ND Spline Evaluator#967
ND Spline Evaluator#967tretre91 wants to merge 31 commits intoCExA-project:mainCExA-project/ddc:mainfrom
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #967 +/- ##
==========================================
- Coverage 95.16% 94.75% -0.41%
==========================================
Files 69 72 +3
Lines 2958 3072 +114
Branches 944 976 +32
==========================================
+ Hits 2815 2911 +96
- Misses 91 104 +13
- Partials 52 57 +5 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
tpadioleau
left a comment
There was a problem hiding this comment.
You also need to declare the new header file in the target_sources to make the installation test pass
40adb73 to
fab1202
Compare
0b394b6 to
41df7f3
Compare
41df7f3 to
3d069d2
Compare
|
The compilation passes, but I got some test failures on cuda that I'm still investigating (the first one is unrelated to this PR), I haven't tested on hip yet. 147 - DeviceForEachSerialDevice.OneDimension (Failed)
776 - BatchedNd1dSplineDeviceGrevilleUniformDegree2.2DXB1 (Failed)
777 - BatchedNd1dSplineDeviceGrevilleUniformDegree2.2DB1X (Failed)
800 - BatchedNd1dSplineDeviceGrevilleUniformDegree5.2DXB1 (Failed)
801 - BatchedNd1dSplineDeviceGrevilleUniformDegree5.2DB1X (Failed)
840 - BatchedNd1dSplineDeviceHermiteUniformDegree5.2DXB1 (Failed)
841 - BatchedNd1dSplineDeviceHermiteUniformDegree5.2DB1X (Failed)
923 - BatchedNd3dSplineDeviceGrevilleUniformDegree2.3DXYZ (Failed)
959 - BatchedNd3dSplineDeviceGrevilleUniformDegree5.3DXYZ (Failed)
1019 - BatchedNd3dSplineDeviceHermiteUniformDegree5.3DXYZ (Failed)Full outputI'm also working on the tuple implementation on a separate repo, I cleaned things up a bit and started working on the clang tidy warnings. |
|
The errors don't actually come from my changes it seems, on adastra mi250x in Release, I have failures on both the Nd and 1d/3d tests (kokkos/kokkos kernels 5.0.2, ginkgo 1.11.0, amdclang++ 6.2.1) Full output |
I think we need rocm 6.3.3 to make the splines tests work. I don't know what is happening for |
I still had failures with 6.4.3, I think I'll put this on hold and investigate after I return from hpsf community summit |
It looks like the errors were related to my installation of the dependencies, I retried with everything installed with spack, but I still got some errors on the ND 3d that I'm investigating Full output |
|
If this can help, you can assume C++20 from now on. |
- works for host backends for the moment
- test the 1d ND evaluator - rename the 3d test file
- Make the DerivDims template argument non-variadic - Add tests for one of the deriv_dim_I overloads
- This will need to be cleaned, some headers (ex: apply.hpp) are not used
- update the tuple to c++20 - fix clang-tidy warnings - do not use CTAD (as nvcc fails with single argument ctors)
|
The tests are passing on Jean-Zay h100 |
This PR adds a SplineEvaluatorND class working in any number of dimensions. Currently, the template arguments to the class and to the methods are passed using
TypeSeqs, and the aliasesalias_1, alias_2, etcare replaced by templated equivalentsalias<0>, alias<1>, etc.The tests for this are just the 3D spline test and one of the 1D spline test modified to use the ND evaluator. We might want to factorize some of the test code or use the same test files with some
#ifdefs to enable the ND or regular Evalutor