eth-cscs · msimberg · Mar 19, 2024 · Mar 21, 2024 · Mar 20, 2024 · Mar 20, 2024
diff --git a/.cmake-format.py b/.cmake-format.py
@@ -17,6 +17,17 @@
         "ARGUMENTS": '*',
         "USE_MAIN": '1',
         "MPIRANKS": '1',
+        "CATEGORY": '1',
+      }
+    },
+    "DLAF_addTargetTest": {
+      "pargs": 1,
+      "flags": [],
+      "kwargs": {
+        "ARGUMENTS": '*',
+        "USE_MAIN": '1',
+        "MPIRANKS": '1',
+        "CATEGORY": '1',
       }
     },
     "DLAF_addMiniapp": {

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -185,6 +185,28 @@ add_subdirectory(external)
 # ---------------------------------------------------------------------------
 add_subdirectory(src)
 
+# ---------------------------------------------------------------------------
+# Test options (apply to unit tests and miniapps as tests)
+# ---------------------------------------------------------------------------
+set(DLAF_PRESET_OPTIONS "plain-mpi" "slurm" "custom")
+set(DLAF_MPI_PRESET "plain-mpi" CACHE STRING "Select a preset to use")
+set_property(CACHE DLAF_MPI_PRESET PROPERTY STRINGS ${DLAF_PRESET_OPTIONS})
+option(
+  DLAF_CI_RUNNER_USES_MPIRUN
+  "Remove mpiexec command for tests executed by ctest. This option is to be used if the CI runner executes the tests with <mpiexec + options> ctest -L RANK_<spawned MPI ranks>"
+  OFF
+)
+
+# On some machines, tests using multiple ranks + oversubscribing run
+# significantly faster when threads are not pinned.
+option(DLAF_TEST_THREAD_BINDING_ENABLED "If OFF disables pika thread binding." ON)
+
+# If DLAF_CI_RUNNER_USES_MPIRUN=on we don't want to use any preset, so we just go for the custom one
+# without setting any variable.
+if(DLAF_CI_RUNNER_USES_MPIRUN)
+  set(DLAF_MPI_PRESET "custom" CACHE STRING "" FORCE)
+endif()
+
 # ---------------------------------------------------------------------------
 # mini Apps
 # ---------------------------------------------------------------------------

diff --git a/ci/ctest_to_gitlab.sh b/ci/ctest_to_gitlab.sh
@@ -48,7 +48,7 @@ upload_reports:
 "
 JOB_TEMPLATE="
 
-{{LABEL}}:
+{{CATEGORY_LABEL_NOPREFIX}}_{{RANK_LABEL}}:
   stage: test
   extends: .daint
   variables:
@@ -61,7 +61,7 @@ JOB_TEMPLATE="
     USE_MPI: 'YES'
     DISABLE_AFTER_SCRIPT: 'YES'
     DLAF_HDF5_TEST_OUTPUT_PATH: \$CI_PROJECT_DIR
-  script: mpi-ctest -L {{LABEL}}
+  script: mpi-ctest -L {{CATEGORY_LABEL}} -L {{RANK_LABEL}}
   artifacts:
     paths:
       - codecov-reports/"
@@ -87,7 +87,7 @@ variables:
 "
 
 JOB_TEMPLATE="
-{{LABEL}}:
+{{CATEGORY_LABEL_NOPREFIX}}_{{RANK_LABEL}}:
   stage: test
   extends: .daint
   variables:
@@ -100,20 +100,30 @@ JOB_TEMPLATE="
     USE_MPI: 'YES'
     DISABLE_AFTER_SCRIPT: 'YES'
     DLAF_HDF5_TEST_OUTPUT_PATH: \$CI_PROJECT_DIR
-  script: mpi-ctest -L {{LABEL}}"
+  script: mpi-ctest -L {{CATEGORY_LABEL}} -L {{RANK_LABEL}}"
 fi
 
 JOBS=""
 
