Eric Marsden <[EMAIL PROTECTED]> writes: > >>>>> "gm" == Gerd Moellmann <[EMAIL PROTECTED]> writes: > > gm> Another data point: The code below, which is my personal direct > gm> translation of the C code, takes 0.5s to run (x86 P4 box, CMUCL from > gm> CVS. FreeBSD -STABLE). The C version, compiled with GCC 2.95.4 takes > gm> 0.43s. > > the problem looks like it will be limited by memory bandwidth, so it > should be possible to make CMUCL run close to C speeds, without having > generated optimal instruction sequences.
For this specific case, memory access is probably a relevant contribution (let's say 30%). But think about applying such filters several times (which is quite usual). Then there are clever ways of doing all applications by only one sweep through memory with only few cache misses. Nicolas.
