3 ms·
A while back, Noah Vawter was using this to make ultra-fast math at the cost of accuracy. [1] He computed each bit of an output function such as addition and m
by chillingeffect 9y ago
A while back, Noah Vawter was using this to make ultra-fast math at the cost of accuracy. [1] He computed each bit of an output function such as addition and multiplication as a function of any two of the N inputs of both operands. Then generated approximations using single-operator functions (and, or, nand, etc.) that minimized error. The result was single-cycle approximations.
For example, he claims 44% accuracy (not sure how this is measured) for 7-bit multiplies in a single cycle.
[1] http://gweep.net/~shifty/ http://gweep.net/~shifty/
- gp2000 9y agoDo you have a more direct link? That one just goes to Noah's home page and for the life of me I couldn't navigate to the ultra-fast math page nor find it via Google. Edit: I think I've found it: http://www.gweep.net/~shifty/portfolio/oddsvf/index.html http://www.gweep.net/~shifty/portfolio/oddsvf/index.html
- JadeNB 9y ago> [1] http://gweep.net/~shifty/ http://gweep.net/~shifty/ Eep, that's an eye-burning page. Where should I go for discussion of the fast-but-inaccurate techniques you mention?