-for label in `ctest --print-labels | egrep -o "RANK_[1-9][0-9]?"`; do
-    N=`echo "$label" | sed "s/RANK_//"`
-    C=$(( THREADS_PER_NODE / N ))
-
-    JOB=`echo "$JOB_TEMPLATE" | sed "s|{{LABEL}}|$label|g" \
-                              | sed "s|{{NTASKS}}|$N|g" \
-                              | sed "s|{{CPUS_PER_TASK}}|$C|g"`
-
-    JOBS="$JOBS$JOB"
+for rank_label in `ctest --print-labels | egrep -o "RANK_[1-9][0-9]?"`; do
+    for category_label in `ctest --print-labels | egrep -o "CATEGORY_[A-Z]+"`; do
+        N=`echo "$rank_label" | sed "s/RANK_//"`
+        C=$(( THREADS_PER_NODE / N ))
+
+        # Skip label combinations that match no tests
+        if [[ "$(ctest -N -L $category_label -L $rank_label | tail -n1)" == "Total Tests: 0" ]]; then
+            continue
+        fi
+
+        category_label_noprefix=`echo "$category_label" | sed "s/CATEGORY_//"`
+        JOB=`echo "$JOB_TEMPLATE" | sed "s|{{CATEGORY_LABEL_NOPREFIX}}|$category_label_noprefix|g" \
+                                  | sed "s|{{CATEGORY_LABEL}}|$category_label|g" \
+                                  | sed "s|{{RANK_LABEL}}|$rank_label|g" \
+                                  | sed "s|{{NTASKS}}|$N|g" \
+                                  | sed "s|{{CPUS_PER_TASK}}|$C|g"`
+
+        JOBS="$JOBS$JOB"
+    done
 done
 
 echo "${BASE_TEMPLATE/'{{JOBS}}'/$JOBS}"
diff --git a/cmake/DLAF_AddTest.cmake b/cmake/DLAF_AddTest.cmake
@@ -15,6 +15,7 @@
 #   [LIBRARIES <arguments for target_link_libraries>]
 #   [MPIRANKS <number of rank>]
 #   [USE_MAIN {PLAIN | PIKA | MPI | MPIPIKA}]
+#   [CATEGORY <category>]
 # )
 #
 # At least one source file has to be specified, while other parameters are optional.
@@ -45,6 +46,8 @@
 #       include/
 # )
 
+set(DLAF_PIKATEST_EXTRA_ARGS "" CACHE STRING "Extra arguments for tests with pika")
+
 # Check if LIST_NAME contains at least an element that matches ELEMENT_REGEX. If not, add FALLBACK
 # to the list.
 function(_set_element_to_fallback_value LIST_NAME ELEMENT_REGEX FALLBACK)
@@ -57,72 +60,70 @@ function(_set_element_to_fallback_value LIST_NAME ELEMENT_REGEX FALLBACK)
   endif()
 endfunction()
 
-function(DLAF_addTest test_target_name)
+function(DLAF_addTargetTest test_target_name)
   set(options "")
-  set(oneValueArgs MPIRANKS USE_MAIN)
-  set(multiValueArgs SOURCES COMPILE_DEFINITIONS INCLUDE_DIRS LIBRARIES ARGUMENTS)
-  cmake_parse_arguments(DLAF_AT "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  set(oneValueArgs CATEGORY MPIRANKS USE_MAIN)
+  set(multiValueArgs ARGUMENTS)
+  cmake_parse_arguments(DLAF_ATT "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
 
   ### Checks
-  if(DLAF_AT_UNPARSED_ARGUMENTS)
-    message(FATAL_ERROR "Unknown arguments ${DLAF_AT_UNPARSED_ARGUMENTS}")
-  endif()
-
-  if(NOT DLAF_AT_SOURCES)
-    message(FATAL_ERROR "No sources specified for this test")
+  if(DLAF_ATT_UNPARSED_ARGUMENTS)
+    message(FATAL_ERROR "Unknown arguments ${DLAF_ATT_UNPARSED_ARGUMENTS}")
   endif()
 
