CMSC 611, Fall 1998
Assigned: Tuesday, October 20, 1998
Due Tuesday, October 27, 1998 (in class)
- Problem 4.4 in the text. In addition, calculate the speedup gained
from unrolling and scheduling the loop.
- Problem 4.8 in the text.
- You are trying to improve the speed of a CPU with branch prediction
by using a branch prediction buffer. The misprediction penalty
for the buffer is 4 cycles, and the miss penalty for the buffer
is 2 cycles. Branches successfully predicted do not stall the
processor. If branches hit in the buffer 85% of the time and the
buffer is 90% accurate, how much faster (or slower) is this scheme
than a CPU that has a fixed penalty of 2 cycles for taken branches
and 0 cycles for non-taken branches? Assume that branches make
up 15% of the instructions executed and that the base CPI is 1.2.
- Problem 4.20 from the text.
- Problem B.2 from the text.
- Problem B.3 (parts a-e only).
12 Oct 1998