00:48 < matches> I will link to it anyway see
00:49 < matches> The odds of anyone actually following the link are slim
00:49 < matches> If it doesn't work I'll tell people to run it in wine
+09:33 < sulix> Those are damn nice graphs!
+09:34 < sulix> I take it there's no 'rat' measurement because it's too slow.
+09:35 < sulix> pxzoom is actually the same as the mouse, the mouse still does random expf() calls to work out the zoom amount.
+09:35 < sulix> (I thought I'd changed that, but apparently not)
+09:37 < sulix> Also, thank god we don't use 'sin': http://randomascii.wordpress.com/2014/10/09/intel-underestimates-error-bounds-by-1-3-quintillion/
+09:38 < matches> There should be a rat measurement?
+09:38 < matches> In the loss of precision graph, the path-rat program loses no precision
+09:39 < matches> But it's underneath all the other implementations that also lose no precision
+09:39 < matches> They are all on the same line...
+09:39 < matches> In the accumulated error graph, it is down near 10^{-2}
+09:39 < matches> I'm not sure why it isn't at zero...
+09:39 < matches> I'm just going to point vaguely at the expf function
+09:41 < matches> I'll probably add MPFR with 10 bits (half floats...) and X bits where X is a lot
+09:41 < matches> But not today
+09:41 < matches> My draft fleshes out a lot more when I uncomment things...
+09:41 < matches> I mean the idea of commenting them was that they'd be removed
+09:42 < matches> Also there are two sources of error in the Direct transformation
+09:42 < matches> I need to think about whether they are both equally bad or one is worse
+09:42 < matches> You need to apply the inverse view transformation to express the svg in document space
+09:43 < matches> *Then* you apply the view transformation to render it
+09:43 < matches> Either of those could be the one that breaks it, but I suspect the rendering transformation is worse
+09:46 < matches> Maybe some of the ipython figures don't have path-rat in them, I forget
+11:40 * sulix sighs.
+11:40 < sulix> cssubmit requires my abstract to be in ASCII, so the é in Bézier means cssubmit rejects it.
+11:51 < matches> Haha
+11:51 < matches> I got excited when Rowan said "Looks great to me" and then I realised he was talking about your abstract not my draft... :P
+11:52 < matches> That would have been impressive speed reading
+11:52 < matches> My abstract is sort of really terrible
+11:52 < matches> It clings to the whole "document formats" thing that the project was alledgedly about but has moved well away from
+11:53 < matches> It also promises that I'll compare arbitrary precision floats as well, so I guess I better hack something together with those...
+11:53 < matches> Maybe a graph showing how you can increase the precision but then the FPS goes down
+11:54 < sulix> That'd ve really good, actually.
+11:55 < matches> Yeah I think I did something similar for CQM once
+12:06 < matches> Hmm, I can probably hack together a CPU renderer only binary for windows
+12:06 < matches> He says, ignoring the tendency for the CPU renderer to shit itself and die
+12:07 < matches> It'd be a Cool Feature for it to fall back to the CPU renderer instead of just segfaulting
+12:08 < matches> Falling back to the backup renderer always sounds impressive
+12:08 < matches> Everyone loved it when HTR did that
+12:08 < matches> (And then I realised the "backup" renderer was actually better)
+12:17 < sulix> I've fixed (hopefully) the OpenGL version detection. It should now Fatal() when the driver doesn't have needed extensions.
+13:43 < sulix> Excellent: the é has gone through!
git.ucc.asn.au / ipdf / documents.git / blobdiff
UCC git Repository :: git.ucc.asn.au