Merge branch 'main' into oshmem/mca/ucx/fix-double-cleanup

iyastreb · iyastreb · commit 5d35c73bdd92 · 2024-02-02T09:52:35.000+02:00
diff --git a/.github/workflows/ompi_mpi4py.yaml b/.github/workflows/ompi_mpi4py.yaml
@@ -0,0 +1,112 @@
+name: GitHub Action CI
+
+on: [pull_request]
+
+jobs:
+  mpi4py:
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+    - name: Configure hostname
+      run:  echo 127.0.0.1 `hostname` | sudo tee -a /etc/hosts > /dev/null
+      if:   ${{ runner.os == 'Linux' || runner.os == 'macOS' }}
+        
+    - name: Install depencencies
+      run:  sudo apt-get install -y -q
+              libnuma-dev
+      
+    - name: Checkout Open MPI
+      uses: actions/checkout@v4
+      with:
+       path: mpi-build
+       submodules: recursive
+
+    - name: Bootstrap Open MPI
+      run:  ./autogen.pl
+      working-directory: mpi-build
+
+    - name: Configure Open MPI
+      run: ./configure
+              --disable-dependency-tracking
+              --disable-sphinx
+              --disable-man-pages
+              --disable-mpi-fortran
+              --disable-oshmem
+              --enable-debug
+              --enable-mem-debug
+              LDFLAGS=-Wl,-rpath,/usr/local/lib
+      working-directory: mpi-build
+
+    - name: Build MPI
+      run:  make -j 2
+      working-directory: mpi-build
+
+    - name: Install MPI
+      run:  sudo make install
+      working-directory: mpi-build
+
+    - name: Tweak MPI
+      run:  |
+        # Tweak MPI
+        mca_params="$HOME/.openmpi/mca-params.conf"
+        mkdir -p "$(dirname "$mca_params")"
+        echo mpi_param_check = true >> "$mca_params"
+        echo mpi_show_handle_leaks = true >> "$mca_params"
+        echo rmaps_base_oversubscribe = true >> "$mca_params"
+        mca_params="$HOME/.prte/mca-params.conf"
+        mkdir -p "$(dirname "$mca_params")"
+        echo rmaps_default_mapping_policy = :oversubscribe >> "$mca_params"
+
+    - name: Show MPI
+      run:  ompi_info
+
+    - name: Show MPICC
+      run:  mpicc -show
+
+    - name: Use Python
+      uses: actions/setup-python@v5
+      with:
+        python-version: 3
+        architecture: x64
+
+    - name: Install Python packages (build)
+      run:  python -m pip install --upgrade
+              setuptools pip wheel
+
+    - name: Install Python packages (test)
+      run:  python -m pip install --upgrade
+              numpy cffi pyyaml
+
+    - name: Checkout mpi4py
+      uses: actions/checkout@v4
+      with:
+        repository: "mpi4py/mpi4py"
+
+    - name: Install mpi4py
+      run:  python -m pip install .
+      env:
+        CFLAGS: "-O0"
+
+    - name: Test mpi4py (singleton)
+      run:  python test/main.py -v
+      if:   ${{ true }}
+    - name: Test mpi4py (np=1)
+      run:  mpiexec -n 1 python test/main.py -v
+    - name: Test mpi4py (np=2)
+      run:  mpiexec -n 2 python test/main.py -v -f
+    - name: Test mpi4py (np=3)
+      run:  mpiexec -n 3 python test/main.py -v -f
+      if:   ${{ true }}
+      timeout-minutes: 20
+    - name: Test mpi4py (np=4)
+      run:  mpiexec -n 4 python test/main.py -v -f
+      if:   ${{ true }}
+      timeout-minutes: 20
+    - name: Test mpi4py (np=5)
+      run:  mpiexec -n 5 python test/main.py -v -f
+      if:   ${{ true }}
+      timeout-minutes: 20
+    - name: Test mpi4py.run
+      run:  python demo/test-run/test_run.py -v
+      if:   ${{ true }}
+      timeout-minutes: 20
diff --git a/ompi/communicator/comm.c b/ompi/communicator/comm.c
@@ -1618,7 +1618,7 @@ int ompi_comm_create_from_group (ompi_group_t *group, const char *tag, opal_info
 
