CUDA_PATH     ?= /usr/local/cuda
HOST_COMPILER  = g++
NVCC           = $(CUDA_PATH)/bin/nvcc

# select one of these for Debug vs. Release
NVCC_DBG       = -g -G
#NVCC_DBG       =

NVCCFLAGS      = $(NVCC_DBG) -m64
GENCODE_FLAGS  = --gpu-architecture=compute_50 --gpu-code=compute_50,sm_50,sm_52

BUILD_DIR = build
SRCS = src/main.cu
INCS = src/vec3.cpp src/ray.cpp

$(BUILD_DIR)/cudart: $(BUILD_DIR)/cudart.o
	$(NVCC) $(NVCCFLAGS) $(GENCODE_FLAGS) -o $(BUILD_DIR)/cudart $(BUILD_DIR)/cudart.o

$(BUILD_DIR)/cudart.o: $(SRCS) $(INCS)
	mkdir -p $(BUILD_DIR)
	$(NVCC) $(NVCCFLAGS) $(GENCODE_FLAGS) -o $(BUILD_DIR)/cudart.o -c src/main.cu

out.ppm: $(BUILD_DIR)/cudart
	rm -f out.ppm
	./$(BUILD_DIR)/cudart

out.jpg: out.ppm
	rm -f out.jpg
	ppmtojpeg out.ppm > out.jpg

profile_basic: $(BUILD_DIR)/cudart
	nvprof ./$(BUILD_DIR)/cudart


# use nvprof --query-metrics
profile_metrics: $(BUILD_DIR)/cudart
	nvprof --metrics achieved_occupancy,inst_executed,inst_fp_32,inst_fp_64,inst_integer ./cudart

clean:
	rm -rf $(BUILD_DIR)
	rm -f out.ppm out.jpg