Paul Jakma wrote: > US IV+ and Opteron both have higher clock frequencies and larger L1 > caches per core and likely would be the better choice for compute > intensive codes, including integer only. > > Most definitely for FP codes. Plus both support SIMD instructions, VIS > for former (for which Sun offer an SDK to make it easy to use without > delving into asm), MMX/SSEn/3DNow for the latter - which iirc does have > 128bit wide registers, though unfortunately for you does not have much > in the way of 128bit arithmetical instructions.
Would it be possible that the next generation of SPARC CPUs will have 128bit integer registers ? Yes, I know - it will require another revision of the ABI similar to SPARCv8 vs. SPARCv8a - but it would also deliver some nice improvements for stuff like IPv6 processing, memory copies, 3D applications (e.g. you could use 128bit integers instead of (slower) floating-point for most 3D stuff) and many other applications... > > (outperforming the old E280R by more than factor three (E280R binary > > was compiled with -xO4, T2000 binary was compiled with -xO2 so this is > > no fair comparisation)), however the resolution isn't good enougth - > > we would need something like 64.64bit fixed point math - or better: > > Plain 128bit integers since it would help getting rid of the shifting > > madness in the code (which makes it a nightmare to maintain). > > There is a GNU Multi-Precision library (GMP - LGPL), which might save > you some work. Yes, I know... but last time I checked it didn't work with Sun Workshop and using gcc for HPC stuff is useless. > Though, it should be fairly easy to roll your own > multi-word integer arithmetic functions if needs be. Yes, but the code would become hard to maintain since this would be a platform/CPU-specific optimisation... ---- Bye, Roland -- __ . . __ (o.\ \/ /.o) [EMAIL PROTECTED] \__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer /O /==\ O\ TEL +49 641 7950090 (;O/ \/ \O;) _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
