Re: MMX speedup for Floyd Steinberg error diffusion

On May 9, 10:03 pm, Terje Mathisen <spamt...@xxxxxxxxxx> wrote:
Indeed.

I was sure that I had the victory to "Dr Dobbs Second Annual Code
Optimization Challenge" all sewn up:

I had very code data compression, I worked on four cells in parallel,
and I used the minimum possible (i.e. 3) updates to modify the
surrounding cells when anything changed.

David Stafford didn't only beat me, but he did so by a factor of two, by
taking advantage of the fact that the 486 target cpu had an 8 KB L1 cache.

My algorithm did less work, and was faster when a lot of cells were
alive, but after the first 200 generations or so, David's working set
started to fit in L1 cache, and he blew me away from that point on. :-)


On May 9, 10:03 pm, Terje Mathisen <spamt...@xxxxxxxxxx> wrote:

Indeed.

I was sure that I had the victory to "Dr Dobbs Second Annual Code
Optimization Challenge" all sewn up:


Wasn't this the one that Mike Abrash wrote about in his Graphics
programming black book?

Recently I came across this game of life algorithm called HASHLIFE,
its been there for quite a while, I wonder how come no one used that
in that challenge.

It apparently can calculate millions of generations per second
(displaying a frame only once in a while). Theres a nice
implementation called golly, which is really fun to play with.

On an other note, I havent seen anything on the web about Michael
Abrash ( apart from some reference on radgametools.com and the fact
that he is ( was? ) in the X-BOX team ). I wonder where he is now.

Vivek

.