   set(IS_AN_MPI_TEST FALSE)
   set(IS_AN_PIKA_TEST FALSE)
-  if(NOT DLAF_AT_USE_MAIN)
-    set(_gtest_tgt gtest)
-  elseif(DLAF_AT_USE_MAIN STREQUAL PLAIN)
-    set(_gtest_tgt gtest_main)
-  elseif(DLAF_AT_USE_MAIN STREQUAL PIKA)
-    set(_gtest_tgt DLAF_gtest_pika_main)
+  if(NOT DLAF_ATT_USE_MAIN OR DLAF_ATT_USE_MAIN STREQUAL PLAIN)
+
+  elseif(DLAF_ATT_USE_MAIN STREQUAL PIKA)
     set(IS_AN_PIKA_TEST TRUE)
-  elseif(DLAF_AT_USE_MAIN STREQUAL MPI)
-    set(_gtest_tgt DLAF_gtest_mpi_main)
+  elseif(DLAF_ATT_USE_MAIN STREQUAL MPI)
     set(IS_AN_MPI_TEST TRUE)
-  elseif(DLAF_AT_USE_MAIN STREQUAL MPIPIKA)
-    set(_gtest_tgt DLAF_gtest_mpipika_main)
+  elseif(DLAF_ATT_USE_MAIN STREQUAL MPIPIKA)
     set(IS_AN_MPI_TEST TRUE)
     set(IS_AN_PIKA_TEST TRUE)
-  elseif(DLAF_AT_USE_MAIN STREQUAL CAPI)
-    set(_gtest_tgt DLAF_gtest_mpi_main)
+  elseif(DLAF_ATT_USE_MAIN STREQUAL CAPI)
     set(IS_AN_MPI_TEST TRUE)
   else()
-    message(FATAL_ERROR "USE_MAIN=${DLAF_AT_USE_MAIN} is not a supported option")
+    message(FATAL_ERROR "USE_MAIN=${DLAF_ATT_USE_MAIN} is not a supported option")
   endif()
 
+  set(_TEST_LABELS)
+
+  if(NOT DLAF_ATT_CATEGORY)
+    set(DLAF_ATT_CATEGORY "UNIT")
+  endif()
+
+  list(APPEND _TEST_LABELS "CATEGORY_${DLAF_ATT_CATEGORY}")
+
   if(IS_AN_MPI_TEST)
-    if(NOT DLAF_AT_MPIRANKS)
+    if(NOT DLAF_ATT_MPIRANKS)
       message(FATAL_ERROR "You are asking for an MPI external main without specifying MPIRANKS")
     endif()
-    if(NOT DLAF_AT_MPIRANKS GREATER 0)
-      message(FATAL_ERROR "Wrong MPIRANKS number ${DLAF_AT_MPIRANKS}")
+    if(NOT DLAF_ATT_MPIRANKS GREATER 0)
+      message(FATAL_ERROR "Wrong MPIRANKS number ${DLAF_ATT_MPIRANKS}")
     endif()
-    if(DLAF_AT_MPIRANKS GREATER MPIEXEC_MAX_NUMPROCS)
+    if(DLAF_ATT_MPIRANKS GREATER MPIEXEC_MAX_NUMPROCS)
       message(
         WARNING
           "\
-      YOU ARE ASKING FOR ${DLAF_AT_MPIRANKS} RANKS, BUT THERE ARE JUST ${MPIEXEC_MAX_NUMPROCS} CORES.
+          YOU ARE ASKING FOR ${DLAF_ATT_MPIRANKS} RANKS, BUT THERE ARE JUST ${MPIEXEC_MAX_NUMPROCS} CORES.
       You can adjust MPIEXEC_MAX_NUMPROCS value to suppress this warning.
       Using OpenMPI may require to set the environment variable OMPI_MCA_rmaps_base_oversubscribe=1."
       )
     endif()
   else()
-    if(DLAF_AT_MPIRANKS)
+    if(DLAF_ATT_MPIRANKS)
       message(FATAL_ERROR "You specified MPIRANKS and asked for an external main without MPI")
     else()
-      set(DLAF_AT_MPIRANKS 1)
+      set(DLAF_ATT_MPIRANKS 1)
     endif()
   endif()
 
   ### Test target
   set(DLAF_TEST_RUNALL_WITH_MPIEXEC OFF CACHE BOOL "Run all tests using the workload manager.")
 
-  set(_TEST_ARGUMENTS ${DLAF_AT_ARGUMENTS})
+  set(_TEST_ARGUMENTS ${DLAF_ATT_ARGUMENTS})
 
   if(DLAF_TEST_RUNALL_WITH_MPIEXEC OR IS_AN_MPI_TEST)
     if(MPIEXEC_NUMCORE_FLAG)
@@ -132,7 +133,7 @@ function(DLAF_addTest test_target_name)
         set(_CORES_PER_RANK 1)
       endif()
 
