designKilla: The 32-bit pipelined processor Brought to you by: Victoria Farthing Dat Huynh Jerry Felker Tony Chen Supervisor: Young Cho

32-Bit RISC Pipelined Processor Reduced Instruction Set allows for faster execution of simple, frequently used instructions which can be combined to achieve the same result as a single, slower CISC instruction Pipelining allows a faster clock cycle and less wasted resources

Datapath Pipeline Stages 5 Stages – Instruction Fetch – Instruction Decode – Execution – Memory Write – Write Back

Unique Data Path Features Next instruction address calculation –For basic incrementation, the address is calculated by a counter

Address Jump Calculations –For address jumps, there is a 19-bit load port on the counter The loaded address comes from an adder with multiplexed inputs Load bit is controlled by a comparator (beq) or-ed with the absolute jump control bit

Double Clocked Memory Interface Problem:Problem: One Memory for both Instruction and Data Solution:Solution: Double Clock! twiceAccess the memory twice during one clock cycle

Fast Clock Clock Fetch Instruction Fetch DataFetch Instruction Write Enable Write Data Fetches Instruction in First Cycle Fetches or Writes Data In Second Cycle Data is output by end of Clock Cycle

Unique Data Path Features Structural Multiplier –16 X 16 bit –Multi-level creation: Four 8 X 8 bit multipliers –Each containing four 4 X 4 bit multipliers Each comprised of a cascaded network of full and half adders, built on logic gates

16-Bit Multiplier Unit Based On Hand Multiplication Made Up of Network of AND Gates and Adders

Why 32  16 bit? 32bit x 32bit = 64 bits! Multiple complex changes to existing architecture would be required Only one register can be written per clock cycle –Could hold value for next cycle or stall the pipeline Would require pseudoinstruction as well as new hardware and multiple control signals

Use pseudo-code instruction mult32 mult 20, 2, 4 mult 21, 4, 1 mult 22, 2, 3 mult 23, 1, 3 and 24, 20, 30 srli 24, 24, 16 and 25, 21, 31 add 25, 24, 25 and 24, 22, 31 add 25, 25, 24 and 5, 25, 31 srli 5, 5, 16 and 20, 20, 31 or 5, 5, 20 srli 25, 25, 16 and 24, 22, 30 srli 24, 24, 16 add 24, 24, 25 and 25, 23, 31 add 24, 24, 25 and 22, 24, 30 srli 22, 22, 16 and 21, 23, 30 srli 21, 21, 16 add 6, 21, 22 slli 6, 6, 16 and 24, 24, 31 or 6, 6, 24

Improve the Multiplier Can decrease the latency of a combinational multiplier with carry-look ahead adding methods. –Small amount of extra hardware needed, worth it if multiplier has largest latency.

Other Multiplier Topologies Shifting multiplication –Shift multiplicand several times based on multiplier bits –Add intermediate shifted values

Other Multiplier Topologies Pipelined multiplication –Store intermediate sums –Allows for faster clock cycle if traditional combinational multiplication presents the critical path

Other Multiplier Topologies Pipelined multiplication –Sequential multiplication Useful to minimize hardware waste if multiplication is an infrequent operation Continues to allow for faster clock cycle if traditional combinational multiplication presents the critical path

Instruction Set Architecture R-Type

I-Type J-Type

Converts assembly code to binary representation The Assembler Add $3,$1,$2 => => High => Low 16-bit wide memory modules Split into high and low bits for output

Allows for labels to be used in loops Automatically calculates offsets based on label position LABEL: add $1,$2,$3 jmp LABEL Resolves hazards created by pipelining 1.Automatically determines the appropriate number of NO-OPS to insert based on relative position of consecutive instructions Assembler Features

Design allows for pseudo-instructions to be used Pseudo Instruction HLT Actual Instructions H1:JMP H1 NOP

Topic 2 Design – Compiler Bison - Parser A compiler compiler A grammar generator Flex – Lexer A Fast lexical analyzer Tool used in pattern matching on text

CompilingThe C Language Interface Lexer and Parser Lex will feed tokens to Bison (YACC) A grammar tree is generated

Source code to run-time

A simple program A simple C program void main ( void ) { int b ; int d; int x; int y = 3; int g; x = b + d; g = y + x; } Assembly Code Equivalent lwi 4, 0, 3 add 6, 1, 2 sw 3, 6, 0 add 6, 4, 3 sw 5, 6, 0 Memory High Memory Low Machine Code Instructions

Could Use a Little Work Currently the Processor could use a little work to improve performance. –Decreased memory latency would be largest and most direct improvement to processor. –Must optimize ALU as well as multiplier unit. –All in all, will work but not ready for commercial usage.

References Computer Organization and Design: The Hardware Software Interface (2 nd Ed) Patterson, David A. and Hennessy, John L. Morgan Kaufman Publishers, 1997 Introduction to Compilers Aaby, Anthony A., 1998 The Compiler Design Handbook Srikant, Y. N. and Shankar, Priti CRC Press, 2002

THE END Questions?