[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [Discuss-gnuradio] SDR on General Purpose Processor Demo
From: |
Ettus, Matt |
Subject: |
RE: [Discuss-gnuradio] SDR on General Purpose Processor Demo |
Date: |
Wed, 13 Mar 2002 18:13:37 -0800 |
> http://murl.microsoft.com/LectureDetails.asp?837
> >
Interesting presentation. You'll of course need a Windows machine to see
it, though.
Major points --
1 Computation is cheap
2 Cache Misses are expensive
3 Memory bandwidth is key
4 Optimize for common case, not worst case like in hardware
5 Ride Moore's law, let AMD/Intel/Motorola do the hard work
He also suggests 3 features for processors, since the lecturees are
in a domain-specific (we used to call that special-purpose, but I guess
that's out of fashion now) processors course.
Branch on random -- I think he's nuts here. Besides, good random
numbers are hard to get in the presence of major EMI on a processor chip
Hashed caches -- sounds cool, but the idea died long ago. He's
forgetting number 4 above... Hashed caches would slow down hit times.
Faster Integer MACs. I strongly agree here, and would add all math,
not just MAC. There's no reason for floating point to be faster than
integer.
Matt