Dynamically scheduling VLIW instructions with dependency information

The paper proposes balancing scheduling effort more evenly between the compiler and the processor, by introducing dynamically scheduled Very Long Instruction Word (VLIW) instructions. Dynamically Instruction Scheduled VLIW (DISVLIW) processor is aimed specifically at dynamic scheduling VLIW instructions with dependency information. The DISVLIW processor dynamically schedules each instruction within long instructions using functional unit and dynamic scheduler pairs. Every dynamic scheduler dynamically checks for data dependencies and resource collisions while scheduling each instruction. This scheduling is especially effective in applications containing loops. We simulate the architecture and show that the DISVLIW processor performs significantly better than the VLIW processor for a wide range of cache sizes and across various numerical benchmark applications.