P-Fetch: Prefetching Guided by Cross-Instruction Dynamic Access
Patterns on the Power5
- funded by: IBM
- funding level: $20,000
- duration: 09/01/2006 - 08/31/2007
- PIs: Frank Mueller
With processor speeds growing much faster than memory, a large
fraction of application execution time is increasingly spent waiting
for the memory system to deliver data. In this proposal, we present a
data prefetching scheme to ameliorate this problem using hardware
support to efficiently identify predictable references to scalars and
non-scalars that may benefit from prefetching.
"PFetch: Software Prefetching Exploiting Temporal Predictability of
Memory Access Streams" by J. Marathe and
F. Mueller in Workshop on MEmory performance: DEaling with
Applications, systems and architecture, Oct 2008 (accepted).