|
| 1 | +# 25.11 |
| 2 | + |
| 3 | + ## Highlights: |
| 4 | + |
| 5 | + * AMReX is now compatible with oneAPI 2025.03 that has deprecated a number |
| 6 | + of things such as the `fsycl-device-lib` flag and the |
| 7 | + `mkl::sparse::set_csr_data` function. |
| 8 | + |
| 9 | + * It used to be an assertion error when the number of ghost cells in the |
| 10 | + MultiFab is different from that in the VisMF file. We now allow them to |
| 11 | + be different. Note that if there are more ghost cells in the MultiFab, |
| 12 | + the extra ghost cell regions are not filled. |
| 13 | + |
| 14 | + * In `amrex::UtilCreateCleanDirectory`, if a directory already exists, we |
| 15 | + preserve it by renaming it to a directory whose name includes a string |
| 16 | + generated by `amrex::UniqueString()`. However, this string is not |
| 17 | + guaranteed to be unique. The behavior of `std::rename` is |
| 18 | + implementation-defined, and it may fail if the destination already |
| 19 | + exists. This happens rarely, but it does happen. We now addresse the |
| 20 | + issue by removing the existing conflicting directory before |
| 21 | + renaming. This makes the file renaming process more robust. |
| 22 | + |
| 23 | + * We have reimplemented SumBoundary. In the old approach, we make a |
| 24 | + temporary copy of the data, zero out the original data, and then perform |
| 25 | + parallel add. In the new approach that should be more efficient, we no |
| 26 | + longer need a temporary copy of the entire data. We have also added an |
| 27 | + optional argument that can be used to make SumBoundary |
| 28 | + deterministic. Note that being deterministic does not mean the data are |
| 29 | + consistent on shared nodes. For that, one still needs to call |
| 30 | + OverrideSync, or FillBounaryAndSync. |
| 31 | + |
| 32 | + ## Other major changes: |
| 33 | + |
| 34 | + * Allow C++20 HIP builds (#4739) |
| 35 | + |
| 36 | + * Make transposeCtoF asynchronous (#4738) |
| 37 | + |
| 38 | + * FFT hybrid Poisson solver: Add support for batched 2d solves (#4732) |
| 39 | + |
| 40 | + * Cray MPICH/ROCm compatibility for GPU-aware MPI (#4729) |
| 41 | + |
| 42 | + * Remove final on Fapply, Fsmooth, FFlux, normalize, supportNSolve (#4727) |
| 43 | + |
| 44 | + * Align fcompare behavior with particle_compare when metadata differ (#4724) |
| 45 | + |
| 46 | + * Robustify file renaming (#4721) |
| 47 | + |
| 48 | + * Allow std::string "name" arguments for ParmParse methods (#4695) |
| 49 | + |
| 50 | + * Add Build with Explicit Multi-Level Geometry for EB (#4714) |
| 51 | + |
| 52 | + * Add reserve methods to ParticleTile, ArrayOfStructs, and StructOfArrays (#4708) |
| 53 | + |
| 54 | + * Fix a bug in EB FillPatch for face MultiFab (#4705) |
| 55 | + |
| 56 | + * Improve usability of amrex::Gpu::Buffer (#4697) |
| 57 | + |
| 58 | + * Reduce::MinMax: fix function name (#4704) |
| 59 | + |
| 60 | + * OpenBC FFT Solver: Fix 2d mode bug (#4703) |
| 61 | + |
| 62 | + * Hypre & PETSc in 1d build (#4692) |
| 63 | + Allow hypre and petsc in 1d build (#4682) |
| 64 | + |
1 | 65 | # 25.10 |
2 | 66 |
|
3 | 67 | ## Highlights: |
|
0 commit comments