freenode/#sicl - IRC Chatlog
Search
4:47:43
ck_
my phase of having a lot of time for things like this is coming to a close (hopefully -- wish me luck). If you'd like me to do anything in particular before then, with the ryu thing for example, it would be good to know.
5:54:50
ck_
At this point, even most of the comments are ported straight from the C source. I'll happily change those.
5:55:43
beach
I would try to make the lines 80 or at most 90 characters. And I would do the comments as complete sentences.
6:00:52
ck_
Yes -- well, the simple algorithm is https://github.com/kc-/ryu-cl/commit/d28056911c7c39acddb82780e37afce6b215fffc
6:03:13
ck_
I'm saying I started with the simple algorithm when I first read the paper, because I intended to go from paper alone.
6:03:46
ck_
Then I hit a snag when some algorithm turned out to not behave as described (we talked about this briefly), and I did the rest from porting the C source
6:04:22
beach
Yes, I remember you said that. I see. So the "simple" algorithm in the paper doesn't quite work. Got it.
6:05:50
ck_
The algorithm is, in principle, still the same, just like the early version I just gave the URL for.
6:06:46
ck_
The only difference, and also the part that accounts for all the code, is that instead of getting at the digits one by one and then doing the 'normalization' (by that I mean finding the shortest decimal string that resolves to that float)
6:07:27
ck_
.. a lot of digits that will not be relevant anyway are cut off by a division. To find this divisor is not that straightforward, but it is the main reason for the speed of ryu
6:09:29
ck_
did you have a chance to watch his video? That might be quicker (you always seem very busy)
6:10:15
beach
I have not had time for that. And my desktop probably crashed a few times, so I lost the link.
6:13:15
beach
It needs better names (u, j, v, i, , vr, vp, vm, etc), but there might not be an easy fix for that.
6:14:01
ck_
v[rpm] I can rename 'easily' because I know exactly what's going on. the single-character variables, I don't have a good name for
7:52:37
ck_
I have the following on my list for some longer train rides in the next days: - better names for vm/vr...; - description of the state of the code in readme.md (for example, tests)
7:53:21
ck_
After that, I'll probably leave it -- unless you have some other suggestions -- and hope that it will be useful some time in the future.
7:59:15
beach
So am I understanding correctly that you have been unemployed, and you are about to start a new employment?
8:02:55
ck_
Yeah, I guess as well :) I changed priorities with respect to that, I'll see how it goes.
8:04:00
ck_
Yes. The change I meant is that previously, I had tried very hard to work in Common Lisp, that was my main objective.
8:05:12
ck_
That probably won't happen again soon. So that's another reason I finally picked up contributing to open source -- to satisfy that urge
8:10:09
ck_
So, how about you, are you still riding the enthusiasm from the speedup introduced by the switch to an interpreter?
8:10:54
beach
I diverged a bit into debugging and Clordane, but I'll be back on HIR-to-MIR and MIR-to-LIR very soon.
8:13:34
beach
Now that I know how to handle multiple values, I think I have removed all the obstacles for generating native code.