Date: Wed, 8 May 1996 10:02:20 +1000 From: Bruce Evans <bde@zeta.org.au> To: rgrimes@GndRsh.aac.dev.com, wscott@ichips.intel.com Cc: asami@cs.berkeley.edu, bde@zeta.org.au, culler@cs.berkeley.edu, current@freebsd.org, ken@area238.residence.gatech.edu, marc@bowtie.nl, nisha@cs.berkeley.edu, pattrsn@cs.berkeley.edu, wollman@lcs.mit.edu Subject: Re: more on fast bcopy Message-ID: <199605080002.KAA12703@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>> size bandwidth >>... >> 4096 69.565217 MB/s >> 8192 71.111111 MB/s >Thats pretty cruddy for 4 way interleaved memory, >1/120nS * 8 bytes/leave * 4 leaves == 266MB/sec. Erm, divide by 2 for read+write. >And I used the worst case there by using the 120nS full cycle time, >if these are sequentail accesses that hit the same DRAM page this >number should be up closer to 500MB/sec. Mabye this accounts for some of the advantages of the FP method - it's easy to read 8*8 bytes off a page at a time, while using integer registers it's impossible to read more than 5*4 bytes at a time (even using %ebp and %esp as scratch registers :-). Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605080002.KAA12703>