Optimizing Capacitance and Switching Activity to Reduce Dynamic Power - Computer Architecture Techniques for Power-Efficiency - page 101

DATA DATA DATA DATA

Step 1

Step 2 Compare

MUX

Step 3

DATA DATA DATA DATA

Step 1

Step 2 Compare

MUX

Step 3

DATA DATA DATA DATA

Step 1

Step 2 Compare

MUX

Step 3

FIGURE 4.25: Sequential access: start from the MRU way and look sequentially.

cache (Calder et al. [ 43 ]), convert a direct-mapped cache into a 2-way set-associative cache
by mapping conflicting lines to two separate sub-banks.
The lines are still accessed with the benefit of direct-mapping: a single probe checks the
tag and reads the data from the most likely location. But if the requested data are not found,
an alternative location is also probed. The three proposals differ on how they choose the most
likely location for the first access and what they do to improve subsequent accesses, with the
PSA proposal fully decoupling location prediction (using a separate prediction structure)
from replacement decisions [ 43 ]. Although these proposals have substantial differences
with respect to power consumption from the corresponding set-associative organizations,
no work is available to systematically compare the various options.

4.9.3 Way Prediction
In 1999, Inhue, Ishihara, and Murakami [ 109 ] recast the Kessler et al. work on sequen-
tially accessed set-associative caches [ 133 ] and the Calder et al. PSA scheme [ 43 ], having