Linux'
        run: |
          sudo apt-get install -y libvulkan-dev glslc mesa-vulkan-drivers
          
      # ----- macOS Dependencies -----  
      - name: Install Dependencies (macOS)
        if: runner.os == 'macOS'
        run: brew install cmake
        
      # ----- Windows Dependencies -----
      - name: Install Dependencies (Windows)
        if: runner.os == 'Windows'
        run: choco install ninja -y
        
      - name: Install Vulkan SDK (Windows)
        if: matrix.backend == 'vulkan' && runner.os == 'Windows'
        run: |
          $VK_VERSION = "1.4.313.2"
          curl.exe -o VulkanSDK.exe -L "https://sdk.lunarg.com/sdk/download/${VK_VERSION}/windows/vulkansdk-windows-X64-${VK_VERSION}.exe"
          Start-Process -FilePath .\VulkanSDK.exe -ArgumentList "--accept-licenses --default-answer --confirm-command install" -Wait
          Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${VK_VERSION}"
          Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${VK_VERSION}\bin"
          
      # ----- Configure -----
      - name: Configure (Unix)
        if: runner.os != 'Windows'
        run: |
          cmake -B build \
            -DCMAKE_BUILD_TYPE=Release \
            -DGGML_DIR=ggml \
            -DGGML_CUDA=OFF \
            -DGGML_VULKAN=${{ matrix.backend == 'vulkan' && 'ON' || 'OFF' }} \
            -DMBR_BUILD_TESTS=ON \
            -DMBR_BUILD_CLI=ON
            
      - name: Configure (Windows)
        if: runner.os == 'Windows'
        run: |
          cmake -B build -G "Ninja Multi-Config" `
            -DGGML_DIR=ggml `
            -DGGML_CUDA=OFF `
            -DGGML_VULKAN=${{ matrix.backend == 'vulkan' && 'ON' || 'OFF' }} `
            -DMBR_BUILD_TESTS=ON `
            -DMBR_BUILD_CLI=ON
            
      # ----- Build -----
      - name: Build (Unix)
        if: runner.os != 'Windows'
        run: cmake --build build --config Release -j $(nproc 2>/dev/null || sysctl -n hw.logicalcpu)
        
      - name: Build (Windows)
        if: runner.os == 'Windows'
        run: cmake --build build --config Release -j $env:NUMBER_OF_PROCESSORS
        
      # ----- Unit Tests -----
      - name: Run Unit Tests
        if: matrix.test
        env:
          MBR_MODEL_PATH: ${{ github.workspace }}/model.gguf
          MBR_TEST_DATA_DIR: ${{ github.workspace }}/test_data
          MBR_FORCE_CPU: ${{ runner.os == 'macOS' && '1' || '' }}
        run: ctest --test-dir build -C Release -V --output-on-failure --timeout 300
        
      # ----- CLI Tests -----
      - name: Test CLI
        if: matrix.test
        shell: bash
        env:
          MBR_MODEL_PATH: ${{ github.workspace }}/model.gguf
          MBR_FORCE_CPU: ${{ runner.os == 'macOS' && '1' || '' }}
        run: |
          echo "=== CLI Test Suite ==="
          
          # Dynamically find CLI executable
          CLI_NAME="mel_band_roformer-cli"
          if [[ "$RUNNER_OS" == "Windows" ]]; then CLI_NAME="mel_band_roformer-cli.exe"; fi
          
          echo "Searching for $CLI_NAME..."
          CLI_PATH=$(find build -name "$CLI_NAME" | head -n 1)
          
          if [[ -z "$CLI_PATH" ]]; then
            echo "Error: CLI executable not found!"
            find build
            exit 1
          fi
          
          echo "Found CLI at: $CLI_PATH"
          chmod +x "$CLI_PATH"
          
          # Setup execution variables
          CLI_DIR=$(dirname "$CLI_PATH")
          CLI_EXE="./$CLI_NAME"
          
          # Debug Info
          echo "Debug: Listing directory $CLI_DIR"
          ls -l "$CLI_DIR"
          
          echo "Debug: Checking dependencies"
          if command -v ldd >/dev/null; then
             ldd "$CLI_PATH" || echo "ldd returned error"
          fi
          
          # Define runner helper
          run_cli() {
              echo "Executing: ./$CLI_NAME $@"
              (cd "$CLI_DIR" && ./$CLI_NAME "$@")
          }
          
