Description
Consider the following code for the following problems
loop:
LW R4, 0(R3)
ADDI R3, R3, 4
SUBI R1, R1, 1
b1:
BEQ R0, R4, b2 ADDI R2, R2, 1
b2:
BNE R0, R1, loop ADDI R3, R3, 4
Assume R3 points to the array with all 0 entries.
R1=R2=5.
# R0 is zero
-
If there is a branch delay slot and result of R2 stays even after branch is taken. What is the value of R2 when the loop finishes?
-
If there is no branch delay slot, what is the value of R2 when the loop finishes?
-
Which behavior makes the most sense?
-
Now assume we use a global branch-predictor with no branch delay slot. Assume
R1 starts out with value 10. R3 points to the array with all 0 entries. How well does the predictor work?
-
Now assume the array has all 1 entries. How well does the predictor work?
-
Now assume the array has alternating 0’s and 1’s in the array entries (starting with 0’s). How well does the predictor work?
-
Consider the following ROB and how the above code will be executed with branch prediction assuming it is predicted as in (6). Fill the table and trace through how it might work with n = 4.
Reorder Buffer (ROB)
use ex op p1 PR1 p2 PR2 Rd LPRd PRd
|
→ |
|||||||||||
|
next to |
|||||||||||
|
commit |
|||||||||||
Assume R1->P0, R2->P1, R3->P2, R4->P3
Free physical registers: P4 – P12
-
How bits are each column in the ROB table to support standard 32-bit MIPS architecture?
9) What type of hazards are detected by ROB and avoided?
2
