feat: optimize MPI communication with non-blocking operations in eigenvalue solvers by laoba657 · Pull Request #7401 · deepmodeling/abacus-develop

feat: optimize MPI communication with non-blocking operations in eigenvalue solvers

laoba657

committed

fix: use type-dispatching MPI helpers for double and complex support

laoba657

committed

fix: remove mixed-precision code from MPI-only branch

laoba657

committed

fix: remove unused wait_some() to resolve std::remove ambiguity with C stdio

laoba657

committed

fix: add extern zheev_ declaration and using namespace hsolver in mpi test

laoba657

committed

fix: add diag_hs_para.cpp to MODULE_HSOLVER_mpi test target

laoba657

committed

fix: also add diago_pxxxgvx.cpp to MODULE_HSOLVER_mpi test for diag_hs_para linking

laoba657

committed

fix: remove unused diago_dav_subspace dependency from mpi test

laoba657

committed

fix: revert para_linear_transform.cpp to develop - non-blocking MPI_Irecv incompatible with GPU device memory

laoba657

committed

fix: restore para_linear_transform.cpp from correct develop commit (71f35241a)

laoba657

committed

fix: skip MPI test when nproc < 2 to prevent hang in single-process CI

laoba657

committed

fix: build MPI test without ctest registration, only run via mpirun script

laoba657

committed

fix: remove MPI test from ctest completely to prevent hang

laoba657

committed

fix: replace non-blocking MPI with blocking to prevent hang

laoba657

committed

fix: wrap reduce_pool/bcast in __MPI guard, add no-op fallbacks

laoba657

committed

fix: move mpi_type traits inside __MPI guard to fix non-MPI build

laoba657

committed

fix: move mpi_type inside __MPI guard to fix non-MPI build

laoba657

committed

Revert to non-blocking MPI: skip test when nproc < 2

laoba657

committed

fix: detect mpirun env before MPI_Init to prevent hang

laoba657

committed

fix: add mpi_type<float> to prevent MPI_BYTE fallback for float tests

laoba657

committed

fix: use MPI_COMM_WORLD instead of POOL_WORLD in mpi test

laoba657

committed

simplify mpi test: remove DiagoDavid-dependent tests, keep only direct MPI communication tests

laoba657

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: optimize MPI communication with non-blocking operations in eigenvalue solvers#7401

feat: optimize MPI communication with non-blocking operations in eigenvalue solvers#7401
laoba657 wants to merge 22 commits into
deepmodeling:developfrom
laoba657:feature/mpi-optimization

Commits on May 30, 2026

feat: optimize MPI communication with non-blocking operations in eigenvalue solvers

fix: use type-dispatching MPI helpers for double and complex support

fix: remove mixed-precision code from MPI-only branch

fix: remove unused wait_some() to resolve std::remove ambiguity with C stdio

fix: add extern zheev_ declaration and using namespace hsolver in mpi test

fix: add diag_hs_para.cpp to MODULE_HSOLVER_mpi test target

fix: also add diago_pxxxgvx.cpp to MODULE_HSOLVER_mpi test for diag_hs_para linking

fix: remove unused diago_dav_subspace dependency from mpi test

fix: revert para_linear_transform.cpp to develop - non-blocking MPI_Irecv incompatible with GPU device memory

fix: restore para_linear_transform.cpp from correct develop commit (71f35241a)

fix: skip MPI test when nproc < 2 to prevent hang in single-process CI

fix: build MPI test without ctest registration, only run via mpirun script

fix: remove MPI test from ctest completely to prevent hang

fix: replace non-blocking MPI with blocking to prevent hang

fix: wrap reduce_pool/bcast in __MPI guard, add no-op fallbacks

fix: move mpi_type traits inside __MPI guard to fix non-MPI build

fix: move mpi_type inside __MPI guard to fix non-MPI build

Revert to non-blocking MPI: skip test when nproc < 2

fix: detect mpirun env before MPI_Init to prevent hang

fix: add mpi_type<float> to prevent MPI_BYTE fallback for float tests

fix: use MPI_COMM_WORLD instead of POOL_WORLD in mpi test

simplify mpi test: remove DiagoDavid-dependent tests, keep only direct MPI communication tests