          # Debug PATH
          echo "PATH: $PATH"
          
          # 1. Test --help
          echo "[1/4] Testing --help..."
          run_cli --help
          
          # 2. Test with missing arguments (should fail)
          echo "[2/4] Testing error handling..."
          if run_cli 2>/dev/null; then
            echo "ERROR: CLI should fail without arguments"
            exit 1
          fi
          
          # 3. Use existing test audio
          echo "[3/4] Using generated test audio..."
          cp test_audio.wav cli_test_input.wav
          
          # 4. Run full inference
          echo "[4/4] Running inference..."
          # Use absolute paths for input/output to avoid directory issues
          ABS_MODEL=$(readlink -f "$MBR_MODEL_PATH" || echo "$(pwd)/model.gguf")
          ABS_INPUT=$(readlink -f cli_test_input.wav || echo "$(pwd)/cli_test_input.wav")
          ABS_OUTPUT=$(readlink -f cli_test_output.wav || echo "$(pwd)/cli_test_output.wav")
          
          # Convert paths for Windows MSVC executables
          if [[ "$RUNNER_OS" == "Windows" ]]; then
              ABS_MODEL=$(cygpath -w "$ABS_MODEL")
              ABS_INPUT=$(cygpath -w "$ABS_INPUT")
              ABS_OUTPUT=$(cygpath -w "$ABS_OUTPUT")
          fi
          
          echo "Running with model: $ABS_MODEL"
          run_cli "$ABS_MODEL" "$ABS_INPUT" "$ABS_OUTPUT" --chunk-size 88200 --overlap 2
          
          # Verify output exists and has reasonable size
          if [[ ! -f cli_test_output.wav ]]; then
            echo "ERROR: Output file not created"
            exit 1
          fi
          
          OUTPUT_SIZE=$(stat -c%s cli_test_output.wav 2>/dev/null || stat -f%z cli_test_output.wav)
          if [[ $OUTPUT_SIZE -lt 1000 ]]; then
            echo "ERROR: Output file too small: $OUTPUT_SIZE bytes"
            exit 1
          fi
          
          echo "=== CLI Tests Passed ==="
          
      # ----- Upload Artifacts -----
      - name: Upload Build Artifacts
        uses: actions/upload-artifact@v4
        with:
          name: build-${{ matrix.name }}
          path: |
            build/bin/
            build/lib*/
            build/*.dll
            build/*.so
            build/*.dylib
            build/mel_band_roformer-cli*
            build/Release/
          retention-days: 7
      
      # ----- Prepare Release Artifact -----
      - name: Prepare Release Artifact (Unix)
        if: runner.os != 'Windows'
        shell: bash
        run: |
          # Create release directory
          mkdir -p release/mel-band-roformer
          
          # Find and copy CLI executable
          CLI_PATH=$(find build -name "mel_band_roformer-cli" -type f | head -n 1)
          if [[ -n "$CLI_PATH" ]]; then
            cp "$CLI_PATH" release/mel-band-roformer/
            chmod +x release/mel-band-roformer/mel_band_roformer-cli
          fi
          
          # Copy shared libraries if exist (only real files, not symlinks)
          find build \( -name "*.so*" -o -name "*.dylib" \) -type f ! -type l | while read lib; do
            cp "$lib" release/mel-band-roformer/ 2>/dev/null || true
          done
          
          # List contents
          echo "Release artifact contents:"
          ls -lh release/mel-band-roformer/
      
      - name: Prepare Release Artifact (Windows)
        if: runner.os == 'Windows'
        shell: pwsh
        run: |
          # Create release directory
          New-Item -ItemType Directory -Force -Path "release\mel-band-roformer"
          
          # Find and copy CLI executable
          $CliPath = Get-ChildItem -Path build -Filter "mel_band_roformer-cli.exe" -Recurse -File | Select-Object -First 1
          if ($CliPath) {
            Copy-Item $CliPath.FullName "release\mel-band-roformer\"
          }
          
          # Copy DLL files
          Get-ChildItem -Path build -Filter "*.dll" -Recurse -File | ForEach-Object {
            Copy-Item $_.FullName "release\mel-band-roformer\" -ErrorAction SilentlyContinue
          }
          
