Advanced Computer Architecture CMSC 611

Homework 3

Due in class Oct 17th, 2012

(Show your work to receive partial credit)

1) For the following code snippet list the data dependencies and rewrite the code to resolve
name dependencies. (15 points)

Loop: LD R2, 0(R7) ----- 0

Add R1, R2, R3 ----- 1
Sub R4, R6, R1 ----- 2
Add R2, R5, R6 ----- 3
LD R4, 32(R1) ----- 4
SD 36(R1), R2 ----- 5
BEQ R4, Loop ----- 6

Data Dependency – 1 & 2, 1 & 4 and 1 & 5 on R1

0 & 1, 3 & 5 on R2

4 & 6 on R4

Name Dependencies- Anti-dependence between 1 and 3 because of R2

Output dependence between 2 and 4 because of R4

Output dependence between 0 and 3 because of R2

Name dependencies can be resolved by renaming registers R2 and R4 with temporary

registers T2 and T4

Loop: LD R2, 0(R7) ----- 0

Add R1, R2, R3 ----- 1
Sub R4, R6, R1 ----- 2
Add T2, R5, R6 ----- 3
LD T4, 32(R1) ----- 4
SD 36(R1), T2 ----- 5
BEQ T4, Loop ----- 6
2) From the course slides. (Multi-cycle FP pipeline for MIPS) (45 points)

 Use the instruction latencies as indicated in the selected slide to first show all the stalls
that is present in the following piece of code if the branch is not taken.
 Now unroll this loop as many times as needed and schedule the instructions to remove all
the stalls. You may rename registers i.e. use new registers and/or change the
immediate/offset values. You can ignore structural hazards as well.
 What is the speed up that you achieved after unrolling and scheduling?
Loop: LD F0, 0(r0)
LD F2, 0(r2)
MULTD F4, F0, F2
LD F6, 0(r4)
ADDD F8, F4, F6
SD 0(r6), F8
ADDI r0, r0, #4
ADDI r2, r2, #4
ADDI r4, r4, #4
ADDI r6, r6, #4
SUBI r8, r8, #1
BNEZ r8, Loop
 Clock
Loop: LD F0, 0(r0) 1
LD F2, 0(r2) 2
Stall
MULTD F4, F0, F2 4
LD F6, 0(r4) 5
Stall
Stall
Stall
Stall
Stall
ADDD F8, F4, F6 11
Stall
Stall
Stall
SD 0(r6), F8 15
ADDI r0, r0, #4 16
ADDI r2, r2, #4 17
ADDI r4, r4, #4 18
ADDI r6, r6, #4 19
SUBI r8, r8, #1 20
Stall
BNEZ r8, Loop 22
Stall 23

Unrolling it twice is sufficient to remove all the stalls.

Loop: LD F0, 0(r0) 1

LD F2, 0(r2) 2
LD F10, 4(r0) 3
LD F12, 4(r2) 4
MULTD F4, F0, F2 5
MULTD F14, F10, F12 6
LD F6, 0(r4) 7
LD F16, 4(r4) 8
ADDI r0, r0, #8 9
ADDI r2, r2, #8 10
ADDI r4, r4, #8 11
ADDD F8, F4, F6 12
ADDD F18, F14, F16 13
 ADDI r6, r6, #8 14
SUBI r8, r8, #2 15
SD -8(r6), F8 16
BNEZ r8, Loop 17
SD -4(r6), F18 18

Unrolled loop is (23*2)/18 = 2.55 times faster than without unrolling with respect to
clock cycles consumed.
(Ignoring pipeline fill time)
CPI for a) is 23/12 ≈ 2
CPI for b) is 1
After unrolling the CPI is now the same as the ideal CPI.

3) Tomasulo’s Algorithm
Consider the following specifications. (40 points)

FU type cycles in EX Number of FUs

Integer 1 1
FP adder 5 1
FP multiplier 8 1
FP Divide 24 1

Assume the following.

 Functional units are not pipelined.
 All stages except EX take one cycle to complete.
 No limit on reservation stations.
 There is no forwarding between functional units. Both integer and floating point
results are communicated through the CDB.
 Memory accesses use the integer functional unit to perform effective address
calculation.
 All loads and stores will access memory during the EX stage. Pipeline stage EX
does both the effective address calculation and memory access for loads/stores.
 There are unlimited load/store buffers and an infinite instruction queue.
 Loads and stores take one cycle to execute. Loads and stores share a memory
access unit.
 If an instruction is in the WR stage in cycle x, then an instruction that is waiting
on the same functional unit (due to a structural hazard) can begin execution in
cycle x , unless it needs to read the CDB, in which case it can only start executing
on cycle x + 1.
  Only one instruction can write to the CDB in a clock cycle. Branches and stores
do not need the CDB since they don’t have WR stage.
 Whenever there is a conflict for a functional unit or the CDB, assume program
order.
 When an instruction is done executing in its functional unit and is waiting for the
CDB, it is still occupying the functional unit and its reservation station. (meaning
no other instruction may enter).
 Treat the BNEZ instruction as an Integer instruction. Assume LD instruction after
the BNEZ can be issued the cycle after BNEZ instruction is issued due to branch
prediction

Fill in the execution profile for the code given in the table which includes the
cycles that each instruction occupies in the IS, EX, and WR stages and comments
to justify your answer such as type of hazards and the registers involved.

# Instruction IS EX WR Comments
1 LD F0, 0(r0) 1 2 3
2 ADDD F2, F0, F4 2 4-8 9 RAW on F0 from #1
RAW on F4 from #2
3 MULTD F4, F2, F6 3 17-24 25 Structural Hazard from MULT at #7
(Only one functional Unit)
Structural Hazard from Adder
4 ADDD F6, F8, F10 4 9-13 14
Add instruction #2
5 DADDI r0,r0, #8 5 6 7
6 LD F1, 0(r1) 6 7 8
7 MULTD F1, F1, F8 7 9-16 17 RAW on F1 from #6
Structural Hazard from adder
8 ADDD F6, F3, F5 8 14-18 19
Add instruction #4
9 DADDI r1, r1, #8 9 10 11