i nerd sniped myself tonight and i imagine the NSA operative who is assigned to me is very confused by the increasingly erratic and frustrated google searches for HOW DO I CALCULATE THE MOON WHERE IS THE MOON
-
@jon_valdes @dotstdy my laptop has a recent-ish Xe gpu
-
-
@aeva @jon_valdes generally cpu trig is not so bad, but does vary between platforms if you're just calling the c stdlib functions (but we're talking differences of 1 ulp or so). so yes, absolutely you can get a warmer fuller soundstage by switching to a different sin.
@dotstdy @jon_valdes maybe I'll call the expensive exact soft sin oscillator "gold plated sin"
-
@dotstdy @jon_valdes maybe I'll call the expensive exact soft sin oscillator "gold plated sin"
@aeva @dotstdy @jon_valdes start a new genre "deca-dance"
-
@aeva @dotstdy @jon_valdes start a new genre "deca-dance"
@demofox @dotstdy @jon_valdes well so far I can't tell the difference by ear between the soft sin and the double precision sin that clang gives me
-
@demofox @dotstdy @jon_valdes well so far I can't tell the difference by ear between the soft sin and the double precision sin that clang gives me
@demofox @dotstdy @jon_valdes assuming I didn't screw anything up adapting Jon's shader toy, the 30 iteration soft sin also has the surprising property of being able to clip out of the expected -1.0 to 1.0 output range https://github.com/Aeva/mollytime/blob/af58e9237903856b6193a116ae6408038dc47d4a/src/patch.cpp#L319
-
@demofox @dotstdy @jon_valdes assuming I didn't screw anything up adapting Jon's shader toy, the 30 iteration soft sin also has the surprising property of being able to clip out of the expected -1.0 to 1.0 output range https://github.com/Aeva/mollytime/blob/af58e9237903856b6193a116ae6408038dc47d4a/src/patch.cpp#L319
@aeva @demofox @dotstdy @jon_valdes There's lots of versions of CPU sin/cos as well. The fastest is basically identical to the GPU one - about 12 bits of precision, IIRC. Then they do iteration internally if you want higher precision. There's no magic - everything costs!
As you noticed, sometimes it's better not to call sin-vs-cos, because you're not guaranteed to get magnitude 1.0. In those cases it's better to get sin and derive the other by doing sqrt(1-sin^2).
-
@aeva @demofox @dotstdy @jon_valdes There's lots of versions of CPU sin/cos as well. The fastest is basically identical to the GPU one - about 12 bits of precision, IIRC. Then they do iteration internally if you want higher precision. There's no magic - everything costs!
As you noticed, sometimes it's better not to call sin-vs-cos, because you're not guaranteed to get magnitude 1.0. In those cases it's better to get sin and derive the other by doing sqrt(1-sin^2).
@aeva @demofox @dotstdy @jon_valdes Also, if you care about precision at all, do range reduction yourself beforehand. Otherwise you have no idea how it's being done internally.
-
@aeva @demofox @dotstdy @jon_valdes Also, if you care about precision at all, do range reduction yourself beforehand. Otherwise you have no idea how it's being done internally.
@TomF @demofox @dotstdy @jon_valdes I already do the range reduction. I learned that the hard way once upon a time XD