|
| 1 | +# 25.07 |
| 2 | + |
| 3 | + ## Highlights: |
| 4 | + |
| 5 | + * Fix GPU communication for non-arithmetic types (#4515) |
| 6 | + |
| 7 | + Previously communication of non-arithmetic types for GPU builds did not |
| 8 | + work if there were race conditions. This has now been fixed. |
| 9 | + |
| 10 | + * Parallel Copy with Offset (#4510) |
| 11 | + |
| 12 | + Add new versions of ParallelCopy and ParallelAdd that effectively shift |
| 13 | + the source multifab first and then perform copy/add. Previously, one |
| 14 | + would need to use NonLocalBC::PareallelCopy, which has a relatively more |
| 15 | + complicated interface. |
| 16 | + |
| 17 | + * Add STLtools::fillSignedDistance (#4501) |
| 18 | + |
| 19 | + This adds a new approach for computing signed distances for STL files. |
| 20 | + |
| 21 | + ## Other major changes: |
| 22 | + |
| 23 | + * AMRErrorTag: Add PARSER (#4534) |
| 24 | + |
| 25 | + * Support Dirichlet BC in z-direction in hybrid FFT Poisson solver (#4503) |
| 26 | + |
| 27 | + * Add getInvalidRandomEngine() for CPU (#4528) |
| 28 | + |
| 29 | + * Spack Smoke Test: Fix HIP and Add SYCL (#4521) |
| 30 | + |
| 31 | + * Modernize roctx Annotations (#4523) |
| 32 | + |
| 33 | + * Add `AMReX_Order.H` (#4517) |
| 34 | + |
| 35 | + * EB: Move EB without regenerating geometry (#4511) |
| 36 | + |
| 37 | + * hypre with hip GNUmake improvements (#4514) |
| 38 | + |
| 39 | + * StateData: add new set functions (#4512) |
| 40 | + |
| 41 | + * Adding a new overload for MultiFab::norm2 (int comp, int numcomp) (#4509) |
| 42 | + |
| 43 | + * Delay gpu stream sync in MFIter (#4493) |
| 44 | + |
| 45 | + * Add paramters for controlling CArena defragmentation (#4479) |
| 46 | + |
| 47 | + * CMake: Need to set CUDA ARCH for plotfile tools (#4485) |
| 48 | + |
| 49 | + * Replaced depreciated cub::Max, cub::Min cub::Sum with cuda::maximum<>… (#4500) |
| 50 | + |
| 51 | + * Added option to use either always use BNorm or ResNorm for calculatin… (#4495) |
| 52 | + |
| 53 | + * Add support for force-inlined methods in GDB (#4492) |
| 54 | + |
| 55 | + * Always FillSignedDistance with max grid size of 32 (#4489) |
| 56 | + |
| 57 | + * Workarounds for Lustre I/O issues (#4426) |
| 58 | + |
| 59 | + * Communication Optimization: Cache GPU kernel metadata (#4462) |
| 60 | + |
| 61 | + * Fix constexpr evaulation bug for is_box_tag() in CPU version of Paral… (#4487) |
| 62 | + |
| 63 | + * Adding new assert check for MLMG to allow for staggered and non-cell … (#4488) |
| 64 | + |
| 65 | + * Add an --output_dir option to makebuildinfo_C.py (#4490) |
| 66 | + |
1 | 67 | # 25.06
|
2 | 68 |
|
3 | 69 | ## Highlights:
|
|
20 | 86 | using a new version of amrex::Initialize that does not need MPI_Comm
|
21 | 87 | introduced in this release can also fix the issue.
|
22 | 88 |
|
23 |
| - ## Major changes: |
| 89 | + ## Other major changes: |
24 | 90 |
|
25 | 91 | * ParallelCopy: Fix a launch region issue (#4481)
|
26 | 92 |
|
|
0 commit comments