Merge branch 'release-2.1.0-rc5'

VERSION

@@ -1,6 +1,6 @@
 MAJOR = 2
 MINOR = 1
-PATCH = 0-rc4
+PATCH = 0-rc5
 # A specific DATE (YYYY-MM-DD) fixes an official release, otherwise
 # it is considered Development version.
 DATE  =

docs/guide/2-user-guide/1-installation/1-install.md

@@ -14,7 +14,7 @@ You absolutely need:
 
 Optionally:
 
-* [libxsmm](https://github.com/hfp/libxsmm) (1.10+) for Small Matrix Multiplication acceleration
+* [libxsmm](https://github.com/hfp/libxsmm) (1.10+, and `pkg-config`) for Small Matrix Multiplication acceleration
 * a LAPACK implementation (reference, OpenBLAS-bundled and MKL have been tested), required when building the tests
 
 To build `libsmm_acc`, DBCSR's GPU backend, you further need:
@@ -64,6 +64,13 @@ Run inside the `dbcsr` directory:
     -DTEST_OMP_THREADS=<2,N>
 ```
 
+When providing a custom build of `libxsmm`, make sure that its library directory is added to the `PKG_CONFIG_PATH` variable prior
+to running `cmake`. An example if `libxsmm` was checked out using Git to your home folder:
+
+```bash
+export PKG_CONFIG_PATH="${PKG_CONFIG_PATH}:${HOME}/libxsmm/lib"
+```
+
 ### CMake Build Recipes
 
 For build recipes on different platforms, make sure to also read the [CMake Build Recipes](./2-cmake-build-recipes.html).

src/acc/acc_init.cpp

@@ -23,10 +23,12 @@
 
 /****************************************************************************/
 extern "C" int acc_init(){
+  int myDevice;
   // Driver boilerplate
   ACC_DRV_CALL(Init, (0));
   ACC_DRV(device) acc_device;
-  ACC_DRV_CALL(DeviceGet, (&acc_device, 0));
+  ACC_API_CALL(GetDevice, (&myDevice));
+  ACC_DRV_CALL(DeviceGet, (&acc_device, myDevice));
   ACC_DRV(context) ctx;
   ACC_DRV_CALL(DevicePrimaryCtxRetain, (&ctx, acc_device));
 
@@ -37,10 +39,11 @@ extern "C" int acc_init(){
 
 /****************************************************************************/
 extern "C" int acc_finalize(){
+  int myDevice;
   // Release driver resources
   ACC_DRV(device) acc_device;
-  ACC_DRV_CALL(DeviceGet, (&acc_device, 0));
+  ACC_API_CALL(GetDevice, (&myDevice));
+  ACC_DRV_CALL(DeviceGet, (&acc_device, myDevice));
   ACC_DRV_CALL(DevicePrimaryCtxRelease, (acc_device));
   return 0;
 }
-

src/tensors/dbcsr_tensor_api.F

@@ -19,10 +19,10 @@ MODULE dbcsr_tensor_api
       dbcsr_t_contract, dbcsr_t_get_block, dbcsr_t_get_stored_coordinates, dbcsr_t_put_block, &
       dbcsr_t_reserve_blocks, dbcsr_t_copy_matrix_to_tensor, dbcsr_t_copy, &
       dbcsr_t_copy_tensor_to_matrix, dbcsr_t_need_contract, dbcsr_t_batched_contract_init, &
-      dbcsr_t_batched_contract_finalize, dbcsr_t_ndims
+      dbcsr_t_batched_contract_finalize, dbcsr_t_ndims, dbcsr_t_contract_index
    USE dbcsr_tensor_block, ONLY: &
       dbcsr_t_iterator_blocks_left, dbcsr_t_iterator_next_block, dbcsr_t_iterator_start, &
-      dbcsr_t_iterator_stop, dbcsr_t_iterator_type
+      dbcsr_t_iterator_stop, dbcsr_t_iterator_type, dbcsr_t_reserved_block_indices
    USE dbcsr_tensor_types, ONLY: &
       dbcsr_t_create, dbcsr_t_destroy, dbcsr_t_distribution_destroy, dbcsr_t_distribution_new, &
