<korvo>
Is there a blessed way to control what IEEE 754 instructions do, e.g. from math.sin(), or am I on my own?
<korvo>
The context is that I want to expose/control, at user level, the difference between IEEE 754 processor instructions and the hundred-line macro-instructions in rlib.
<korvo>
Fully aware that there will be differences in accuracy; that's part of the goal.
<korvo>
Also fully aware that I might have to go down to libc to figure this one out.
<korvo>
I guess I should zoom out. One of my Cammy demos is raytracing/animating. Some demos are in the megapixel/second range, which means that individual pixels do not have a lot of room for overhead in processing.
<korvo>
I would like to be able to control whether I'm getting e.g. a cheap fast native sine or a fancy low-error high-precision sine, as if I were controlling a GPU or DSP.
<korvo>
(BTW are other folks interested in GPUs? I've written a GPU driver, I could imagine writing more GPU code...)
glyph has quit [Quit: End of line.]
glyph has joined #pypy
itamarst has quit [Quit: Connection closed for inactivity]
jcea has quit [Ping timeout: 272 seconds]
dmalcolm_ has quit [Ping timeout: 252 seconds]
dmalcolm has joined #pypy
<cfbolz>
korvo: you're on your own, I think
<korvo>
Fair enough, no worries. I got lolremez working again, so that should make it much easier to write a cheap fast non-native sine. I'll do some reading on the libc side of things.