Introduce support for executing the IPC conduit without MPI (#153)

* relax MPI dependency from code

This commit (series) removes the strict dependency on MPI in code base.
rocSHMEM will still be compiled with MPI, but the goal is to make the
code work even if MPI_Init_thread has not been invoked, at least for
certain, well-defined scenarios. Hence, the goal is not remove any
mentioning of MPI from rocSHMEM, but to ensure correct execution of the
ipc conduit even if the library has been initialized using other means.

Details:
 - add non-MPI version of remote_heap and WindowInfo classes
 - host interfaces work on WindowInfoMPI, they will not work with the
   non-MPI code path. Since it is unclear whether we plan to support the
   host interfaces at all, this is probably not a major limitation.

* update symmetric_heap structures and backend

* first cut on initialization

and enabling non-MPI initialization of the IPCBackend

* add non-MPI hostInterface methods

at the moment, only barrier_all and sync_all are explicitely supported.

* add non-mpi version of ipc_policy

and a number of smaller fixes required in other files.
A small init/finalize test already passes now with the branch.

* add non-mpi team_split_strided code

* minor fixes for non-MPI use-case

* disable symmetric-heap-window-ionfo test

disable this test for now just to make the compilation pass. Will have
to rework it.

* make no-mpi great again

after rebasing on top of the MPI singleton changes.

* enable running functional tests with uuid init

to run the functional tests using rocshmem_init_attr and the uuid
mechanism requires
a) a PMIx installation on the system
b) setting the environment variable ROCSHMEM_TEST_UUID=1

* fix multi-team creation bug

fix a bug occuring when creating many teams, which was the result of
incorrectly applying two indices in our own implementation of Allreduce.

* make unit tests pass again

* reverse offload was impacted by code change

fix the RO conduit to cope wioth the non-MPI path introduced for the IPC
conduit.

* update to cmake logic to find pmix

* Update src/memory/window_info.hpp

Co-authored-by: Yiltan <ytemucin@amd.com>

* Update CMakeLists.txt

Co-authored-by: Yiltan <ytemucin@amd.com>

* document ROCSHMEM_UNIQUEID_NO_MPI

* rename env. variable to UNIQUEID_WITH_MPI

* update host.cpp to use USE_HDP_FLUSH macro

instead of the deprecated USE_COHERENT_HEAP.

* add note for running example with RO conduit

add a note clarifying that running init_attr_test from the example
directory requires setting an additional environment variable with the
RO conduit.

* Find PMIx in more cases, only apply pmix build options to the test that
needs it, if OMPI_COMM_WORLD_LOCA_RANK is not setenv, abort

---------

Co-authored-by: Yiltan <ytemucin@amd.com>
Co-authored-by: Aurelien Bouteiller <abouteil@amd.com>

tests/functional_tests/CMakeLists.txt

@@ -71,6 +71,9 @@ target_sources(
 ###############################################################################
 # ROCSHMEM
 ###############################################################################
+include(${CMAKE_SOURCE_DIR}/cmake/find_pmix.cmake)
+check_pmix()
+
 if (BUILD_TESTS_ONLY)
   find_package(MPI REQUIRED)
   find_package(hip REQUIRED PATHS /opt/rocm)
@@ -84,8 +87,21 @@ if (BUILD_TESTS_ONLY)
   )
 endif()
 
+target_compile_definitions(
+  ${PROJECT_NAME}
+  PRIVATE
+    $<$<BOOL:HAVE_PMIX>:HAVE_PMIX=1>
+)
+
+target_include_directories(
+  ${PROJECT_NAME}
+  PRIVATE
+    ${PMIX_INCLUDE_DIRECTORIES}
+)
+
 target_link_libraries(
   ${PROJECT_NAME}
   PRIVATE
     roc::rocshmem
+    ${PMIX_LIBRARIES}
 )

tests/functional_tests/test_driver.cpp

@@ -28,6 +28,104 @@
 #include "tester.hpp"
 #include "tester_arguments.hpp"
 
