Last weekend I was visiting Pat, the surveyor who asked me to write
PerfectTIN, and he gave me a 56 Mpoint cloud of some site. I ran 0.4.0rc1 on
it and found some bugs that showed up when processing a point cloud that big,
which I fixed. (The largest cloud I'd run it on before has 13 million points.)
He showed me the spikes when an earlier version processed it and verified that
0.4.0rc1 does not produce spikes. I just released 0.4.0rc2 with the bugs fixed.
The reason I'm not calling this 0.4.0 is that twice, once each on two different
computers, the program crashed, but did not dump core because of ulimit. I ran
the debugger, but could not get it to crash again. I ran it without the
debugger, but it still didn't crash. Could you run it in the debugger and see
if it crashes? https://github.com/phma/perfecttin
I'm planning to implement speed improvement for 0.5.0. When it starts
processing a point cloud, each of the six triangles has a huge number of dots,
and it can't use more than one thread, because it has to lock all the
triangles, since they share a corner. So I'm going to partition the work into
tasks of about 65536 dots each and let any thread that would otherwise be
sleeping work on a task. Later I'd like to have the GPU, if present, process
tasks, but I don't know how. Do any of you know how to use OpenCL to write GPU
Por H o por B, los campos magnéticos se difieren dentro de un imán.
I don't know if it's a memory problem, since I couldn't reproduce the crash. I
ran it in Valgrind, but that was to find out how much time it was spending
checking whether triangles are in tolerance. The bug may have disappeared when
I cleaned some lint or made some other modification.
Lanthanidia deliciosa: What the kiwifruit would be
if it weren't so radioactive.