fix(workflow): update Windows CUDA build configuration to use Ninja generator and adjust build command

Author: Gioee

.github/workflows/main.yml

@@ -96,7 +96,7 @@ jobs:
           - os: windows-latest
             arch: x86_64
             name: windows-cuda
-            make: LLAMA="-DGGML_NATIVE=OFF -DGGML_CPU=OFF -DGGML_CUDA=ON"
+            make: LLAMA="-G \"Ninja Multi-Config\" -DGGML_NATIVE=OFF -DGGML_CPU=OFF -DGGML_CUDA=ON"
           - os: windows-latest
             arch: x86_64
             name: windows-vulkan
@@ -222,9 +222,10 @@ jobs:
         shell: pwsh
 
       - name: windows build ggml-cuda
-        if: matrix.name == 'windows-cuda'
-        run: make build/llama.cpp.stamp LLAMA="-G Ninja Multi-Config -DGGML_NATIVE=OFF -DGGML_CPU=OFF -DGGML_CUDA=ON" LLAMA_ARGS="--target ggml-cuda"
-        shell: pwsh
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
+          make build/llama.cpp.stamp ${{ matrix.make && matrix.make || ''}} LLAMA_ARGS="--target ggml-cuda"
 
       - name: windows install hip toolkit
         if: matrix.name == 'windows-hip'