Performance Modelling - RISC-V processor

Performance Modelling - RISC-V processor
This project will require you to implement cycle-accurate simulators of a 32-bit RISC-V processor in C++ or
Python. The skeleton code for the assignment is given in file (NYU_RV32I_6913.cpp or
NYU_RV32I_6913.py).
The simulators should take in two files as inputs: imem.text and dmem.txt files
The simulator should give out the following:
● cycle by cycle state of the register file (RFOutput.txt)
● Cycle by cycle microarchitectural state of the machine (StateResult.txt)
● Resulting dmem data after the execution of the program (DmemResult.txt)
The imem.txt file is used to initialize the instruction memory and the dmem.txt file is used to initialize the
data memory of the processor. Each line in the files contain a byte of data on the instruction or the data
memory and both the instruction and data memory are byte addressable. This means that for a 32 bit
processor, 4 lines in the imem.txt file makes one instruction. Both instruction and data memory are in
“Big-Endian” format (the most significant byte is stored in the smallest address).
The instructions to be supported by the processor are categorized into the following types:
The simulator should support the following set of instructions.
Mnemonic Type Full Name Psuedocode Details
ADD R Addition rd = rs1 + rs2 Store the result of rs1 + rs2 in register rd.
SUB R Subtraction rd = rs1 - rs2 Store the result of rs1 - rs2 in register rd.
XOR R Bitwise XOR rd = rs1 ^ rs2 Store the result of rs1 ^ rs2 in register rd.
OR R Bitwise OR rd = rs1 | rs2 Store the result of rs1 | rs2 in register rd.
AND R Bitwise AND rd = rs1 & rs2 Store the result of rs1 & rs2 in register rd.
ADDI I Add Immediate rd = rs1 + sign_ext(imm)
Add the sign-extended immediate to
register rs1 and store in rd. Overflow bits ignored.
XORI I XOR Immediate rd = rs1 ^ sign_ext(imm)
Bitwise XOR the sign-extended immediate to
register rs1 and store result in rd.
ORI I OR Immediate rd = rs1 | sign_ext(imm)
Bitwise OR the sign-extended immediate to
register rs1 and store result in rd.
ANDI I AND Immediate rd = rs1 & sign_ext(imm)
Bitwise AND the sign-extended immediate to
register rs1 and store result in rd.
JAL J Jump and Link
rd = PC + 4;
PC = PC + sign_ext(imm)
Jump to PC = PC + sign_ext(imm) and store the
current PC + 4 in rd.
BEQ B Branch if equal
PC = (rs1 == rs2)? PC +
sign_ext(imm) : PC + 4
Take the branch (PC = PC + sign_ext(imm)) if rs1 is
equal to rs2.
BNE B Branch if not equal
PC = (rs1 != rs2)? PC +
sign_ext(imm) : PC + 4
Take the branch (PC = PC + sign_ext(imm)) if rs1 is
not equal to rs2.
LW I Load Word
rd = mem[rs1 +
signa(imm)][31:0]
Load 32-bit value at memory address [rs1 +
signPext(imm)] and store it in rd.
SW S Store Word
data[rs1 +
sign_ext(imm)][31:0] =
rs2
Store the 32 bits of rs2 to memory address [rs1 value
+ sign_ext(imm)].
HALT - Halt execution
Instruction encoding:
Mnemonic
Bit Fields
31:27 26:25 24:20 19:15 14:12 11:7 6:0
ADD 0000000 rs2 rs1 000 rd 0110011
SUB 0100000 rs2 rs1 000 rd 0110011
XOR 0000000 rs2 rs1 100 rd 0110011
OR 0000000 rs2 rs1 110 rd 0110011
AND 0000000 rs2 rs1 111 rd 0110011
ADDI imm[11:0] rs1 000 rd 0010011
XORI imm[11:0] rs1 100 rd 0010011
ORI imm[11:0] rs1 110 rd 0010011
ANDI imm[11:0] rs1 111 rd 0010011
JAL imm[20|10:1|11|19:12] rd 1101111
BEQ imm[12|10:5] rs2 rs1 000 imm[4:1|11] 1100011
BNE imm[12|10:5] rs2 rs1 001 imm[4:1|11] 1100011
LW imm[11:0] rs1 000 rd 0000011
SW imm[11:0] rs2 rs1 010 imm[4:0] 0100011
HALT x x x xxx x 1111111
The simulator should have the following five stages in its pipeline:
● Instruction Fetch: Fetches instruction from the instruction memory using PC value as address.
● Instruction Decode/ Register Read: Decodes the instruction using the format in the table above
and generates control signals and data signals after reading from the register file.
● Execute: Perform operations on the data as directed by the control signals.
● Load/ Store: Perform memory related operations.
● Writeback: Write the result back into the destination register. Remember that R0 in RISC-V can
only contain the value 0.
Each stage must be preceded by a group of flip-flops to store the data to be passed on to the next stage in the
next cycle. Each stage should contain a nop bit to represent if the stage should be inactive in the following
cycle.
The simulator must be able to deal with two types of hazards.
1. RAW Hazards: RAW hazards are dealt with using either only forwarding (if possible) or, if not,
using stalling + forwarding. Use EX-ID forwarding and MEM-ID forwarding appropriately.
2. Control Flow Hazards: The branch conditions are resolved in the ID/RF stage of the pipeline.
The simulator deals with branch instructions as follows:
1. Branches are always assumed to be NOT TAKEN. That is, when a beq is fetched in the IF stage, the
PC is speculatively updated as PC+4.
2. Branch conditions are resolved in the ID/RF stage.
3. If the branch is determined to be not taken in the ID/RF stage (as predicted), then the pipeline
proceeds without disruptions. If the branch is determined to be taken, then the speculatively
fetched instruction is discarded and the nop bit is set for the ID/RR stage for the next cycle. Then
the new instruction is fetched in the next cycle using the new branch PC address.
Tasks:
1) Draw the schematic for a single stage processor and fill in your code in the to run the simulator. (20
points)
2) Draw the schematic for a five stage pipelined processor and fill in your code to run the simulator. The
processor should be able ot take care of RAW and control hazards by stalling and forwarding. (20
points)
3) Measure and report average CPI, Total execution cycles, and Instructions per cycle for both these cores
by adding performance monitors to your code. (Submit code and print results to console or a file.) (5
points)
4) Compare the results from both the single stage and the five stage pipelined processor implementations
and explain why one is better than the other. (5 points)
5) What optimizations or features can be added to improve performance? (Extra credit 1 point)
Your work will be evaluated against the 10 test cases, 3 of which will be revealed one week before thedeadline. (50 points - 5 points each)Useful References:
WX：codehelp