Adds 3D winding number example to quest #1768
Conversation
kennyweiss
added
enhancement
| SLIC_ASSERT_MSG( | ||
| isValidParameter_u(u, 1e-5), | ||
| axom::fmt::format("Requested u-parameter {} is outside valid range [{},{}] with tolerance {}", | ||
| u, | ||
| getMinKnot_u(), | ||
| getMaxKnot_u(), | ||
| 1e-5)); | ||
| SLIC_ASSERT_MSG( | ||
| isValidParameter_v(v, 1e-5), | ||
| axom::fmt::format("Requested v-parameter {} is outside valid range [{},{}] with tolerance {}", | ||
| v, | ||
| getMinKnot_v(), | ||
| getMaxKnot_v(), | ||
| 1e-5)); |
There was a problem hiding this comment.
@jcs15c and I spoke about this -- this is triggered called when evaluating the normal for a trimmed patch, and was failing for some files due to trimming curves that are slightly outside the parametric bounding box. Since the user has no control over this, we should probably preprocess/fix this before processing the NURBS patches,
There was a problem hiding this comment.
To further keep record of this conversation, at this point it's unrealistic to clip or clamp the trimming curves to the correct parameter space for a given patch. Instead, the methods in NURBSPatch (which already clamp the inputs to the parameter space) should probably give a one-time warning if out-of-bounds input is recorded, then move on. Since #1765 is mostly concerned with the STEPReader, I'll make those changes here instead.
kennyweiss
force-pushed
the
feature/kweiss/wn-example-3D
branch
2 times, most recently
from
4d0dd2d to
59746cf
Compare
| "NURBSPatchGWNCache::getKnots_v() must match NURBSPatch signature"); | ||
| static_assert(std::is_same_v<decltype(std::declval<const Cache&>().getTrimmingCurves()), | ||
| decltype(std::declval<const Patch&>().getTrimmingCurves())>, | ||
| "NURBSPatchGWNCache::getTrimmingCurves() must match NURBSPatch signature"); |
There was a problem hiding this comment.
These are great tests, thank you! It's worth keeping in mind that a future implementation of NURBSPatchGWNCache might have trimming curves of type NURBSCurveGWNCache instead of NURBSCurve, since that seems easier than templating the type of NURBSPatch on trimming curve type.
jcs15c
left a comment
jcs15c
left a comment
There was a problem hiding this comment.
Thanks for the optimizations of winding_number(NURBSPatch)! It all looks great.
kennyweiss
requested review from
Arlie-Capps,
BradWhitlock,
adayton1,
bmhan12,
publixsubfan,
rhornung67 and
white238
* Allows user to omit bounding box and will use input bounding box instead. * Checks for consistency in query grid bounds and resolution (min/max/res) * Adds a few timers and annotations
Scales query bounding box when user does not supply one.
…or u- and v- The function already clamps the parametric coordinates to the knot-span, so there's no real loss here. In this case, the function was called when computing the patch normal of the trimmed patch. We should probably preprocess the trimming curves to be withing the parametric bounding box since the user of the 3D winding numbers has no control over the trimming curves.
We can now choose not to provide a query_mesh bounding box, and it will instead use the input mesh's bounding box (slightly inflated). We also print the query times and rates.
I tracked down the changes in containment determination exclusively to samples on/near the boundary. atan2 is more stable, robust and performant than acos for the linear_winding_number calculations.
| * \return a*b+c computed using an fma operation | ||
| */ | ||
| template <typename T> | ||
| inline AXOM_HOST_DEVICE T fma(const T& a, const T& b, const T& c) |
There was a problem hiding this comment.
I can think of a bunch of places to use this!
kennyweiss
force-pushed
the
feature/kweiss/wn-example-3D
branch
from
8716c0a to
c63ef54
Compare
Summary
quest::STEPReader#1765 (all the changes in this PR toSTEPReader.cpp), so that should be merged firstNURBSPatch.hppto accommodate the fact that many in-the-wild CAD examples have trimming curves which extend beyond the range of the patch's parameter space. In the future, these could be handled by the methods described in Add slic macros that only log the first time they're called #1685, which would alert the user without cluttering output.linear_winding_numberis now based onatan2instead ofacos. This was the bottleneck in 2D WN computations and shaves about 25% of runtimesaxom::utilities::fmafunction (with the expectation that this will improve robustness)