...Behavior
1st revision June 26, 1994. This work was supported in part by the Office of Naval Research under contract # N00014-94-1-0006
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...functions
We will use the term function rather than procedure, subroutine, subprogram, or other equivalent notions.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...blocks
A basic block is a sequence of instructions where only the first instruction may be preceded by a label and only the last instruction may be a transfer of control.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...misses.
This additional requirement is a correction to the version of this paper published in the LCTS'94 workshop.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...cycles.
The semantics has been changed for a set fetch-from-memory bit since the publication in the LCTS'94 workshop.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...cleared
It is possible in such a situation that the merged instruction could be safely classified as a first-miss and have its bit cleared. An example of this situation is the first instruction in block 8 of Figure 3. It is the authors' intention to analyze the control flow to recognize these situations in the future.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...respectively
h'and m' are approximately the same as the number of instructions executed with the fetch-from-memory bit clear and set respectively with the exception of first-misses which are counted as misses on the first reference and hits on subsequent references.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...n=9.
The latency for a memory fetch is assumed to be n=9 cycles, a cache look-up takes one cycle, and thus a cache hit also consumes one cycle while a miss takes n+1=10 cycles. These assumptions are described as realistic by other researchers [, ]. A memory fetch in an uncached system fetches exactly one instruction while a memory fetch in a cached system fetches a line of 4 instructions. Fetching a line of multiple instructions is typically accomplished through a wider bus between cache and main memory for a cached system.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Robert Palmer
Mon May 19 10:08:14 EDT 1997