chapter3-m5-ziavras

Vliw very long instruction word or epic explicitly

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: achieve ideal CPI of 1 Here: Decrease CPI to less than 1 1. Superscalar processors – Statically scheduled superscalars: in-order execution – Dynamically scheduled superscalars: out-oforder execution 2. VLIW (Very Long Instruction Word) or EPIC (Explicitly Parallel Instruction Computers) (Explicitly Parallel Instruction Computers) processors: statically scheduled S. Ziavras Statically Scheduled Superscalars • Typically the HW issues 0-8 instrs./clock cycle • Instrs. issue in order • All pipeline hazards are checked for at issue time S. Ziavras Statically Scheduled Superscalar MIPS • • Assumptions – 2 instrs. can be issued per clock cycle • One of these instrs. can be a L, S, branch, or integer ALU • The other can be any FP operation • L/S here can be integer of FP operation 3 steps in fetch & issue 1. Fetch 2 instrs. from cache • High-end superscalars rely on independent instr. prefetch unit 2. Determine whether 0, 1, or 2 instrs. can issue 3. Issue them to respective FUs S. Ziavras Statically Scheduled Superscalar MIPS (2) • Either – pipelined FPUs or – multiple FUs are needed • Otherwise the FP data path will quickly become the bottleneck advantages gained by dual issue will be gained by dual issue will be diminished S. Ziavras Dynamic Scheduling & Multiple Instruction Issue • • • Objective – Extend Tomasulo’s algorithm for 2-issue superscalar pipeline (any 2 instrs.) In-order instr. issue to preserve program semantics ti 2 approaches 1. Assign an RS & update the pipeline control an RS update the pipeline control tables in ½ of clock cycle 2 instrs. can be issued in the same cycle 2. Build logic to handle 2 instrs. simultaneously, including data dependences Both approaches often used in modern processors approaches often used in modern processors S. Ziavras Dynamic Scheduling & Multiple Instruction Issue (2) Important question – How should dynamic branch prediction be integrated into a dynamical...
View Full Document

This document was uploaded on 02/09/2014.

Ask a homework question - tutors are online