Final Presentation Part-A

Final PresentationPart-A

Infrastructure design & implementation of MIPS processors for students lab based on

Bluespec HDL

Students: Danny Hofshi, Shai ShachrurSupervisor: Mony Orbach

Winter 2012

FPGA

Bluespec Scemi

Bluespec HDL

C++

Reminder

Project Goals (from characteristic presentation)

Part A:

Creating the Laboratory working environment.

1. Hardware environment – RTL , interface to outer world ( pci - express ), Xilinx utilities.

2. Software environment – simulation of a MIPS processor on the above RTL using simplified commands, the same environment will be used for emulation and simulation.

Project Characterization

Multi cycle MIPS hardware

Adjusting the experiment flow

Test, Sync & Conclusions

Designing the MIPS improvements

performance questionsProgram for testing

Running a pilot group

Danny & ShaiLab staff

Part A

Part B

We are here

What has been done

Hardware infrastructure

Hardware

PCIe cable

Virtex 5 FPGA

Linux Environment

What has been done

RTL Design

MIPS Design

Xilinx BRAM

Xilinx BRAM

BlueSpecS C E M I

- PCIe

-Clk control

Communicating with a c++ environment using generic interface

BlueSpec

Co-Process

or

The MIPS state machine

What has been done

Software environment

Control interface Ability to Put & get data from the instruction

memory.Ability to Put & get data from the data

memory.Ability to control the MIPS operation (Co-

processor)- stop- load PC, start & wait for interrupt - load PC & start for a specific number of cycles

Retrieve performances (Co-processor).- Number of clock cycles. - Number of Instructions.

SCEMI

CompilerPerl Script compiler – translate assembler

code to our specific SCEM interface.

addi $t1 $zero 2 addi $t2 $zero 2 add $t3 $t1 $t2sw $t3 $zero 0 stop

604569602 0 0 1 0604635138 4 4 1 01952289 8 8 1 02 8 86402048 12 12 1 0

1082130432 16 16 1 0

0 1 1 1 22 0 0 1 21 0 0 1 2

Processor commands

Stop commandwill generate an interrupt

Load PC Set PCStart

User work FlowWrite Assembler Code & Data memory

Type executable command

Receive performances and data

addi $t1 $zero 2 addi $t2 $zero 2 add $t3 $t1 $t2sw $t3 $zero 0 stop

Compiler604569602 0 0 1 0

604635138 4 4 1 01952289 8 8 1 02 8 86402048 12 12 1 0

1082130432 16 16 1 0

0 1 1 1 22 0 0 1 21 0 0 1 2

Communication module C++ MIPS Data

received

On the PC

Project Characterization

Multi cycle MIPS hardware

Adjusting the experiment flow

Test, Sync & Conclusions

Designing the MIPS improvements

performance questionsProgram for testing

Running a pilot group

Danny & ShaiLab staff

Part A

Part B

We are here

Test programThe Bubble sort Assembler code:// size of arrayaddi $s0 $zero 64 // address counteraddi $s1 $zero 0 // swap indicatorstart : addi $s2 $zero 0 start_in_iteration : lw $t0 $s1 0lw $t1 $s1 4 // $t0=1 if we need to swap, $t0=0 elseslt $t3 $t1 $t0beq $t3 $zero no_swapsw $t0 $s1 4sw $t1 $s1 0// set swap indicatoraddi $s2 $zero 1no_swap : beq $s1 $s0 end_iterationaddi $s1 $s1 4beq $zero $zero start_in_iteration  end_iteration : beq $s2 $zero stop_sortaddi $s1 $zero 0addi $s0 $s0 -4beq $s0 $zero stop_sortbeq $zero $zero start 

RunThe program Run on the CPU. Typing the command line to compile and run the above code [s03992455@diglabl2 fpga]$ perl parser.pl -source bubble_sort.S -end -meminit mem_init.txtStarting to parse code ********* Running command_file.txt ************ (Writing Data to the I-MEM)Loading data to I-mem: 605028416 0 0 1 0Loading data to I-mem: 605093888 4 4 1 0Loading data to I-mem: 605159424 8 8 1 0Loading data to I-mem: 2384986112 12 12 1 0Loading data to I-mem: 2385051652 16 16 1 0Loading data to I-mem: 19421226 20 20 1 0Loading data to I-mem: 291504131 24 24 1 0Loading data to I-mem: 2921857028 28 28 1 0Loading data to I-mem: 2921922560 32 32 1 0Loading data to I-mem: 605159425 36 36 1 0Loading data to I-mem: 305135618 40 40 1 0Loading data to I-mem: 640745476 44 44 1 0Loading data to I-mem: 268500982 48 48 1 0Loading data to I-mem: 306184196 52 52 1 0Loading data to I-mem: 605093888 56 56 1 0Loading data to I-mem: 638648316 60 60 1 0Loading data to I-mem: 301989889 64 64 1 0Loading data to I-mem: 268500976 68 68 1 0Loading data to I-mem: 1082130432 72 72 1 0Loading to arg register 0 1 1 1 2  (Loading the start value of the MIPS Program counter)Load PC from arg register2 0 0 1 2 (Applying the start value to the PC)Start CPU 1 0 0 1 2 (Starting the run on the MIPS) ####### Program run on CPU finished ####### (Receiving an interrupt from the PC) SceMi Service thread finished!SceMi Service thread finished! 

Results ********* inquiring check_mem.txt ************ (Reading the performance counters sort DATA) requesting number of cycles  (Sending a data request to the Co-processor)number of clock cycles is 591number of instructions is 122{data 32'h00000011 resp_type 4'h1 resp_addr 32'h00000000}{data 32'h00000014 resp_type 4'h1 resp_addr 32'h00000004}{data 32'h00000015 resp_type 4'h1 resp_addr 32'h00000008}{data 32'h00000037 resp_type 4'h1 resp_addr 32'h0000000c}{data 32'h0000003c resp_type 4'h1 resp_addr 32'h00000010}{data 32'h0000003d resp_type 4'h1 resp_addr 32'h00000014}{data 32'h0000003d resp_type 4'h1 resp_addr 32'h00000018}{data 32'h00000045 resp_type 4'h1 resp_addr 32'h0000001c}{data 32'h00000047 resp_type 4'h1 resp_addr 32'h00000020}{data 32'h0000004a resp_type 4'h1 resp_addr 32'h00000024}{data 32'h0000004d resp_type 4'h1 resp_addr 32'h00000028}{data 32'h00000050 resp_type 4'h1 resp_addr 32'h0000002c}{data 32'h00000053 resp_type 4'h1 resp_addr 32'h00000030}{data 32'h00000058 resp_type 4'h1 resp_addr 32'h00000034}{data 32'h00000059 resp_type 4'h1 resp_addr 32'h00000038}{data 32'h0000005d resp_type 4'h1 resp_addr 32'h0000003c}{data 32'h00000060 resp_type 4'h1 resp_addr 32'h00000040}{data 32'h00000061 resp_type 4'h1 resp_addr 32'h00000044}Finished parsing

591 Cycles

122 Instructions

4.844 Cycles per instruction

RemarksWe can also run in simulation mode for RTL

debug.A full verification on the whole MIPS

command set was made.

DiscussionWe

implement Meow

prediction ?