     /*
      * setup predefined keyvals - see MPI Standard for predefined keyvals cached on 
-     * communicators created via MPI_Comm_from_group or MPI_Intercomm_create_from_groups
+     * communicators created via MPI_Comm_create_from_group or MPI_Intercomm_create_from_groups
      */
     ompi_attr_hash_init(&newcomp->c_keyhash);
     ompi_attr_set_int(COMM_ATTR,
diff --git a/ompi/communicator/comm_cid.c b/ompi/communicator/comm_cid.c
@@ -367,7 +367,7 @@ static int ompi_comm_ext_cid_new_block (ompi_communicator_t *newcomm, ompi_commu
             opal_show_help("help-comm.txt",
                            "MPI function not supported",
                            true,
-                           "MPI_Comm_from_group/MPI_Intercomm_from_groups",
+                           "MPI_Comm_create_from_group/MPI_Intercomm_create_from_groups",
                            msg_string);
             ret = MPI_ERR_UNSUPPORTED_OPERATION;
             break;
@@ -490,7 +490,7 @@ int ompi_comm_nextcid_nb (ompi_communicator_t *newcomm, ompi_communicator_t *com
 
     /* old CID algorighm */
 
-    /* if we got here and comm is NULL then that means the app is  invoking MPI-4 Sessions or later
+    /* if we got here and comm is NULL then that means the app is invoking MPI-4 Sessions or later
        functions but the pml does not support these functions so return not supported */
     if (NULL == comm) {
        char msg_string[1024];
@@ -499,7 +499,7 @@ int ompi_comm_nextcid_nb (ompi_communicator_t *newcomm, ompi_communicator_t *com
        opal_show_help("help-comm.txt",
                       "MPI function not supported",
                       true,
-                      "MPI_Comm_from_group/MPI_Intercomm_from_groups",
+                      "MPI_Comm_create_from_group/MPI_Intercomm_create_from_groups",
                       msg_string);
 
         return MPI_ERR_UNSUPPORTED_OPERATION;
diff --git a/opal/mca/accelerator/rocm/accelerator_rocm_component.c b/opal/mca/accelerator/rocm/accelerator_rocm_component.c
@@ -6,7 +6,7 @@
  *                         reserved.
  * Copyright (c) 2017-2022 Amazon.com, Inc. or its affiliates.
  *                         All Rights reserved.
- * Copyright (c) 2022      Advanced Micro Devices, Inc. All Rights reserved.
+ * Copyright (c) 2022-2024 Advanced Micro Devices, Inc. All Rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -17,13 +17,12 @@
 #include "opal_config.h"
 
 #include <stdio.h>
-#include <dlfcn.h>
 
 #include "opal/mca/dl/base/base.h"
 #include "opal/runtime/opal_params.h"
 #include "accelerator_rocm.h"
 
-int opal_accelerator_rocm_memcpy_async = 1;
+int opal_accelerator_rocm_memcpy_async = 0;
 int opal_accelerator_rocm_verbose = 0;
 size_t opal_accelerator_rocm_memcpyD2H_limit=1024;
 size_t opal_accelerator_rocm_memcpyH2D_limit=1048576;
@@ -149,9 +148,9 @@ static int accelerator_rocm_component_register(void)
                                  &opal_accelerator_rocm_memcpyH2D_limit);
 
     /* Use this flag to test async vs sync copies */
-    opal_accelerator_rocm_memcpy_async = 1;
+    opal_accelerator_rocm_memcpy_async = 0;
     (void) mca_base_var_register("ompi", "mpi", "accelerator_rocm", "memcpy_async",
-                                 "Set to 0 to force using hipMemcpy instead of hipMemcpyAsync",
+                                 "Set to 1 to force using hipMemcpyAsync instead of hipMemcpy",
                                  MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_9,
                                  MCA_BASE_VAR_SCOPE_READONLY, &opal_accelerator_rocm_memcpy_async);
 
