Refactor occupy.h/cpp and update MPI kmesh_world

source/Makefile.Objects

@@ -231,6 +231,10 @@ OBJS_CELL=atom_pseudo.o\
     mdcell.o\
     cif_io.o\
     ucell_io.o\
+    read_cube.o\
+    write_cube.o\
+    write_pao.o\
+    output_log.o\
 
 OBJS_DEEPKS=lcao_deepks.o\
         deepks_basic.o\
@@ -626,7 +630,6 @@ OBJS_IO=module_parameter/input_conv.o\
     module_bessel/numerical_basis_output.o\
     output.o\
     module_output/print_info.o\
-    module_output/read_cube.o\
     module_wf/read_wfc_pw.o\
     module_wf/read_wf2rho_pw.o\
     module_restart/restart.o\
@@ -645,17 +648,14 @@ OBJS_IO=module_parameter/input_conv.o\
     module_wannier/to_w90_pw_setup.o\
     module_wannier/fr_overlap.o\
     module_unk/unk_overlap_pw.o\
-    module_output/write_pao.o\
     module_wf/write_wfc_pw.o\
-    module_output/write_cube.o\
     module_elf/write_elf.o\
     module_dipole/write_dipole.o\
     module_current/td_current_io.o\
     module_current/td_current_io_comm.o\
     td_efield_io.o\
     td_vector_pot_io.o\
     module_chgpot/write_libxc_r.o\
-    module_output/output_log.o\
     module_hs/output_mat_sparse.o\
     module_ctrl/ctrl_scf_lcao.o\
     module_ctrl/ctrl_runner_lcao.o\
@@ -692,7 +692,7 @@ OBJS_IO=module_parameter/input_conv.o\
     module_hs/cal_plpr.o\
 
 OBJS_IO_LCAO=module_hs/cal_r_overlap_r.o\
-      module_output/write_orb_info.o\
+      write_orb_info.o\
       module_dos/write_dos_lcao.o\
       module_energy/write_proj_band_lcao.o\
       module_energy/write_eig_occ.o\
@@ -797,7 +797,7 @@ OBJS_PARALLEL=parallel_common.o\
     para_pw_world.o\
     para_diag_world.o\
     para_rgrid_world.o\
-    para_bgroup_world.o\
+    para_bdiff_ksame_world.o\
     para_matrix_world.o\
     para_mpi_func.o\
     para_setup.o\

source/source_base/CMakeLists.txt

@@ -79,7 +79,7 @@ add_library(
     module_parallel/para_pw_world.cpp
     module_parallel/para_diag_world.cpp
     module_parallel/para_rgrid_world.cpp
-    module_parallel/para_bgroup_world.cpp
+    module_parallel/para_bdiff_ksame_world.cpp
     module_parallel/para_matrix_world.cpp
     module_parallel/para_mpi_func.cpp
     module_parallel/para_setup.cpp

source/source_base/module_parallel/para_bdiff_ksame_world.cpp

@@ -0,0 +1,33 @@
+#include "para_bdiff_ksame_world.h"
+
+namespace Parallel
+{
+
+ParaBdiffKsameWorld::ParaBdiffKsameWorld()
+    : ParaWorld("bdiff_ksame"), my_bndgroup_(0), nbndgroup_(1)
+{
+}
+
+#ifdef __MPI
+ParaBdiffKsameWorld::ParaBdiffKsameWorld(const MPI_Comm& intra_comm, const MPI_Comm& inter_comm, int nbndgroup)
+    : ParaWorld("bdiff_ksame", intra_comm), inter_comm_(inter_comm), nbndgroup_(nbndgroup)
+{
+    if (inter_comm != MPI_COMM_NULL)
+    {
+        MPI_Comm_rank(inter_comm, &my_bndgroup_);
+    }
+}
+#endif
+
+void ParaBdiffKsameWorld::reduce_across_bdiff_ksame(double& value) const
+{
+#ifdef __MPI
+    if (inter_comm_ == MPI_COMM_NULL || nbndgroup_ <= 1)
+    {
+        return;
+    }
+    MPI_Allreduce(MPI_IN_PLACE, &value, 1, MPI_DOUBLE, MPI_SUM, inter_comm_);
+#endif
+}
+
+} // namespace Parallel

