[Beowulf] [rjdyer@vcu.edu: Re: Intel?]
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Greg Lindahl lindahl at pathscale.comWed Jun 8 10:58:57 PDT 2005
- Previous message: [Beowulf] [rjdyer@vcu.edu: Re: Intel?]
- Next message: [Beowulf] [gallatin@cs.duke.edu: Re: Intel?]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On Wed, Jun 08, 2005 at 11:21:34AM -0400, Mark Hahn wrote: > for instance, altivec has 32x128 and x86-64 has "only" 16x128. > but how relevant is this? studies of conventional reg-set usage > show diminishing returns after about 16 regs. These studies are always contaminated by compiler quality. If you look back you can find papers where the authors used a horrible compiler and "proved" that you didn't need more than 4 registers. Even with a really aggressive compiler, you can find many example apps that run out of registers with only 16. Then the question becomes: How many customers are hurt by this? -- greg
- Previous message: [Beowulf] [rjdyer@vcu.edu: Re: Intel?]
- Next message: [Beowulf] [gallatin@cs.duke.edu: Re: Intel?]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
