Optimization question for bitblit optimization
bhurt at spnz.org
Thu Jan 29 15:38:26 CET 2004
This is in regards to the bitblit routine I posted yesterday. In general,
which should I be optimizing for: long bitblits (where the O(1) costs are
insignifigant, the dominate cost is the main loops), or short bitblits
(where the O(1) costs dominate)?
I'm looking at doing hand-tuned x86 versions of the main loops. What I've
come up with so far (generic 386 code) increases the speed per bit by 20%
for long (1K word) bitblits- 0.277 clocks/bit vr.s 0.345 clocks/bit. But
it slows down the short (<3 words) bitblits by 10%- 1.963 clocks/bit vr.s
I'm inclined to optimize for the long blits myself. But I wanted to give
the development community an opportunity to comment.
"Usenet is like a herd of performing elephants with diarrhea -- massive,
difficult to redirect, awe-inspiring, entertaining, and a source of
mind-boggling amounts of excrement when you least expect it."
- Gene Spafford
More information about the gmp-devel