source/source_base/module_parallel/para_bdiff_ksame_world.h

@@ -0,0 +1,87 @@
+#ifndef PARA_BDIFF_KSAME_WORLD_H
+#define PARA_BDIFF_KSAME_WORLD_H
+
+#include "para_world.h"
+
+namespace Parallel
+{
+
+/**
+ * @brief bdiff_ksame parallel domain: band-group communication topology
+ * inside one k-pool.
+ *
+ * Self-contained replacement for INT_BGROUP + BP_WORLD +
+ * GlobalV::MY_BNDGROUP/NPROC_IN_BNDGROUP/RANK_IN_BPGROUP.
+ *
+ * The domain has two communicators:
+ *   - intra: INT_BGROUP (bsame_kdiff; same band group, different k/pw)
+ *   - inter: BP_WORLD  (bdiff_ksame; different band groups, same k)
+ *
+ * Tests only need this header.
+ */
+class ParaBdiffKsameWorld : public ParaWorld
+{
+public:
+    /**
+     * @brief Construct a serial domain (single band group).
+     */
+    ParaBdiffKsameWorld();
+
+#ifdef __MPI
+    /**
+     * @brief Construct a domain from intra and inter communicators.
+     *
+     * @param[in] intra_comm  intra-group communicator (e.g. INT_BGROUP)
+     * @param[in] inter_comm  inter-group communicator (e.g. BP_WORLD)
+     * @param[in] nbndgroup   number of band groups
+     */
+    ParaBdiffKsameWorld(const MPI_Comm& intra_comm, const MPI_Comm& inter_comm, int nbndgroup);
+#endif
+
+    /// Band group index of this process.
+    int my_bndgroup() const { return my_bndgroup_; }
+
+    /// Number of band groups.
+    int nbndgroup() const { return nbndgroup_; }
+
+    /// Rank within the band group (alias for rank()).
+    int rank_in_bpgroup() const { return rank(); }
+
+    /// Number of processes in the band group (alias for size()).
+    int nproc_in_bndgroup() const { return size(); }
+
+#ifdef __MPI
+    /// Inter-group communicator (BP_WORLD / bdiff_ksame equivalent).
+    MPI_Comm inter_comm() const { return inter_comm_; }
+#endif
+
+    /**
+     * @brief Sum a scalar across the band groups of this k-pool.
+     *
+     * Band-parallel eigensolvers (bpcg) shard the band range across the
+     * BNDPAR band groups of a k-pool: every process only accumulates the
+     * partial sum over its own band window. This reduction combines those
+     * partial sums on the bdiff_ksame (BP_WORLD) communicator, which links
+     * the same rank position of every band group inside one k-pool, so each
+     * band window contributes exactly once.
+     *
+     * It must run BEFORE ParaKmeshWorld::reduce_across_pools so that the
+     * k-pool reduction receives one complete per-k-pool partial sum.
+     * No-op when there is only a single band group.
+     *
+     * @param[in,out] value  local partial sum, overwritten with the
+     *                       k-pool-wide total
+     */
+    void reduce_across_bdiff_ksame(double& value) const;
+
+private:
+    int my_bndgroup_ = 0;
+    int nbndgroup_ = 1;
+#ifdef __MPI
+    MPI_Comm inter_comm_ = MPI_COMM_NULL;
+#endif
+};
+
+} // namespace Parallel
+
+#endif // PARA_BDIFF_KSAME_WORLD_H

source/source_base/module_parallel/para_bridge.cpp

@@ -2,6 +2,7 @@
 #include "para_tag.h"
 
 #ifdef __MPI
+#include "source_base/global_variable.h"
 #include "source_base/parallel_comm.h"
 #endif
 
@@ -19,4 +20,63 @@ ParaWorld make_pw_world()
 #endif
 }
 
+// Reduce-only overload: no k-point distribution data.
+ParaKmeshWorld make_kmesh_world()
+{
+#ifdef __MPI
+    int mpi_initialized = 0;
+    MPI_Initialized(&mpi_initialized);
+    // Any distributed layout (k pools or band groups) may need the
+    // world-wide max/min reductions, so build the MPI domain whenever
+    // more than one process is running. The sum reduction no-ops for
+    // kpar <= 1 on its own.
+    if (mpi_initialized && GlobalV::NPROC > 1)
+    {
+        // Build from globals but skip distribute_kpoints (nkstot=0).
+        return ParaKmeshWorld(MPI_COMM_WORLD, GlobalV::KPAR, GlobalV::MY_POOL, 0, 1);
+    }
+#endif
+    return ParaKmeshWorld();
+}
+
+// Temporary bridge: construct a kmesh-domain ParaKmeshWorld from the old
+// globals. Delete this file once ParaCollection is wired into driver init.
+ParaKmeshWorld make_kmesh_world(int nkstot, int nspin)
+{
+#ifdef __MPI
+    // Fall back to a serial single-pool domain when MPI is not initialized
+    // (e.g. unit tests linked against the MPI-compiled base library) or when
+    // there is only one k-pool, so that no MPI call is made on an unset
+    // communicator.
+    int mpi_initialized = 0;
+    MPI_Initialized(&mpi_initialized);
+    if (mpi_initialized && GlobalV::KPAR > 1)
+    {
+        return ParaKmeshWorld(MPI_COMM_WORLD, GlobalV::KPAR, GlobalV::MY_POOL,
+                              nkstot, nspin);
+    }
+#endif
+    return ParaKmeshWorld(nkstot, nspin);
+}
+
+// Temporary bridge: construct a bdiff_ksame-domain ParaBdiffKsameWorld from
+// the old globals. Delete this file once ParaCollection is wired into driver init.
+ParaBdiffKsameWorld make_bdiff_ksame_world()
+{
+#ifdef __MPI
+    int mpi_initialized = 0;
+    MPI_Initialized(&mpi_initialized);
+    // NPROC_IN_BNDGROUP stays 0 until divide_pools has run, which also
+    // guards unit tests that link the MPI base library without a layout.
+    if (mpi_initialized && INT_BGROUP != MPI_COMM_NULL && BP_WORLD != MPI_COMM_NULL
+        && GlobalV::NPROC_IN_BNDGROUP > 1)
+    {
+        int nbndgroup = 1;
+        MPI_Comm_size(BP_WORLD, &nbndgroup);
+        return ParaBdiffKsameWorld(INT_BGROUP, BP_WORLD, nbndgroup);
+    }
+#endif
+    return ParaBdiffKsameWorld();
+}
+
 } // namespace Parallel