          # List contents
          Write-Host "Release artifact contents:"
          Get-ChildItem "release\mel-band-roformer" | Format-Table Name, Length
      
      - name: Upload Release Artifact
        uses: actions/upload-artifact@v4
        with:
          name: MelBandRoformer-${{ matrix.name }}
          path: release/mel-band-roformer/
          retention-days: 30

  # ===========================================================================
  # CUDA Build: Linux (Verification Only - Not for Release)
  # ===========================================================================
  # Note: We do not provide precompiled CUDA binaries for Linux.
  # This build exists solely to verify the code compiles with CUDA on Linux.
  # Linux users should build from source for optimal performance.
  build-cuda-linux:
    name: build-cuda-linux-verify
    runs-on: ubuntu-latest
    
    steps:  
      - name: Checkout
        uses: actions/checkout@v4
      
      - name: Clone GGML
        run: git clone --depth 1 https://github.com/ggerganov/ggml.git ggml

      - name: Install CUDA Toolkit
        uses: Jimver/cuda-toolkit@master
        with:
          cuda: "12.9.1"
          method: network
          sub-packages: '["nvcc", "cudart", "thrust"]'
          non-cuda-sub-packages: '["libcublas", "libcublas-dev"]'
      
      - name: Install Dependencies
        run: |
          sudo apt-get install -y cmake build-essential ninja-build
          
      - name: Configure
        run: |
          ls -ld ggml
          cmake -B build -G Ninja \
            -DCMAKE_BUILD_TYPE=Release \
            -DGGML_DIR=ggml \
            -DGGML_CUDA=ON \
            -DGGML_CUDA_FORCE_MMQ=ON \
            -DCMAKE_CUDA_RUNTIME_LIBRARY=Static \
            -DCMAKE_CUDA_ARCHITECTURES="61;75;80;86;89;120" \
            -DMBR_BUILD_TESTS=OFF \
            -DMBR_BUILD_CLI=ON
            
      - name: Build
        run: cmake --build build --config Release -j $(nproc)
        
      - name: Upload Build Artifacts (Debug Only)
        uses: actions/upload-artifact@v4
        with:
          name: build-linux-cuda-verify
          path: |
            build/bin/
            build/lib*/
            build/*.so
          retention-days: 7

  # CUDA Build: Windows (Single-Architecture Distribution)
  # ===========================================================================
  # Strategy: Each architecture built separately for minimal file size
  # - CUDA 11.8: GTX 10 / RTX 20 / RTX 30 / RTX 40 (Driver >= 520)
  # - CUDA 12.9: RTX 50 only (Driver >= 575)
  build-cuda-windows:
    name: build-cuda-windows-${{ matrix.arch_name }}
    runs-on: windows-2022
    strategy:
      fail-fast: false
      matrix:
        include:
          # CUDA 11.8 builds (Driver >= 520)
          - { cuda_version: "11.8.0", arch: "61", arch_name: "gtx10-pascal" }
          - { cuda_version: "11.8.0", arch: "75", arch_name: "rtx20-turing" }
          - { cuda_version: "11.8.0", arch: "80", arch_name: "rtx30-desktop" }
          - { cuda_version: "11.8.0", arch: "86", arch_name: "rtx30-mobile" }
          - { cuda_version: "11.8.0", arch: "89", arch_name: "rtx40-ada" }
          
          # CUDA 12.9 build (Driver >= 575)
          - { cuda_version: "12.9.1", arch: "120", arch_name: "rtx50-blackwell" }
    
    steps:
      - name: Checkout
        uses: actions/checkout@v4
      
      - name: Setup MSVC
        uses: ilammy/msvc-dev-cmd@v1

      - name: Clone GGML
        run: git clone --depth 1 https://github.com/ggerganov/ggml.git ggml
      
      - name: Install CUDA Toolkit
        uses: Jimver/cuda-toolkit@master
        with:
          cuda: ${{ matrix.cuda_version }}
          method: network
          sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'