-      math(EXPR DLAF_CORE_PER_RANK "${_CORES_PER_RANK}/${DLAF_AT_MPIRANKS}")
+      math(EXPR DLAF_CORE_PER_RANK "${_CORES_PER_RANK}/${DLAF_ATT_MPIRANKS}")
 
       if(NOT DLAF_CORE_PER_RANK)
         set(DLAF_CORE_PER_RANK 1)
@@ -148,29 +149,29 @@ function(DLAF_addTest test_target_name)
     else()
       separate_arguments(MPIEXEC_PREFLAGS)
       set(_TEST_COMMAND
-          ${MPIEXEC_EXECUTABLE} ${MPIEXEC_NUMPROC_FLAG} ${DLAF_AT_MPIRANKS} ${_MPI_CORE_ARGS}
+          ${MPIEXEC_EXECUTABLE} ${MPIEXEC_NUMPROC_FLAG} ${DLAF_ATT_MPIRANKS} ${_MPI_CORE_ARGS}
           ${MPIEXEC_PREFLAGS} ${DLAF_TEST_PREFLAGS} $<TARGET_FILE:${test_target_name}>
           ${DLAF_TEST_POSTFLAGS} ${MPIEXEC_POSTFLAGS}
       )
     endif()
-    set(_TEST_LABEL "RANK_${DLAF_AT_MPIRANKS}")
+    list(APPEND _TEST_LABELS "RANK_${DLAF_ATT_MPIRANKS}")
 
   else()
     # ----- Classic test
     set(_TEST_COMMAND ${DLAF_TEST_PREFLAGS} $<TARGET_FILE:${test_target_name}> ${DLAF_TEST_POSTFLAGS})
-    set(_TEST_LABEL "RANK_1")
+    list(APPEND _TEST_LABELS "RANK_1")
   endif()
 
   if(IS_AN_PIKA_TEST)
     separate_arguments(_PIKA_EXTRA_ARGS_LIST UNIX_COMMAND ${DLAF_PIKATEST_EXTRA_ARGS})
 
     # --pika:bind=none is useful just in case more ranks are going to be allocated on the same node.
-    if(IS_AN_MPI_TEST AND (DLAF_AT_MPIRANKS GREATER 1) AND (NOT DLAF_TEST_THREAD_BINDING_ENABLED))
+    if(IS_AN_MPI_TEST AND (DLAF_ATT_MPIRANKS GREATER 1) AND (NOT DLAF_TEST_THREAD_BINDING_ENABLED))
       _set_element_to_fallback_value(_PIKA_EXTRA_ARGS_LIST "--pika:bind" "--pika:bind=none")
     endif()
 
     if(IS_AN_MPI_TEST AND DLAF_MPI_PRESET STREQUAL "plain-mpi")
-      math(EXPR _DLAF_PIKA_THREADS "${MPIEXEC_MAX_NUMPROCS}/${DLAF_AT_MPIRANKS}")
+      math(EXPR _DLAF_PIKA_THREADS "${MPIEXEC_MAX_NUMPROCS}/${DLAF_ATT_MPIRANKS}")
 
       if(_DLAF_PIKA_THREADS LESS 2)
         set(_DLAF_PIKA_THREADS 2)
@@ -186,16 +187,16 @@ function(DLAF_addTest test_target_name)
 
   # Special treatment for C API tests
   # C API tests require pika arguments to be hard-coded in the test file
-  if(DLAF_AT_USE_MAIN STREQUAL CAPI)
+  if(DLAF_ATT_USE_MAIN STREQUAL CAPI)
     separate_arguments(_PIKA_EXTRA_ARGS_LIST_CAPI UNIX_COMMAND ${DLAF_PIKATEST_EXTRA_ARGS})
 
     # --pika:bind=none is useful just in case more ranks are going to be allocated on the same node.
-    if((DLAF_AT_MPIRANKS GREATER 1) AND (NOT DLAF_TEST_THREAD_BINDING_ENABLED))
+    if((DLAF_ATT_MPIRANKS GREATER 1) AND (NOT DLAF_TEST_THREAD_BINDING_ENABLED))
       _set_element_to_fallback_value(_PIKA_EXTRA_ARGS_LIST_CAPI "--pika:bind" "--pika:bind=none")
     endif()
 
     if(IS_AN_MPI_TEST AND DLAF_MPI_PRESET STREQUAL "plain-mpi")
-      math(EXPR _DLAF_PIKA_THREADS "${MPIEXEC_MAX_NUMPROCS}/${DLAF_AT_MPIRANKS}")
+      math(EXPR _DLAF_PIKA_THREADS "${MPIEXEC_MAX_NUMPROCS}/${DLAF_ATT_MPIRANKS}")
 
       if(_DLAF_PIKA_THREADS LESS 2)
         set(_DLAF_PIKA_THREADS 2)
@@ -215,11 +216,50 @@ function(DLAF_addTest test_target_name)
 
   endif()
 
