Architectural Verification of Four-instruction Superscalar Processor for MIPS I Instruction Set

Abstract

The study undertaken in this thesis tries to tackle this inefficiency by having extra register locations other than the architectural registers called pseudo-registers, and a pointer scheme is followed to reference both architectural and pseudo registers. This scheme renames each logical destination register of an incoming instruction, to a pseudo register referenced by pointers called pseudo-pointers. Two separate lists of these pointers are maintained, one for all types of instructions and the other for only unspeculated instructions. When a branch instruction preceding the speculated instruction is evaluated and it is established that the prediction was correct, the machine state is altered by updating the pointer lists instead of moving the data. As the pointes are only 6-bits, the inefficiency is considerably reduced. This processor scheme is implemented using the Verilog hardware description language (HDL). The following study provides architectural details of each component used in the processor, stressing issues involved in the implementation and methods used to overcome these issues. This study also discusses verification methodology, documenting steps involved in compiling a 'c' program and loading it onto the simulated instructions cache and data cache for simulation. Finally, simulation results are presented for a sample 'c' program verifying the design.