High Speed low Precision computations?

[nick__]

I was wondering if there is a way to allow computations to be performed faster with lower precision. Some c++ compilers have an option for doing quick vs. precise computations on floating point values and I was wondering if MKL supported something like this. I am trying to get a dense matrix multiplication with single precision values to run as quickly as possible. I am using Intel MKL 9.1 within Matlab r2008a.
Thanks,
Nick

[crispybits]

Not really an answer, but I usually find that when you'retrying to squeeze the last 5% of optimization out of something, the better solution is to change things at a higher level. Like maybe representing the problem with smaller matrices. I have plenty of code that I've spent days tweaking to speed it up, then eventually totally discarded it for enterely new solutions that are already fast without tweaks.

However, have to tried aligning the data bytes in the way MKL reccommends?

Quoting - nick__

I was wondering if there is a way to allow computations to be performed faster with lower precision. Some c++ compilers have an option for doing quick vs. precise computations on floating point values and I was wondering if MKL supported something like this. I am trying to get a dense matrix multiplication with single precision values to run as quickly as possible. I am using Intel MKL 9.1 within Matlab r2008a.
Thanks,
Nick