+  add_test(NAME ${test_target_name} COMMAND ${_TEST_COMMAND} ${_TEST_ARGUMENTS})
+  set_tests_properties(${test_target_name} PROPERTIES LABELS "${_TEST_LABELS}")
+endfunction()
+
+function(DLAF_addTest test_target_name)
+  set(options "")
+  set(oneValueArgs MPIRANKS USE_MAIN)
+  set(multiValueArgs SOURCES COMPILE_DEFINITIONS INCLUDE_DIRS LIBRARIES ARGUMENTS)
+  cmake_parse_arguments(DLAF_AT "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+
+  ### Checks
+  if(DLAF_AT_UNPARSED_ARGUMENTS)
+    message(FATAL_ERROR "Unknown arguments ${DLAF_AT_UNPARSED_ARGUMENTS}")
+  endif()
+
+  if(NOT DLAF_AT_SOURCES)
+    message(FATAL_ERROR "No sources specified for this test")
+  endif()
+
+  if(NOT DLAF_AT_USE_MAIN)
+    set(_gtest_tgt gtest)
+  elseif(DLAF_AT_USE_MAIN STREQUAL PLAIN)
+    set(_gtest_tgt gtest_main)
+  elseif(DLAF_AT_USE_MAIN STREQUAL PIKA)
+    set(_gtest_tgt DLAF_gtest_pika_main)
+  elseif(DLAF_AT_USE_MAIN STREQUAL MPI)
+    set(_gtest_tgt DLAF_gtest_mpi_main)
+  elseif(DLAF_AT_USE_MAIN STREQUAL MPIPIKA)
+    set(_gtest_tgt DLAF_gtest_mpipika_main)
+  elseif(DLAF_AT_USE_MAIN STREQUAL CAPI)
+    set(_gtest_tgt DLAF_gtest_mpi_main)
+  else()
+    message(FATAL_ERROR "USE_MAIN=${DLAF_AT_USE_MAIN} is not a supported option")
+  endif()
+
   ### Test executable target
   add_executable(${test_target_name} ${DLAF_AT_SOURCES})
   target_link_libraries(
     ${test_target_name} PRIVATE ${_gtest_tgt} DLAF_test ${DLAF_AT_LIBRARIES} dlaf.prop_private
   )
+  set(IS_AN_MPI_TEST FALSE)
+  if(DLAF_AT_USE_MAIN MATCHES MPI OR DLAF_AT_USE_MAIN STREQUAL CAPI)
+    set(IS_AN_MPI_TEST TRUE)
+  endif()
   target_compile_definitions(
     ${test_target_name} PRIVATE ${DLAF_AT_COMPILE_DEFINITIONS} $<$<BOOL:${IS_AN_MPI_TEST}>:
                                 NUM_MPI_RANKS=${DLAF_AT_MPIRANKS}>
@@ -229,8 +269,6 @@ function(DLAF_addTest test_target_name)
   )
   target_add_warnings(${test_target_name})
   DLAF_addPrecompiledHeaders(${test_target_name})
-  add_test(NAME ${test_target_name} COMMAND ${_TEST_COMMAND} ${_TEST_ARGUMENTS})
-  set_tests_properties(${test_target_name} PROPERTIES LABELS "${_TEST_LABEL}")
 
   ### DEPLOY
   include(GNUInstallDirs)
@@ -242,4 +280,12 @@ function(DLAF_addTest test_target_name)
             RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
     )
   endif()
+
+  ### Test
+  DLAF_addTargetTest(
+    ${test_target_name}
+    MPIRANKS ${DLAF_AT_MPIRANKS}
+    USE_MAIN ${DLAF_AT_USE_MAIN}
+    ARGUMENTS ${DLAF_AT_ARGUMENTS}
+  )
 endfunction()
diff --git a/include/dlaf/communication/communicator_grid.h b/include/dlaf/communication/communicator_grid.h
@@ -131,6 +131,11 @@ class CommunicatorGrid {
       return col_communicator_pipeline();
   }
 
+  /// Synchronization barrier for all communicators in the grid
+  ///
+  /// This function returns when all communications on the communicators in the grid have completed.
+  void wait_all_communicators();
+
   /// Prints information about the CommunicationGrid.
   friend std::ostream& operator<<(std::ostream& out, const CommunicatorGrid& grid) {
     return out << "position=" << grid.position_ << ", size=" << grid.grid_size_;