build: align CUDA build outputs and CI checks

Unify Makefile output paths with the CMake layout, broaden default CUDA architecture coverage, and restore a containerized CUDA compile-only CI job. This keeps local build commands consistent with docs while adding a realistic non-GPU validation step in CI.

Author: shijiashuai

.github/workflows/ci.yml

@@ -29,3 +29,26 @@ jobs:
           check-path: '.'
           exclude-regex: '(^|/)(build|third_party|external|vendor|\.git)(/|$)'
           fallback-style: 'LLVM'
+
+  cuda-build:
+    name: CUDA Build (compile only)
+    runs-on: ubuntu-latest
+    container:
+      image: nvidia/cuda:12.4.1-devel-ubuntu22.04
+    steps:
+      - name: Install build tools
+        run: apt-get update && apt-get install -y cmake git build-essential
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Configure CMake
+        run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=OFF
+
+      - name: Build benchmark target
+        run: cmake --build build --target sgemm_benchmark -j2
+
+      - name: Document runtime limitation
+        run: |
+          echo "This workflow validates CUDA compilation in a containerized toolkit environment."
+          echo "GPU runtime tests still require a CUDA-capable local machine or dedicated GPU runner."

CMakeLists.txt

@@ -20,7 +20,7 @@ if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
     if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.24")
         set(CMAKE_CUDA_ARCHITECTURES native)
     else()
-        set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89)
+        set(CMAKE_CUDA_ARCHITECTURES 70 75 80 86 89 90)
     endif()
 endif()
 

Makefile

@@ -27,8 +27,9 @@ MAIN_SRC = $(SRC_DIR)/main.cu
 TEST_SRC = $(TEST_DIR)/test_sgemm.cu
 
 # Targets
-MAIN_TARGET = $(BUILD_DIR)/sgemm_benchmark
-TEST_TARGET = $(BUILD_DIR)/test_sgemm
+BIN_DIR = $(BUILD_DIR)/bin
+MAIN_TARGET = $(BIN_DIR)/sgemm_benchmark
+TEST_TARGET = $(BIN_DIR)/test_sgemm
 
 # Header files (for dependency tracking)
 HEADERS = $(wildcard $(KERNEL_DIR)/*.cuh) $(wildcard $(UTILS_DIR)/*.cuh)
@@ -39,6 +40,7 @@ all: dirs $(MAIN_TARGET)
 
 dirs:
 	@mkdir -p $(BUILD_DIR)
+	@mkdir -p $(BIN_DIR)
 	@mkdir -p $(KERNEL_DIR)
 	@mkdir -p $(UTILS_DIR)
 	@mkdir -p $(TEST_DIR)
@@ -79,7 +81,7 @@ help:
 	@echo "  profile   - Build with profiling info"
 	@echo ""
 	@echo "Variables:"
-	@echo "  GPU_ARCH  - GPU architecture (default: sm_70)"
+	@echo "  GPU_ARCH  - GPU architecture (default: sm_86)"
 	@echo ""
 	@echo "Example:"
 	@echo "  make GPU_ARCH=sm_80 benchmark"