      - name: Install Ninja
        run: choco install ninja -y
        
      - name: Configure and Build
        shell: cmd
        run: |
          REM CUDA 11.8 requires compatibility flags for newer MSVC
          if "${{ matrix.cuda_version }}" == "11.8.0" (
            set CUDAFLAGS=-allow-unsupported-compiler -D_ALLOW_COMPILER_AND_STL_VERSION_MISMATCH -D_DISABLE_CONSTEXPR_MUTEX_CONSTRUCTOR
          ) else (
            set CUDAFLAGS=
          )
          
          cmake -B build -G "Ninja Multi-Config" ^
            -DGGML_DIR=ggml ^
            -DGGML_CUDA=ON ^
            -DGGML_CUDA_FORCE_MMQ=ON ^
            -DCMAKE_CUDA_RUNTIME_LIBRARY=Static ^
            -DCMAKE_CUDA_ARCHITECTURES="${{ matrix.arch }}" ^
            -DMBR_BUILD_TESTS=OFF ^
            -DMBR_BUILD_CLI=ON
          cmake --build build --config Release -j %NUMBER_OF_PROCESSORS%
        
      - name: Verify Binary Dependencies
        shell: pwsh
        run: |
          Write-Host "=== Verifying Binary Dependencies ===" -ForegroundColor Cyan
          
          # Find all DLLs and EXEs
          $binaries = Get-ChildItem -Path build -Include *.dll,*.exe -Recurse -File
          
          if ($binaries.Count -eq 0) {
            Write-Host "WARNING: No binaries found!" -ForegroundColor Yellow
            exit 0
          }
          
          $hasProblems = $false
          $forbiddenDeps = @("cudart64", "cudart32", "cublas64", "cublas32", "cublasLt64")
          
          foreach ($binary in $binaries) {
            Write-Host "`n--- $($binary.Name) ---" -ForegroundColor Green
            
            # Use dumpbin to get dependencies
            $deps = & dumpbin /dependents $binary.FullName 2>&1
            
            # Extract DLL names
            $dllDeps = $deps | Select-String -Pattern "^\s+(\S+\.dll)" | ForEach-Object { $_.Matches.Groups[1].Value }
            
            if ($dllDeps) {
              Write-Host "Dependencies:"
              foreach ($dep in $dllDeps) {
                # Check for forbidden dependencies
                $isForbidden = $false
                foreach ($forbidden in $forbiddenDeps) {
                  if ($dep -like "$forbidden*") {
                    Write-Host "  [FAIL] $dep" -ForegroundColor Red
                    $isForbidden = $true
                    $hasProblems = $true
                  }
                }
                if (-not $isForbidden) {
                  Write-Host "  [OK] $dep" -ForegroundColor Gray
                }
              }
            } else {
              Write-Host "  No DLL dependencies found (static build)" -ForegroundColor Gray
            }
          }
          
          Write-Host "`n=== Summary ===" -ForegroundColor Cyan
          if ($hasProblems) {
            Write-Host "FAILED: Found forbidden CUDA runtime dependencies!" -ForegroundColor Red
            Write-Host "The build should use static CUDA runtime linking." -ForegroundColor Red
            exit 1
          } else {
            Write-Host "PASSED: No forbidden CUDA dependencies found." -ForegroundColor Green
          }
      
      - name: Upload Build Artifacts
        uses: actions/upload-artifact@v4
        with:
          name: build-windows-cuda-${{ matrix.arch_name }}
          path: |
            build/bin/
            build/Release/
            build/*.dll
          retention-days: 7
      
      # ----- Prepare Release Artifact -----
      - name: Prepare Release Artifact
        shell: pwsh
        run: |
          # Create release directory
          New-Item -ItemType Directory -Force -Path "release\mel-band-roformer"
          
          # Find and copy CLI executable
          $CliPath = Get-ChildItem -Path build -Filter "mel_band_roformer-cli.exe" -Recurse -File | Select-Object -First 1
          if ($CliPath) {
            Copy-Item $CliPath.FullName "release\mel-band-roformer\"
          }
          
          # Copy DLL files
          Get-ChildItem -Path build -Filter "*.dll" -Recurse -File | ForEach-Object {
            Copy-Item $_.FullName "release\mel-band-roformer\" -ErrorAction SilentlyContinue
          }
          
          # List contents
          Write-Host "Release artifact contents:"
          Get-ChildItem "release\mel-band-roformer" | Format-Table Name, Length
      
      - name: Upload Release Artifact
        uses: actions/upload-artifact@v4
        with:
          name: MelBandRoformer-windows-cuda-${{ matrix.arch_name }}
          path: release\mel-band-roformer\
          retention-days: 30