Merge pull request #7 from AIComputing101/coketaste/profile-cuda

Fix the Makefile for profile on cuda

Author: coketaste

modules/module1/examples/Makefile

@@ -206,7 +206,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif

modules/module2/examples/Makefile

@@ -159,7 +159,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif
@@ -260,8 +260,8 @@ profile_memory: cuda
 	@echo "  ncu --metrics l1tex__throughput.avg.pct_of_peak_sustained_elapsed ./02_memory_coalescing_cuda"
 	@echo "  ncu --metrics dram__bytes_read.sum,dram__bytes_write.sum ./05_memory_bandwidth_optimization_cuda"
 	@echo ""
-	@echo "Legacy nvprof (if available):"
-	@echo "  nvprof --metrics achieved_occupancy,gld_efficiency,gst_efficiency ./03_texture_memory_cuda"
+	@echo "Modern NVIDIA Nsight Systems:"
+	@echo "  nsys profile --cuda-event-trace=false -o profile.nsys-rep ./03_texture_memory_cuda"
 
 
 

modules/module3/examples/Makefile

@@ -164,7 +164,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif

modules/module4/examples/Makefile

@@ -173,7 +173,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif
@@ -419,9 +419,9 @@ profile_examples: all
 	@echo "  nsys profile --trace=cuda,nvtx --stats=true ./02_multi_gpu_programming"
 	@echo "  nsys profile --trace=cuda,nvtx,osrt --stats=true ./03_unified_memory"
 	@echo ""
-	@echo "Legacy nvprof:"
-	@echo "  nvprof --print-gpu-trace ./01_cuda_streams_basics"
-	@echo "  nvprof --print-api-trace ./02_multi_gpu_programming"
+	@echo "Modern NVIDIA Nsight Systems:"
+	@echo "  nsys profile --cuda-event-trace=false -o trace.nsys-rep ./01_cuda_streams_basics"
+	@echo "  nsys profile --cuda-event-trace=false -o multi_gpu.nsys-rep ./02_multi_gpu_programming"
 	@echo ""
 	@echo "Multi-GPU Analysis:"
 	@echo "  nsys profile --trace=cuda,nvtx --stats=true -o multi_gpu_trace ./02_multi_gpu_programming"

modules/module5/examples/Makefile

@@ -57,7 +57,7 @@ HIP_DEBUG_FLAGS += --offload-arch=$(GPU_ARCH)
 CXX_FLAGS = -std=c++17 -O3 -fopenmp
 
 # Profiling flags
-NVPROF_FLAGS = --print-gpu-trace --log-file %s.nvprof
+NSYS_FLAGS = --cuda-event-trace=false --force-overwrite
 ROCPROF_FLAGS = --hip-trace --stats --output-file %s.csv
 
 # Directories
@@ -144,7 +144,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif
@@ -171,11 +171,12 @@ run: all
 # Performance profiling targets
 .PHONY: profile-cuda
 profile-cuda: $(CUDA_TARGETS)
-	@echo "Profiling CUDA examples with nvprof..."
+	@echo "Profiling CUDA examples with nsys..."
+	@mkdir -p $(PROFILE_DIR)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof $(NVPROF_FLAGS) $$target > $(PROFILE_DIR)/$$(basename $$target).nvprof 2>&1; \
+			nsys profile $(NSYS_FLAGS) -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target > $(PROFILE_DIR)/$$(basename $$target).nsys.log 2>&1; \
 		fi; \
 	done
 
@@ -330,7 +331,7 @@ help:
 	@echo "  validate           - Validate optimization correctness"
 	@echo ""
 	@echo "Profiling Targets:"
-	@echo "  profile-cuda       - Profile CUDA examples with nvprof"
+	@echo "  profile-cuda       - Profile CUDA examples with nsys"
 	@echo "  profile-hip        - Profile HIP examples with rocprof"
 	@echo "  profile-detailed-cuda - Detailed profiling with Nsight Compute"
 	@echo "  memcheck-cuda      - Run CUDA memory checker"

modules/module6/examples/Makefile

@@ -142,7 +142,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif

modules/module7/examples/Makefile

@@ -150,7 +150,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif

modules/module8/examples/Makefile

@@ -207,7 +207,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif

modules/module9/examples/Makefile

@@ -222,7 +222,7 @@ ifeq ($(BUILD_CUDA),1)
 	@for target in $(CUDA_TARGETS); do \
 		if [ -f $$target ]; then \
 			echo "Profiling $$target..."; \
-			nvprof --csv -o $(PROFILE_DIR)/$$(basename $$target).csv $$target 2>/dev/null || echo "nvprof completed"; \
+			nsys profile --cuda-event-trace=false -o $(PROFILE_DIR)/$$(basename $$target).nsys-rep $$target 2>/dev/null || echo "nsys completed"; \
 		fi; \
 	done
 endif