Estudo e avaliação de arquiteturas de processadores quanto ao uso de memória em sistemas embarcados

Abstract

Embedded systems have many constraints and requirements. In addition to the significant cost, these systems must meet hard performance restrictions and limitations in the energy consumption. The memory system is one of the major contributing factors to the performance and power consumption in embedded software. Virtual platforms offer an alternative to hardware prototypes and they offer many advantages such as early development, performance and accessibility. There are many virtual platforms and simulators that provide several processor architecture models. This work presents a study and evaluation on the use of memory in embedded processor architectures. This work presents different memory optimization techniques for embedded systems. ARM and x86 architectures were evaluated to access to memory and energy consumption. For these evaluations, instructions and data trace were generated by different embedded applications from MiBench package. The evaluation tools of embedded processor architectures used were Simics and CACTI. The results show that 34,42% of all instructions to ARM architecture are load and store instructions. In addition, 64,41% of access, on average, to instruction memory are caused by only 8KB of the program, as 64KB for the percentage is 84,58%. In relation to data memory, 8KB of data for this architecture are responsible for 70,66% of the memory accesses. In the x86 architecture predominate move instructions, which represent, on average, 88,06% of executed instructions. In the x86 architecture, 95,94% of access to instruction memory are caused by only 256 bytes, while an 8 KB of data memory is responsible for 73,62% of the memory accesses. Experimental results showed great optimization potential of memory access and high possibility of the design space exploration for ARM architecture with the use of an SPM memory.