PaulMullowney
PaulMullowney
Not sure how to resolve this. I add +rocm as my spack variant. CMake Error in CMakeLists.txt: Target "blt_hip" INTERFACE_INCLUDE_DIRECTORIES property contains path: "/gpfs/alpine/cfd116/proj-shared/mullowne/spack-manager-hma/stage/spack-stage-umpire-2022.03.1-tqmh5ayb6ho36rsixipec3swpyo6p7y6/spack-src/HIP_CLANG_INCLUDE_PATH-NOTFOUND/.." which is prefixed in the source...
Hi, I have the following cmake code for finding umpire: set(CMAKE_PREFIX_PATH ${UMPIRE_ROOT_DIR} ${CMAKE_PREFIX_PATH}) message(STATUS "Looking for Umpire at ${UMPIRE_ROOT_DIR}") find_package(umpire REQUIRED CONFIG) message(STATUS "Found Umpire = ${UMPIRE_DIR}") I supply UMPIRE_ROOT_DIR...
…d multifabs. ## Summary ## Additional background ## Checklist The proposed changes: - [ ] fix a bug or incorrect behavior in AMReX - [x] add new capabilities to AMReX...
I had to write wrapper implementations that execute the loops over the box, when compiling for CPU. Also, the compactify method does not compile for GPU. So, the 3D impl...
Why is hypre_ForceSyncComputeStream necessary when using GPU-aware MPI? Alternatively, why isn't hypre_SyncComputeStream sufficient? -Paul
Implementation of GMRES, COGMRES GramSchmidt using MassDotpTwo, MassInnerProduct and MassAxpy. This includes the an option to choose classical (CGS) or modified Gram-Schmidt (MGS) in GMRES. There is also a Thrust...
Jacobi iterative solver for ILU smoother/preconditioner implemented on CPU/GPU currently for Nvidia only since we don't have AMD ILU factorization yet. This will occur in the next PR. This works...
slight fix to compile with umpire and pinned memory pools
I've been able to get gpu-aware-mpi working through spack builds. Umpire seems a little trickier. Any help on this would be appreciated. ```` diff --git a/var/spack/repos/builtin/packages/hypre/package.py b/var/spack/repos/builtin/packages/hypre/package.py index b38ae55ab0..339147acfb 100644...
Hi, I've been running running GMRES-BoomerAMG computations through the cuda-memcheck race check tool. I've tried 5 different compiler/cuda combos on 2 different tests (6 GPUs on a small test, 18...