LAM/MPI General User's Mailing List Archives

From: Tim Prince (n8tm_at_[hidden])
Date: 2007-04-10 08:54:07

• Next message: Heroux, Michael A: "LAM: The selected RPI failed to initialize during MPI_INIT message"
• Previous message: Yuan Wan: "LAM: How to optimize LAM on SMP machine"
• In reply to: Yuan Wan: "LAM: How to optimize LAM on SMP machine"

ywan_at_[hidden] wrote:
> Hi,
>
> I'm running benchmark code on single node of 4 cores with LAM-7.1.3.
> I'm wondering if LAM can be optimized to make use of the SMP shared memory
> feature. Anyone has good idea?
With a standard lam build, shared memory is used automatically. The -O
run-time flag can be used to avoid byte reversing the messages, when all
nodes are little-endian. Otherwise, there is a define to use at build
time, to use shifts for the byte reversing, which is much faster.

• Next message: Heroux, Michael A: "LAM: The selected RPI failed to initialize during MPI_INIT message"
• Previous message: Yuan Wan: "LAM: How to optimize LAM on SMP machine"
• In reply to: Yuan Wan: "LAM: How to optimize LAM on SMP machine"