freenode/#sbcl - IRC Chatlog
Search
20:36:55
mfiano
How would I remove the notes from this function? Not sure why the range is expected to be very high here: https://gist.github.com/mfiano/66c0c1897c190e68438d405f604ddb20
20:42:02
mfiano
So I gues some transform makes it impossible to get the integer portion of a limited domain single-float in fast code?
22:01:43
scymtym
i saw that (floor (floor x) y) is truncate then integer division while (floor x y) is float division then truncate
22:02:16
scymtym
so there is a difference, but i'm not sure what problem in terms of derived types you are seeing
22:03:31
mfiano
I don't think there is any problem, apart from the fact that I have to explicitly denote the domain of the float to keep it in non-bignum territory, but that's expected
22:05:22
scymtym
there may still be a problem in that the first two variants (that is SINGLE-FLOAT with vs. without explicit bounds) in https://techfak.de/~jmoringe/sbcl-floor-float-derivation.png produce different results
22:06:24
scymtym
SINGLE-FLOAT derives INTEGER as the overall result, (single-float <bounds>) derives (integer <bounds>)
22:08:11
mfiano
The last few times I did, it crashed my display driver, exiting out of Xorg, with not even a core dumped or trace
22:09:42
scymtym
sorry to hear that. i wouldn't know where to start with something like that. but yeah, more appropriate for #clim if you want to attack the problem again in the future
22:11:13
mfiano
Thanks for the help everyone. I appreciate it. It was one of those days a stupid mistake caused me to try everything but see the obvious mistake in hindsight
22:16:31
scymtym
maybe NUMERIC-TYPE->INTERVAL can add missing bounds for specific floating point types