       dbcsr_t_distribution_type, dbcsr_t_nd_mp_comm, dbcsr_t_nd_mp_free, dbcsr_t_type, &
@@ -31,7 +31,7 @@ MODULE dbcsr_tensor_api
       dbcsr_t_mp_environ_pgrid => mp_environ_pgrid, dbcsr_t_blk_sizes, dbcsr_t_get_info, &
       dbcsr_t_finalize, dbcsr_t_scale, dbcsr_t_get_nze, dbcsr_t_get_nze_total, &
       dbcsr_t_get_num_blocks, dbcsr_t_get_num_blocks_total, dbcsr_t_clear, &
-      dbcsr_t_mp_dims_create
+      dbcsr_t_mp_dims_create, dbcsr_t_pgrid_change_dims
    USE dbcsr_tensor_test, ONLY: &
       dbcsr_t_contract_test, dbcsr_t_checksum
    USE dbcsr_tensor_split, ONLY: &
@@ -92,5 +92,8 @@ MODULE dbcsr_tensor_api
    PUBLIC :: dbcsr_t_batched_contract_init
    PUBLIC :: dbcsr_t_batched_contract_finalize
    PUBLIC :: dbcsr_t_ndims
+   PUBLIC :: dbcsr_t_pgrid_change_dims
+   PUBLIC :: dbcsr_t_reserved_block_indices
+   PUBLIC :: dbcsr_t_contract_index
 
 END MODULE dbcsr_tensor_api

src/tensors/dbcsr_tensor_types.F

@@ -78,6 +78,7 @@ MODULE dbcsr_tensor_types
       dbcsr_t_mp_dims_create, &
       dbcsr_t_nd_mp_comm, &
       dbcsr_t_nd_mp_free, &
+      dbcsr_t_pgrid_change_dims, &
       dbcsr_t_pgrid_create, &
       dbcsr_t_pgrid_create_expert, &
       dbcsr_t_pgrid_destroy, &
@@ -1230,6 +1231,34 @@ SUBROUTINE dbcsr_t_pgrid_remap(pgrid_in, map1_2d, map2_2d, pgrid_out)
       ENDIF
    END SUBROUTINE
 
+   SUBROUTINE dbcsr_t_pgrid_change_dims(pgrid, pdims)
+      !! change dimensions of an existing process grid.
+
+      TYPE(dbcsr_t_pgrid_type), INTENT(INOUT) :: pgrid
+         !! process grid to be changed
+      INTEGER, DIMENSION(:), INTENT(INOUT)    :: pdims
+         !! new process grid dimensions, should all be set > 0
+      TYPE(dbcsr_t_pgrid_type)                :: pgrid_tmp
+      INTEGER                                 :: nsplit, dimsplit
+      INTEGER, DIMENSION(:), ALLOCATABLE      :: map1_2d, map2_2d
+      TYPe(nd_to_2d_mapping)                  :: nd_index_grid
+      INTEGER, DIMENSION(2)                   :: pdims_2d
+
+      DBCSR_ASSERT(ALL(pdims > 0))
+      CALL dbcsr_tas_get_split_info(pgrid%tas_split_info, nsplit=nsplit, split_rowcol=dimsplit)
+      CALL dbcsr_t_get_mapping_info(pgrid%nd_index_grid, map1_2d=map1_2d, map2_2d=map2_2d)
+      CALL create_nd_to_2d_mapping(nd_index_grid, pdims, map1_2d, map2_2d, base=0, col_major=.FALSE.)
+      CALL dbcsr_t_get_mapping_info(nd_index_grid, dims_2d=pdims_2d)
+      IF(MOD(pdims_2d(dimsplit),nsplit) == 0) THEN
+         CALL dbcsr_t_pgrid_create_expert(pgrid%mp_comm_2d, pdims, pgrid_tmp, map1_2d=map1_2d, map2_2d=map2_2d, &
+                                          nsplit=nsplit, dimsplit=dimsplit)
+      ELSE
+         CALL dbcsr_t_pgrid_create_expert(pgrid%mp_comm_2d, pdims, pgrid_tmp, map1_2d=map1_2d, map2_2d=map2_2d)
+      ENDIF
+      CALL dbcsr_t_pgrid_destroy(pgrid)
+      pgrid = pgrid_tmp
+   END SUBROUTINE
+
    SUBROUTINE dbcsr_t_distribution_remap(dist_in, map1_2d, map2_2d, dist_out)
       TYPE(dbcsr_t_distribution_type), INTENT(IN)    :: dist_in
       INTEGER, DIMENSION(:), INTENT(IN) :: map1_2d, map2_2d