camierjs
351806ce77
[MemorySanitizer] use-of-uninitialized-value InvTNewtonSolverBase
2025-06-19 10:59:27 -07:00
Andrew Ho
537112ebb8
cleaned up kernel specialization registration
...
Added assert for unsupported NewtonSegmentProject
2025-03-28 16:03:33 -07:00
Andrew Ho
2ce31f6841
one more old code comment
2025-03-28 13:01:10 -07:00
Andrew Ho
6859612af9
cleanup old code
2025-03-28 12:58:47 -07:00
Andrew Ho
d9d963cc88
Address review comments
...
- changed use_dev to use_device
- added a way to directly set initial guess order rather than have
relative order only
- cleanup newton solves
- updated some documentation
2025-03-28 12:47:49 -07:00
Andrew Ho
e386c1a655
Default to closed uniform points for edgescan if btype is invalid
...
Update documentation
2025-03-17 13:33:07 -07:00
Andrew Ho
2ced1ceb84
Re-arranged interface to UpdateNodes from a GridFunction
...
Added tests for 2D embedded in 3D and 3D mesh
2025-03-12 17:42:45 -07:00
Andrew Ho
cd799dba86
removed ::mfem in batchitrans
...
isn't needed here
2025-03-12 11:32:21 -07:00
Andrew Ho
00008800b8
fixed typo
2025-03-12 11:11:29 -07:00
Andrew Ho
52c1a3c0b8
fixed element dof indexing
2025-02-24 12:16:38 -08:00
Andrew Ho
83f553f4a9
fixed non-power of 2 team sizes
...
these fail during the tree reduction steps
2025-02-24 02:14:44 -08:00
Andrew Ho
bcde2ec4f1
fixed step counting in findpts for some cases
2025-02-21 06:51:50 -08:00
Andrew Ho
cdbc8c710c
implemented iteration count tracking in findpts and batch element transformations
2025-02-20 18:57:47 -08:00
Andrew Ho
c75c5f2ea8
added optimization for guesses co-located at dofs
...
this matches findpts, seems like a worthwhile performance upgrade in
most situations over center
2025-02-20 05:15:46 -08:00
Andrew Ho
34ea1d5b07
fixed batch edgescan
...
Adding a native nodal DOFs init guess type would work well for some elements
2025-02-19 20:58:58 -08:00
Andrew Ho
d1f560d652
implemented batch edgescan
...
Could have a smarter traversal order
2025-02-19 17:29:40 -08:00
Andrew Ho
8048e1d433
return the element result from InvTNewtonSolver
...
this allows an EdgeScan to re-use the same solver code
2025-02-19 15:10:22 -08:00
Andrew Ho
c322d7cbb0
using max_q1d speeds up physnode initial guess finder
2025-02-19 01:59:17 -08:00
Andrew Ho
baa2c246f9
changed newtonsolve parallelization strategy
...
uses a lot less shared memory, hopefully performs better. Need to tune
max_team_x.
2025-02-18 17:57:47 -08:00
Andrew Ho
337dd7c274
changed loop order
...
try to improve memory access pattern
2025-02-15 08:26:58 -08:00
Andrew Ho
6f3ecfe129
fixed 3d cube linsolve
2025-02-14 20:27:40 -08:00
Andrew Ho
ccdaa48f69
batch inverse element transformation should now work for all square and cube element geometries
...
assumes that evaluating at GLL nodes is equivalent, which I think for
NURBS meshes is not true
2025-02-14 12:32:09 -08:00
Andrew Ho
490b60c272
re-arrangement to make doxygen happy
2025-02-14 00:01:53 -08:00
Andrew Ho
9aaa7f13a3
fixed -Wall -Werror for some config options
2025-02-13 16:46:51 -08:00
Andrew Ho
07605d3097
fixed compile bugs wtih various configurations
2025-02-13 14:58:55 -08:00
Andrew Ho
02aed483e6
fixed newtonsolver
2025-02-13 09:20:16 -08:00
Andrew Ho
facb0b5434
implemented newtonproject solves
2025-02-12 22:25:00 -08:00
Andrew Ho
cb32b5ee8e
implemented square element newton project solve
2025-02-12 21:00:46 -08:00
Andrew Ho
cd50b6b806
setup infrastructure for calling newtonsolve
...
still need to implement remaining specializations
2025-02-12 13:40:44 -08:00
Andrew Ho
74ef6490a7
implemented elementproject for segments
2025-02-12 13:09:14 -08:00
Andrew Ho
f963df773c
implemented raw newton solve for segments
2025-02-12 12:52:33 -08:00
Andrew Ho
31786ff0e0
enabled specifying 1D forall thread block size
...
left as an optional param
2025-02-12 11:15:41 -08:00
Andrew Ho
3c17ed88c5
fixed some bugs
...
switched qaud1d type with basis type
fixed computing coordinate interpolation not summing basis values
2025-02-11 11:56:25 -08:00
Andrew Ho
0a754a78b9
implemented tensor basis optimization
2025-02-11 11:11:34 -08:00
Andrew Ho
35f508ca8c
fixed bugs in batch closestphysnode kernel
...
still could perform better...
2025-02-11 00:31:06 -08:00
Andrew Ho
acea00376b
fixed serial reduction save index
...
should be the thread x id, not just i
2025-02-10 12:36:27 -08:00
Andrew Ho
13fed2e691
implemented PhysNodeFinder for tensor elements
2025-02-10 12:25:19 -08:00
Andrew Ho
1332ba5ee7
stride_sdim needs all element dofs, not just ndof1d
2025-02-10 12:17:14 -08:00
Andrew Ho
a252286472
implemented ClosestPhysNodeImpl for segments in SDim
2025-02-10 12:10:31 -08:00
Andrew Ho
8bab223de4
switched init condition kernels to template on geometry instead of dim
2025-02-10 11:52:33 -08:00
Andrew Ho
96392055d6
think I want separate kernels per geometry shape
2025-02-10 10:13:53 -08:00
Andrew Ho
36f1b769c5
specialize on device vs. non-device
2025-02-09 08:36:30 -08:00
Andrew Ho
3e074b3c7e
implemented refinement points for batch itransform
2025-02-09 08:31:24 -08:00
Andrew Ho
d8773cad73
initialize batch inverse transformation kernels in setup
...
this is guaranteed to be called before the kernels are used
2025-02-09 07:44:14 -08:00
Andrew Ho
b1f7cb2766
moved batch inverse transform implementation into separate cpp file
2025-02-09 07:34:37 -08:00
camierjs