celeritas-project
Add support for surface-based VecGeom
VecGeom is implementing a new surface-based navigation mechanism designed to be more efficient on GPU: see 10.2172/2204991 and associated AdePT integration https://github.com/apt-sim/AdePT/issues/298 . This issue will track the progress of that integration.
@mrguilima please document the issues you're seeing as part of this update (upload log files of tests, etc.) Thanks!
The first three failing tests need support for Orb (FourLevels*) and many other shapes (solids.gdml). 54 - geocel/vg/Vecgeom:FourLevelsGeantTest.* (SEGFAULT) <<== needs Orb 55 - geocel/vg/Vecgeom:SolidsGeantTest.* (SEGFAULT) <<== solids.gdml, needs support for ~all shapes 157 - celeritas/ext/GeantVolumeMapper (Failed) <<== also uses solids.gdml, needs support for ~all shapes
Thanks! Did you see Severin's comment about recent changes negating the need for the stack size? How recent is the build of vecgeom you're using?
Yes, as I said yesterday, the stack size issue appears only in Debug mode, not in Release mode.
For the missing "orb" we should be able to trivially build a sphere instead, and for the unsupported volumes (do you have a list?) we can likewise modify the geometry converter.
I can apply this change.
The next group of tests are failing because they load more than one geometry, but the reset_geometry() function is not working properly for the Surface infrastructure. Using the --gtest_filter=Cmse* etc.. works for most of the geometries though (exception: Cmse), so I assume that the BVHNavigator from VecGeom is working fine at this point.
162 - celeritas/field/FieldPropagator (SEGFAULT) 165 - celeritas/geo/Geometry (SEGFAULT)
All these work well: test/celeritas/field_FieldPropagator --gtest_filter=TwoBox* test/celeritas/field_FieldPropagator --gtest_filter=LayersTest* test/celeritas/field_FieldPropagator --gtest_filter=SimpleCmsTest* test/celeritas/field_FieldPropagator --gtest_filter=CmseTest*
test/celeritas/geo_Geometry --gtest_filter=SimpleCmsTest*
test/celeritas/geo_Geometry --gtest_filter=ThreeSpheres*
and this one fails: test/celeritas/geo_Geometry --gtest_filter=Cmse*
Cmse geometry needs some attention.
Can you run ctest with --output-on-failure and print the within-test failures?
This is the typical test output when a second geometry is loaded:
Start 53: geocel/vg/Vecgeom:SimpleCmsTest.*
1/4 Test #53: geocel/vg/Vecgeom:SimpleCmsTest.* ......... Passed 0.94 sec
Start 54: geocel/vg/Vecgeom:FourLevelsGeantTest.*
2/4 Test #54: geocel/vg/Vecgeom:FourLevelsGeantTest.* ...***Exception: SegFault 0.46 sec
Celeritas version 0.5.0-dev.225+0f952bc0
Note: Google Test filter = FourLevelsGeantTest.*
[==========] Running 2 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 2 tests from FourLevelsGeantTest
[ RUN ] FourLevelsGeantTest.accessors
/global/homes/l/lima/work/local/src/vecgeom/repo/VecGeom/surfaces/conv/SolidConverter.h:95: error: CreateSolidSurfaces: solid type not supported UnplacedOrb {5}
/global/homes/l/lima/work/local/src/vecgeom/repo/VecGeom/surfaces/cpp/BrepHelper.cpp:1290: critical: Could not convert volume 0: Shape20x14a1d50
Table size index: 3032
unknown file: Failure
C++ exception with description "celeritas: runtime error: failed to convert VecGeom to surfaces
/global/homes/l/lima/work/cele/src/geocel/vg/VecgeomParams.cc:375: 'brep_helper.Convert()' failed" thrown in the test body.
[ FAILED ] FourLevelsGeantTest.accessors (39 ms)
And here is the output for Geometry:CmseTest*
==>>> test/celeritas/geo_Geometry --gtest_filter=CmseTest*
Celeritas version 0.5.0-dev.225+0f952bc0
Note: Google Test filter = CmseTest*
[==========] Running 2 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 2 tests from CmseTest
[ RUN ] CmseTest.host
status: Loading VecGeom geometry from GDML at /global/homes/l/lima/work/cele/test/geocel/data/cmse.gdml
Table size index: 4080
status: Initializing tracking information
/global/homes/l/lima/work/cele/test/celeritas/geo/HeuristicGeoTestBase.cc:74: Failure
Values in: avg_path
Expected: ref_path
14 of 18 elements differ
by 0.0050000000000000001 relative error or 5.0000000000000002e-05 absolute error
i ref_path avg_path Difference
0 74.17136 74.6817891125235 0.00688175479758624
1 13.25306 13.4613848262866 0.0157189982001622
2 76.67924 67.7890370814263 -0.115940154317827
3 449.5464 460.345985002743 0.0240232932634832
4 0.09551618 0.0752032526740877 -0.212664779160057
5 0.3231404 0.395826227083629 0.224935746454573
6 0.310899 0.258379633374872 -0.16892742216967
7 0.3844357 0.51484801200587 0.339230492916942
9 11.09485 10.6629583649317 -0.0389272171384311
10 9.101073 9.3044714865428 0.0223488468384776
12 0.3033329 0.258743528862538 -0.146998136824137
14 228.7892 226.528020448794 -0.0098832442755416
16 563.0746 550.756536464024 -0.0218764326005393
17 2858.592 2824.10663159091 -0.0120637602040073
[ FAILED ] CmseTest.host (210 ms)
[ RUN ] CmseTest.device
/global/homes/l/lima/work/cele/test/celeritas/geo/HeuristicGeoTestBase.cc:98: Failure
Values in: avg_path
Expected: this->reference_avg_path()
14 of 18 elements differ
by 0.0050000000000000001 relative error or 5.0000000000000002e-05 absolute error
i EXPECTED avg_path Difference
0 74.17136 74.6817891125236 0.00688175479758797
1 13.25306 13.4613848262866 0.0157189982001608
2 76.67924 67.7890370814264 -0.115940154317826
3 449.5464 460.345985002744 0.0240232932634845
4 0.09551618 0.0752032526740867 -0.212664779160068
5 0.3231404 0.395826227083628 0.22493574645457
6 0.310899 0.258379633374872 -0.16892742216967
7 0.3844357 0.51484801200587 0.339230492916943
9 11.09485 10.6629583649317 -0.0389272171384296
10 9.101073 9.30447148654281 0.0223488468384788
12 0.3033329 0.258743528869017 -0.146998136802776
14 228.7892 226.528020448788 -0.00988324427556943
16 563.0746 550.756536464034 -0.0218764326005231
17 2858.592 2824.10663159092 -0.0120637602040014
[ FAILED ] CmseTest.device (99 ms)
[----------] 2 tests from CmseTest (310 ms total)
Awesome, thanks @mrguilima . If you open a draft pull request, I can push patches to fix the "unplaced orb" issue and we can work from there?
@mrguilima Now that 2.0.0-rc.4 is out, and celeritas is patched, could you take a look at this again? Have you merged develop recently and can you make sure this is up to date? I've got the spack update ready to go.