ND Spline Evaluator#967
Open
tretre91 wants to merge 31 commits into
Open
Conversation
Codecov Report❌ Patch coverage is
❌ Your patch check has failed because the patch coverage (83.92%) is below the target coverage (100.00%). You can increase the patch coverage or adjust the target coverage. 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
reviewed
Nov 10, 2025
Member
tpadioleau
left a comment
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
tretre91
force-pushed
the
nd_evaluator
branch
from
40adb73 to
fab1202
Compare
tretre91
force-pushed
the
nd_evaluator
branch
2 times, most recently
from
0b394b6 to
41df7f3
Compare
tretre91
force-pushed
the
nd_evaluator
branch
from
41df7f3 to
3d069d2
Compare
Member
Author
|
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. |
Member
Author
|
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 |
Member
I think we need rocm 6.3.3 to make the splines tests work. I don't know what is happening for |
Member
Author
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 |
Member
Author
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 |
Member
|
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)
Member
Author
|
The tests are passing on Jean-Zay h100 |
Member
Thanks, ready to review then ? |
Member
Author
Yes |
tpadioleau
requested changes
May 21, 2026
Member
tpadioleau
left a comment
tpadioleau
left a comment
There was a problem hiding this comment.
Great job, let me know if some comments are not clear or need discussion
| @@ -0,0 +1,159 @@ | |||
| // NOLINTBEGIN(readability-identifier-naming) | |||
| // SPDX-FileCopyrightText: 2026 CExA-project | |||
Member
There was a problem hiding this comment.
I don't see the copyrights of the initial implementation ?
Also I don't think we need the date, we don't put it in the rest of the repo and it looks confusing to me.
| @@ -0,0 +1,159 @@ | |||
| // NOLINTBEGIN(readability-identifier-naming) | |||
| // SPDX-FileCopyrightText: 2026 CExA-project | |||
| // SPDX-License-Identifier: MIT or Apache-2.0 with LLVM-exception | |||
Member
There was a problem hiding this comment.
Was it originally multi-licensed ?
Comment on lines
+20
to
+23
| template <class> | ||
| constexpr bool is_reference_wrapper_v = false; | ||
| template <class U> | ||
| constexpr bool is_reference_wrapper_v<std::reference_wrapper<U>> = true; |
Member
There was a problem hiding this comment.
Suggested change
| template <class> | |
| constexpr bool is_reference_wrapper_v = false; | |
| template <class U> | |
| constexpr bool is_reference_wrapper_v<std::reference_wrapper<U>> = true; |
Can we really support std::reference_wrapper in a device compatible tuple ?
| #include <array> | ||
| #include <cstddef> | ||
| #include <ranges> | ||
| #include <type_traits> // integral_constant |
Member
There was a problem hiding this comment.
Suggested change
| #include <type_traits> // integral_constant | |
| #include <type_traits> |
|
|
||
| if constexpr (std::is_function_v<Pointed>) { | ||
| if constexpr (is_derived_object) { | ||
| return (std::forward<Object>(object).*member)(std::forward<Args>(args)...); |
Member
There was a problem hiding this comment.
For DDC usage of std::forward is ok on device code but in Kokkos I guess it is not ?
| auto const spline_coef_ND = spline_coef[j]; | ||
| ddc::device_for_each( | ||
| evaluation_domain(spline_eval.domain()), | ||
| [=, *this](evaluation_domain::discrete_element_type const i) { |
Member
There was a problem hiding this comment.
Suggested change
| [=, *this](evaluation_domain::discrete_element_type const i) { | |
| [&](evaluation_domain::discrete_element_type const i) { |
| auto const spline_coef_ND = spline_coef[j]; | ||
| ddc::device_for_each( | ||
| evaluation_domain(spline_eval.domain()), | ||
| [=, *this](evaluation_domain::discrete_element_type const i) { |
Member
There was a problem hiding this comment.
Suggested change
| [=, *this](evaluation_domain::discrete_element_type const i) { | |
| [&](evaluation_domain::discrete_element_type const i) { |
| integrals(j) = 0; | ||
| ddc::device_for_each( | ||
| ddc::DiscreteDomain<BSplines...>(), | ||
| [=](ddc::DiscreteDomain<BSplines...>::discrete_element_type const i) { |
Member
There was a problem hiding this comment.
Suggested change
| [=](ddc::DiscreteDomain<BSplines...>::discrete_element_type const i) { | |
| [&](ddc::DiscreteDomain<BSplines...>::discrete_element_type const i) { |
Member
There was a problem hiding this comment.
I see multiple extended lambdas inside member functions of a class with multiple variadic template parameters, I hope this will work fine for nvcc.
| KOKKOS_LAMBDA(batch_domain_type<BatchedDDom>::discrete_element_type const j) { | ||
| integrals(j) = 0; | ||
| ddc::device_for_each( | ||
| ddc::DiscreteDomain<BSplines...>(), |
Member
There was a problem hiding this comment.
Suggested change
| ddc::DiscreteDomain<BSplines...>(), | |
| ddc::DiscreteDomain<BSplines...>(spline_coef.domain()), |
I think it would even be better to construct this object outside of the parallel_for_each to avoid each thread paying for this ?
Member
|
I think it also deserves a new contribution entry in the AUTHORS |
Member
|
May need to rebase on main to make the CI work |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
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