How to debug GPGPU-Sim

Use GDB

When debuging GPGPU-Sim, the first thing is compile the whole project under debug environment, this could be done at the beginning using source setup_environment debug. Then compile the targeted application program using nvcc -g flag. Now use gdb to debug the program, by simply gdb prog.

As for making break point, one method is break at file line level, like b gpgpusim_entrypoint.cc:199, which is the struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() function, it’s the logical entry point of the whole simualtion.

Inside gdb, we could use info frame to check the current stack and previous stack, use up could jumped to previous stack frame, and domn could jump to the next. It’s useful if we want to track the function calls. You could also use bt to show all the stack frames.

Add Log library

It’s also possible to add a log module into the GPGPU-Sim project. For example, if we would like to integrate the simple log.c, first we could create a new directory inside the src, named liblog. Then clone the log.c project into that folder. To simplyfy the makefile structure, just move all the *.c and .h files in the liblog folder. Then copy a makefile from any same level folder, like gpgpu-sim, change the following lines:

OUTPUT_DIR=$(SIM_OBJ_FILES_DIR)/liblog

liblog.a:$(OBJS)
	ar rcs  $(OUTPUT_DIR)/liblog.a $(OBJS)

remove other unrelated .o genegration, later this make file could be called from src level makefile. Now edit the src level Makefile, add these lines:

gpu_uarch_simlib:
	make   -C ./gpgpu-sim

# for the new module
liblog:
	make -C ./liblog

Then we could modify the highest level Makefile, add some lines:

# link to the final .libcudart.so
$(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib
    # ...
	$(SIM_OBJ_FILES_DIR)/liblog/*.o \
    #...
    -o $(SIM_LIB_DIR)/libcudart.so

# create corresponding build directory
liblig: makedirs 
	$(MAKE) -C ./src/liblig/ depend
	$(MAKE) -C ./src/liblig/

makedirs:
	# ...
    if [ ! -d $(SIM_OBJ_FILES_DIR)/liblog ]; then mkdir -p $(SIM_OBJ_FILES_DIR)/liblog; fi;
	# ...

Now if we want to use the log module in cuda-sim/cuda-sim.cc, first include the coresponding header file, and then modify the makefile to include the log module object file, by adding these lines:

OBJS	:= ... $(OUTPUT_DIR)/cuda_device_runtime.o $(OUTPUT_DIR)/../liblog/log.o
#...
$(OUTPUT_DIR)/../liblog/log.o:
	make -C ../liblog

Then we could re-compile the simulator, and the log module could work.