11:49RSpliet: nvuser: interesting, the PLL nouveau chooses has a massive multiplier
11:50RSpliet: *PLL coefficients
11:52RSpliet: the VBIOS says the PLL should be coping just fine with such coeffs
12:00RSpliet: 1006MHz is a PITA to get to though. Dividing 2012 by 405 gives 4,967901235 - which means you need a big multiplier to get to something that ends with ~xyz,000000. And the divider only goes up to 32, unless you use the P divider multiplier. Better coeffs would be n = 154, m = 31.
12:00RSpliet: Wonder if the blob would select sth like that if asked to set 1006MHz
17:57fdobridge: <gfxstrand> We'll be moving a lot of codegen magic to NIR.
20:37nvuser: RSpliet: sounds complicated hah, I guess it's more informational then if I provide that in an issue