chapter3-m2-ziavras

Ziavras basic compiler techniques for exposing ilp

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ce exception move DSUBU before branch without changing data flow (compiler SPECULATION) S. Ziavras Basic Compiler Techniques for Exposing ILP • Objective: Compiler technology to improve the performance of – Pipelines Pipelines – Simple multiple-issue processors • Applicable primarily to static-issue processors – Can also help processors with dynamic issue decisions & static scheduling • Basic idea idea – Extract sequences of unrelated instrs. that can be overlapped in pipeline or can be executed at the same time Pipeline: if there is a data dependence, separate the 2 instrs. by the appropriate # of cycles S. Ziavras Assumptions • • • • Standard 5-stage integer pipeline Branches have 1 clock cycle delay No structural hazards (pipelined or ∞ FUs) structural hazards or Latency of FP L to S: 0 (L result can be bypassed without stalling S) • Integer load latency: 1 • Integer ALU operation latency: 0 • Required distance between FP instrs. (avoid stalls): distance between FP instrs stalls Result producing instr. Inst...
View Full Document

This document was uploaded on 02/09/2014.

Ask a homework question - tutors are online