| 摘要 |
A pipelined processor uses pipe control and register translation to maximize instruction flow through, in an exemplary embodiment, the execution pipelines of a superscalar, superpipelined microprocessor compatible with the x86 instruction set architecture. The pipe control logic simultaneously issues instructions into X and Y pipelines without regard to data dependencies between those instructions issued instructions, and in particular without regard to read-after-write dependencies. Pipe switching allows instructions to be issued into either execution pipeline so as to minimize stalls due to data dependencies between instructions. Instruction flow is controlled independently for each pipeline, with data dependencies between instructions being monitored at each stage of a pipeline. A register translation unit implements register renaming to eliminate write-after-read and write-after-read dependencies, and data forwarding (both operand forwarding and result forwarding) to eliminate read-after-write dependencies through the pipelines is controlled independently. The register translation unit controls the mapping of the eight x86 logical registers into 32 physical registers, including maintaining for each physical register status as to availability, logical register allocation, and write pending -- allocations of physical registers are based on operand size. Instructions are allowed to proceed to execution once it has been determined that the instruction can no longer cause an exception, permitting out-of-order completion. Checkpoint and Exception registers in the register translation unit support speculative execution, permitting pipeline recovery and repair after an instruction exception, branch misprediction , or floating point error. A microcontrol unit provides independent microinstruction flows for both pipelines, enabling, for selected instruction: (a) both execution stages to be used to execute the same instruction, and (b) control of the register translation unit. <IMAGE> |
