<InPhase>
hramrach: Are you using the nightly for that? I know there was work done already on sorting the outputs for reliable testing purposes.
<InPhase>
hramrach: Off the top of my head I don't recall which year that was in, but it was a good bit ago.
<Scopeuk>
I believe the manifold dev (sorry terrible with names) thought the latest version of that engine (which I believe is now in the nightlies) had resolved all of that and anything left was configured a bug. I think the "sort vertex" experimental feature is still in as well
kintel has joined #openscad
kintel has quit [Client Quit]
<hramrach>
ugh, OpenSCAD version 2021.01
<hramrach>
was tehre no release since then?
<InPhase>
hramrach: It's being worked on. There are a lot of major improvements, which adds a bit more work to get a release cleaned up.
<InPhase>
hramrach: But there are snapshot releases which are pretty good, and might solve your problem.
<hramrach>
but not tagged in git
<InPhase>
Yeah, only main releases and RCs are getting tagged.
<InPhase>
hramrach: The nightly snapshots autorun every night. Or you can just self-build if capable. I primarily do that.
<hramrach>
only debian packages, not rpm
<teepee>
flatpak, snap, appimage
<InPhase>
hramrach: Just grab the appimage. They are superb.
<teepee>
we had rpm at some point, I took them off as nobody cared when they were multiple month broken
<InPhase>
hramrach: You just +x and run as-is, and you can archive large quantities of the appimages to compare versions. This is very handy.
<InPhase>
hramrach: My /usr/local/bin contains quite a lot of OpenSCAD appimage files, and then I just symlink my primary one.
<InPhase>
hramrach: I assume you know what that means and how to do it from your other comments, but feel free to ask if not. ;)
kintel has joined #openscad
<kintel>
hramrach ..but to answer your question: Yes, in the latest nighties do support deterministic output. Deterministic triangulation is enabled by default, and there's a cmd-line flag to also output deterministic vertex- and triangle order.
<kintel>
Note: Deterministic triangulation is only supported on the same machine when using Manifold. If you run on a different CPU, your results may have variations in triangulation
<hramrach>
what causes teh difference across different CPUs?
<hramrach>
Aren't floats standardized these days?
<kintel>
floats are standardized, but floating-point operations vary between CPUs
<kintel>
e.g. the result of a combined multiply+add vs. performing a mult, then add with temporary storage in-between won't yield the same result.
<kintel>
and CPUs have all kinds of different optimized float SIMD instructions these days.
<kintel>
..plus threading models may differ.
<kintel>
pca006132 may have a more nuanced answer :(
<kintel>
:) (typo)
kintel has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
<hramrach>
madd instructions
<hramrach>
yes, I recall that some arithemetic units have higher precision internally than what fits to the float register
<hramrach>
are the --enable parameters documented somewhere? It seems 'predictible-output' does make the output deterministic but don't see any description in the help
ToAruShiroiNeko has joined #openscad
<ali1234>
"IEEE 754-2008 specifies that it must be performed with one rounding, yielding a more accurate result."