Thread: [Algorithms] P3 Prefetching.
Brought to you by:
vexxed72
From: Jason R. <jas...@vo...> - 2001-01-29 16:56:55
|
Hi all This is a bit of a long-shot question. - I think I'm maybe asking the wrong mail list - (if I am mailing the wrong list - can anyone mail me saying where I should be asking ! :) heres the problem... I've been trying to get Pentium 3 Prefetching working for a while now and I'm a bit confused why it wont work. what I've got is an algorithm that reads short strips of memory (about 20 bytes each) from 6 seperate locations - performs a calculation using the data and writes a row of result values to a 7th address. the volume of data is huge - upto 1Gb in size. the output address is continuous (therefore I get no stalls on writing) BUT the addresses of the strips of input data is pretty much random for each successive strip (unavoidable) (therefore none of it is in any level of the cache) BUT - I know the address of the strips I'll be wanting after the current one. So I figured I ought to be able to prefech the 6 addresses whilst calculating on the current lines... I've put the prefetch calls into the loop but they only give a slight speed improvement. and when I look in VTune I see that I'm still stalling waiting for the data as I was before... and I have no explaination as to why the prefetches are not taking place... the loop takes about 200 cycles to use a strip of data so there ought to be plenty time for the prefetch to have completed... I'm most puzzled. is this perhaps related to the TLB ? (translation lookaside buffer) any clues would be most appreciated !!! Thanks in advance... Jason |
From: Christopher H. <chu...@ca...> - 2001-01-29 17:21:12
|
Hi all, Before I spend the time playing with this, I thought I'd might ask if anyone else has run into any problems before. Can you do a quad-tree clod algorithm with triangle nodes instead of square nodes. I can't see why you shouldn't be able to, but I thought maybe the guarentees you have that lod's between adjacent blocks don't differ by more than one may not hold true with triangles. Or maybe there are other problems that I haven't thought of. Thanks in advance, Chris |
From: Thatcher U. <tu...@tu...> - 2001-01-30 02:17:11
|
On Jan 29, 2001 at 11:21 -0600, Christopher Husband wrote: > > Before I spend the time playing with this, I thought I'd might ask if > anyone else has run into any problems before. Can you do a quad-tree clod > algorithm with triangle nodes instead of square nodes. I can't see why > you shouldn't be able to, but I thought maybe the guarentees you have that > lod's between adjacent blocks don't differ by more than one may not hold > true with triangles. Or maybe there are other problems that I haven't > thought of. Look up Binary Triangle Trees and/or ROAM... -- Thatcher Ulrich <tu...@tu...> == Soul Ride -- pure snowboarding for the PC -- http://soulride.com == real-world terrain -- physics-based gameplay -- no view limits |