1 |
Improving the Fetching Performance of Instruction Stream Buffer for VLIW Architectures with Compressed InstructionsYang, Kai-Ming 25 August 2006 (has links)
Because of the restriction on structure hazard and instruction data dependence, the quantity of NOP instructions fills up a program for VLIW Architectures. This problem causes a waste of program memory, so that an instruction compression mechanism is a must for VLIW Architectures. The vectorized instruction in DVB-T (Digital Video Broadcasting - Terrestrial) DSP will collect the discrete vectors into one continuous vector. This mechanism is based on the software-pipeline of the zero overhead looping mode. It is important to improve the efficiency of instruction fetcher. Additionally, the branch instruction can cause the non-continuous behavior of a program and the damage of the efficiency of instruction fetcher. The mechanism of compressed instructions causes the irregular length of long instruction in fetch packet. The problem becomes difficult designed. The thesis implements a design of improving instruction stream buffer, which can keep the repeat block in buffer. This mechanism overcomes the effects of zero overhead looping and branch instruction. It can also improve the efficiency of continuously fetch instructions. The simulation result shows that the mechanism has a good efficiency in FFT, FIR and DCT.
|
Page generated in 0.0858 seconds