submited by

Style Pass

This is a pretty straightforward method. This one can be done on VU1, VU0 micro mode or VU0 macro mode. For simplicity, I will do it with VU0 macro mode, where VU0 is used as a coprocessor. That way it can be done directly on the EE CPU without dealing with VU programs.

There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be accurate.

It’s a little bit loaded but the gist of it is, with multiplication operations 1 * X is not guaranteed to result in X, unlike X * 1 which is.

The exact reason why a bit is lost is not known to me, whoever decides to implement software floating point for the PS2 will have to figure that out and report back :^)

First things first, we need to figure out a number that triggers this issue. The easiest way to do that is to brute force it. Thankfully I’ve already done that in the past. The list of the first 250 numbers (with 0.5 increments) that have the issue can be found in a gist here.

Read more fobes.dev/ps...