Recip/Divide routines.

by **rcp** » Jul 17, 2003 @ 10:28pm

I've noticed a few people stating that recip tables are faster than doing a divide. While I understand the obvious here, I would have thought that the cache hit would negate much of the advantage.

What size recip tables are people using? Is there a routine out there that uses a small recip table and then some code to get a high bit answer? We usually are looking for about 14->16 bits for the recip. ie;

long invW= ((1<<30)/w)<<2; //invW is 16:16
verts->x=imul16(verts->x*invW);
etc...

Actually, everthing is done is ASM to take advantage of the 32x32->64 bit multiply (and accumulate) functions of the ARM.... but you get the idea.

Cheers,

rcp

by **bluescrn** » Jul 17, 2003 @ 11:20pm

by **cryo** » Jul 18, 2003 @ 8:27am

by **rcp** » Jul 18, 2003 @ 8:55am

by **bluescrn** » Jul 19, 2003 @ 8:56am

by **refractor** » Jul 19, 2003 @ 9:20am

by **Guest** » Jul 20, 2003 @ 7:17pm

a simple use of logarithms in optimization is to make use of this property:

log(a*b) = log(a) + log(b)

That way, you can code a multiplication (or division using log(a) - log(b)) with 3 table lookups. One for log(a), one for log(b) and one to execute the exponential function.

The advantage of such a system is that you can lookup any multiplication between two integers between 0 and n with log tables that have only n entries.

cryo

by **rcp** » Jul 20, 2003 @ 11:17pm

by **Sergey Chaban** » Jul 21, 2003 @ 1:40am

by **rcp** » Jul 21, 2003 @ 2:31am

by **refractor** » Jul 21, 2003 @ 1:00pm

by **Sergey Chaban** » Jul 21, 2003 @ 2:09pm

by **refractor** » Jul 21, 2003 @ 3:37pm

by **rcp** » Jul 22, 2003 @ 5:39am

by **rcp** » Jul 22, 2003 @ 5:40am