libera/#commonlisp - IRC Chatlog
Search
3:50:24
hayley
I just got SIMD scanning working in my regex engine. It is now about three times as fast as Hyperscan at finding resolutions in my Xorg.0.log apparently.
3:52:24
White_Flame
hobo: are you sure you used raw telnet for muds, and not something like tinyfugue?
4:09:36
phantomics
Interesting thing I just discovered in LispWorks: (expt 10.0d0 10) gives 1.000000000000004D10, the 4 is not there in other CLs. Causes problems when taking the ceiling of the result
4:11:35
Nilby
hayley: just don't sell out to intel, or if you do make sure to get a big ask like qpx
4:13:47
hayley
Not quite as "feature"ful, given that I have to generate a DFA from everything, but it's pretty fast in return, even without SIMD.
4:14:41
phantomics
Cool, I'll have tackle SIMD at some point for April, I take it the SIMD is using define-vop and only works with SBCL?
4:16:46
semz
phantomics: A program that relies on this exact behavior sounds misdesigned to me, the same way a program using exact FP comparisons would.
4:18:11
phantomics
It also has a tendency to lock up a lot during intensive April computations, maybe it's a problem with the multithreading
4:22:12
phantomics
ECL sometimes locks up once or twice when doing the intensive demo test suite, but LW locks up a dozen times or so needing manual restarts. SBCL, CCL, and ABCL get through it fine.
4:23:24
beach
I think you should report it to LispWorks. I am sure they will appreciate the feedback.
4:27:15
beach
I don't think so. Martin Simmons seems like a very reasonable person. At least when we have diner together at ELS. :)
6:31:36
White_Flame
phantomics: CLHS says that "a floating point approximation might result" for anything but (expt <rational> <integer>)
7:49:56
jackdaniel
n.b I'm not "a fan of stable things" - most notably I'm not that cool (a pun towards the word "fan"), but changing stable apis breaks existing code - mind that no common lisp implementation changed nth argument order to match elt, even if someone could have thought that it is a good idea :)
7:51:03
jackdaniel
or that changing the interface of with-output-to-presentation in a non-backward compatible way would break many preexisting clim applications
8:06:03
beach
jackdaniel: It would be much easier to parse what you wrote if you would capitalize NTH, ELT, API, and CLIM.
11:25:08
pjb
jackdaniel: beach: would I suggest to use instead 𝐧𝐭𝐡, 𝐞𝐥𝐭, 𝐀𝐏𝐈, 𝐂𝐋𝐈𝐌, given the flame you can get when using upper case?
13:16:29
pdietz
phantomics: in SBCL, expt will sometimes give different results on the same arguments depending on whether they appear as variables or constants in the form.