Heterogeneous Multi-processor Embedded System

In recent years,with the development of large scale integrated circuits,the number of transistors on the chip is expected to develop in accordance with the Moore's Law.Also,according to the Moore's Law,the speed of processing on the chip can not be further improved because of the quantum effects.So when the number of transistors on the chip is in a certain number,it is another solution that making the chip perform well.At the beginning of the 20 th century,the performance of the CPU architecture can not fit in the calculated amount of the game and graphics.And GPU began to develop rapidly.In the next few years,some scholars found that the GPU system is very consistent with the general purpose computing mode.And some GPU manufactures also began to support general purpose computation.Therefore,in the past 10 years,GPGPU has been developed,the concept of multi-core Heterogeneous prevailed.Special application applying to special architecture make the chip perform well.So FPGA that can be programmed at any time is to used widely in some specific areas.The work in this thesis is to customize an open-source GPGPU(nyuzi)that is from the Github,Which is connected to the ARM core in the board of XILINX via the interrupt and AXI bus.The GPGPU is a soft core on the FPGA,which is a computing core for general purpose computation in multi-threads.And the ARM core in the system is the hard core on the XILINX ZYNQ board.In the ZYNQ chip,ARM core and some peripherals such as memory are connected through the AXI bus,the system control the GPGPU core operation through the AXI bus,and return to the ARM core with a interrupt when the computation task is finished.The ARM core share one physical memory with the GPGPU core.Then,a multi-core heterogeneous system come true with the ARM core as the controller and GPGPU core as the computation core.In the system,with the control of The ARM core,push the data on the shared memory,handle the GPGPU core to execute the computation task,then push the result of the computation task on the memory when it is finished.The advantage of the method is that no data is copied.At same time,it build a set of software running on the hardware system.Although the test show that it is not good as some general processor like AMD-A10 and GPU,the nyuzi core is a soft core which runs at a frequency of 50 M,it has a big practical significance.The efficiency of the system depend on the scheduling on the software,which is need to be improved.