13 Apr
2006
13 Apr
'06
9:09 p.m.
On 4/13/2006 at 7:28 PM Richard wrote:
Browse around these places and then you can't tell
me there's noone left
who understands how to squeeze cycles out with careful thought!
Ah, but small size does not equate to fast code--or even the fastest code
for a simple task.
Take the lowly job of moving some bytes on a Pentium-class processor.
Anyone can write "rep movsb', but while very small, it's not usually the
fastest. What's the fastest byte string move on a Z80? (Assume a
non-overlapping move).
Clearly, in many applications "rep movsb" will be perfectly adequate and
won't affect the performance of a program one bit. So not every lily
should be gilded.
I'm more impressed with a cleverly fast FFT that betrays the coder's level
of understanding of the problem than an implementation that slavishly
implements the basic algorithm in the smallest number of bytes.
Cheers,
Chuck