+#if defined(HAVE_PMIX)
+#include <pmix.h>
+
+static pmix_proc_t pmix_myproc;
+static pmix_proc_t pmix_proc;
+
+static void init_pmix(int *rank, int *nranks)
+{
+    pmix_status_t rc;
+    pmix_value_t *val;
+
+    if (PMIX_SUCCESS != (rc = PMIx_Init(&pmix_myproc, NULL, 0))) {
+      std::cerr << "Rank " << pmix_myproc.rank << " PMIx_Init failed: " << rc << std::endl;
+      abort();
+    }
+#ifdef VERBOSE
+    printf("Client ns %s rank %d: Running\n", pmix_myproc.nspace, pmix_myproc.rank);
+#endif
+    PMIX_PROC_CONSTRUCT(&pmix_proc);
+    PMIX_LOAD_PROCID(&pmix_proc, pmix_myproc.nspace, PMIX_RANK_WILDCARD);
+
+    /* get our job size */
+    if (PMIX_SUCCESS != (rc = PMIx_Get(&pmix_proc, PMIX_JOB_SIZE, NULL, 0, &val))) {
+      std::cerr << "Rank " << pmix_myproc.rank << " PMIx_Get universe size failed: "
+                <<  rc << std::endl;
+        abort();
+    }
+
+    *nranks = val->data.uint32;
+    *rank   = pmix_myproc.rank;
+
+    PMIX_VALUE_RELEASE(val);
+    return;
+}
+
+static void pmix_bcast(void *buf, size_t nbytes, char *key, int root)
+{
+    pmix_status_t rc;
+    pmix_value_t value;
+    pmix_value_t *val;
+    pmix_info_t *info;
+    bool flag;
+
+    if (pmix_myproc.rank == root) {
+      value.type = PMIX_BYTE_OBJECT;
+      value.data.bo.bytes = (char *) (buf);
+      value.data.bo.size = nbytes;
+
+      rc = PMIx_Put(PMIX_GLOBAL, key, &value);
+      if (PMIX_SUCCESS != rc) {
+        std::cerr << "Rank " << pmix_myproc.rank << " PMIx_Put failed: " << rc << std::endl;
+        abort();
+      }
+
+      /* push the data to our PMIx server */
+      if (PMIX_SUCCESS != (rc = PMIx_Commit())) {
+        std::cerr <<  "Rank " << pmix_myproc.rank << " PMIx_Commit failed: " << rc << std::endl;
+        abort();
+      }
+    }
+
+    /* call fence to synchronize with our peers - instruct
+     * the fence operation to collect and return all "put"
+     * data from our peers */
+    PMIX_INFO_CREATE(info, 1);
+    flag = true;
+        PMIX_INFO_LOAD(info, PMIX_COLLECT_DATA, &flag, PMIX_BOOL);
+    if (PMIX_SUCCESS != (rc = PMIx_Fence(&pmix_proc, 1, info, 1))) {
+      std::cerr <<  "Rank " << pmix_myproc.rank << " PMIx_Fence failed: " << rc << std::endl;
+      abort();
+    }
+    PMIX_INFO_FREE(info, 1);
+
+    pmix_proc.rank = 0;
+    if (PMIX_SUCCESS != (rc = PMIx_Get(&pmix_proc, key, NULL, 0, &val))) {
+      std::cerr <<  "Rank " << pmix_myproc.rank << " PMIx_Get failed: " << rc << std::endl;
+      abort();
+    }
+    if (PMIX_BYTE_OBJECT != val->type) {
+      std::cerr <<  "Rank " << pmix_myproc.rank << " PMIx_Get returned wrong type: " << val->type  << std::endl;
+      PMIX_VALUE_RELEASE(val);
+      abort();
+    }
+
+    if (pmix_myproc.rank != root) {
+      if (NULL == val->data.bo.bytes) {
+        std::cerr <<  "Rank " << pmix_myproc.rank << " PMIx_Get %d returned NULL pointer\n";
+        PMIX_VALUE_RELEASE(val);
+        abort();
+      }
+      memcpy (buf, val->data.bo.bytes, val->data.bo.size);
+    }
+    PMIX_VALUE_RELEASE(val);
+
+    return;
+}
+#endif
+
 using namespace rocshmem;
 
 int main(int argc, char *argv[]) {
@@ -40,12 +138,61 @@ int main(int argc, char *argv[]) {
    * Select a GPU
    */
   char* ompi_local_rank = getenv("OMPI_COMM_WORLD_LOCAL_RANK");
+  if (nullptr == ompi_local_rank) {
+    printf("Could not determine local rank, use Open MPI `mpiexec`\n");
+    abort();
+  }
   CHECK_HIP(hipSetDevice(atoi(ompi_local_rank)));
 
   /**
    * Must initialize rocshmem to access arguments needed by the tester.
    */
+#ifdef HAVE_PMIX
+  int test_uuid = 0;
+  char *rocshmem_test_uuid = getenv("ROCSHMEM_TEST_UUID");
+  if (rocshmem_test_uuid != nullptr) {
+    test_uuid = atoi(rocshmem_test_uuid);
+  }
+
+  if (test_uuid) {
+    int ret;
+    int rank, nranks;
+    rocshmem_uniqueid_t uid;
+    rocshmem_init_attr_t attr;
+
+    init_pmix(&rank, &nranks);
+    if (rank == 0) {
+      ret = rocshmem_get_uniqueid (&uid);
+      if (ret != ROCSHMEM_SUCCESS) {
+        std::cout << rank << ": Error in rocshmem_get_uniqueid. Aborting.\n";
+        abort();
+      }
+    }
+
+    char key[] = "rocshmem-uuid";
+    pmix_bcast(&uid, sizeof(rocshmem_uniqueid_t), key, 0);
+
+    ret = rocshmem_set_attr_uniqueid_args(rank, nranks, &uid, &attr);
+    if (ret != ROCSHMEM_SUCCESS) {
+      std::cout << rank << ": Error in rocshmem_set_attr_uniqueid_args. Aborting.\n";
+      abort();
+    }
+
+    ret = rocshmem_init_attr(ROCSHMEM_INIT_WITH_UNIQUEID, &attr);
+    if (ret != ROCSHMEM_SUCCESS) {
+      std::cout << rank << ": Error in rocshmem_init_attr. Aborting.\n";
+      abort();
+    }
+
+#ifdef VERBOSE
+    std::cout << rank << ": rocshmem_init_attr SUCCESS\n";
+#endif
+  } else {
+    rocshmem_init();
+  }
+#else
   rocshmem_init();
+#endif
 
   /**
    * Now grab the arguments from rocshmem.
@@ -77,5 +224,11 @@ int main(int argc, char *argv[]) {
    */
   rocshmem_finalize();
 
+#ifdef HAVE_PMIX
+  if (test_uuid) {
+    PMIx_Finalize(NULL, 0);
+  }
+#endif
+
   return 0;
 }

tests/functional_tests/tester.cpp

@@ -617,7 +617,7 @@ void flush_hdp() {
 }
 
 void Tester::barrier() {
-  MPI_Barrier(MPI_COMM_WORLD);
+  rocshmem_barrier_all();
   flush_hdp();
 }