Merge pull request #1469 from lplewa/hwloc_filter

make topology_init faster

Author: KFilipek

src/libumf.c

@@ -74,7 +74,11 @@ umf_result_t umfInit(void) {
     if (TRACKER) {
         LOG_DEBUG("UMF library initialized");
     }
-
+#if !defined(UMF_NO_HWLOC)
+    // some benchmarks uses multiple forks, and topology initialization is very slow
+    // so if we initialize topology before the first fork, we can get significant performance gain.
+    umfGetTopologyReduced();
+#endif
     return UMF_RESULT_SUCCESS;
 }
 

src/memspaces/memspace_host_all.c

@@ -37,7 +37,7 @@ static umf_result_t umfMemspaceHostAllCreate(umf_memspace_handle_t *hMemspace) {
 
     umf_result_t umf_ret = UMF_RESULT_SUCCESS;
 
-    hwloc_topology_t topology = umfGetTopology();
+    hwloc_topology_t topology = umfGetTopologyReduced();
     if (!topology) {
         // TODO: What would be an approrpiate err?
         return UMF_RESULT_ERROR_UNKNOWN;

src/memtargets/memtarget_numa.c

@@ -218,7 +218,7 @@ static umf_result_t numa_get_capacity(void *memTarget, size_t *capacity) {
         return UMF_RESULT_ERROR_INVALID_ARGUMENT;
     }
 
-    hwloc_topology_t topology = umfGetTopology();
+    hwloc_topology_t topology = umfGetTopologyReduced();
     if (!topology) {
         return UMF_RESULT_ERROR_NOT_SUPPORTED;
     }

src/provider/provider_os_memory.c

@@ -98,6 +98,7 @@ umf_result_t umfOsMemoryProviderParamsSetPartitions(
 #include "critnib.h"
 #include "libumf.h"
 #include "provider_os_memory_internal.h"
+#include "topology.h"
 #include "utils_common.h"
 #include "utils_concurrency.h"
 #include "utils_log.h"
@@ -629,27 +630,19 @@ static umf_result_t os_initialize(const void *params, void **provider) {
 
     memset(os_provider, 0, sizeof(*os_provider));
 
-    int r = hwloc_topology_init(&os_provider->topo);
-    if (r) {
-        LOG_ERR("HWLOC topology init failed");
-        ret = UMF_RESULT_ERROR_OUT_OF_HOST_MEMORY;
-        goto err_free_os_provider;
-    }
-
-    r = hwloc_topology_load(os_provider->topo);
-    if (r) {
+    os_provider->topo = umfGetTopologyReduced();
+    if (!os_provider->topo) {
         os_store_last_native_error(UMF_OS_RESULT_ERROR_TOPO_DISCOVERY_FAILED,
                                    0);
         LOG_ERR("HWLOC topology discovery failed");
         ret = UMF_RESULT_ERROR_MEMORY_PROVIDER_SPECIFIC;
-        goto err_destroy_hwloc_topology;
     }
 
     os_provider->fd_offset_map = critnib_new(NULL, NULL);
     if (!os_provider->fd_offset_map) {
         LOG_ERR("creating file descriptor offset map failed");
         ret = UMF_RESULT_ERROR_OUT_OF_HOST_MEMORY;
-        goto err_destroy_hwloc_topology;
+        goto err_free_os_provider;
     }
 
     ret = translate_params(in_params, os_provider);
@@ -692,8 +685,6 @@ static umf_result_t os_initialize(const void *params, void **provider) {
     free_bitmaps(os_provider);
 err_destroy_critnib:
     critnib_delete(os_provider->fd_offset_map);
-err_destroy_hwloc_topology:
-    hwloc_topology_destroy(os_provider->topo);
 err_free_os_provider:
     umf_ba_global_free(os_provider);
     return ret;
@@ -717,7 +708,7 @@ static umf_result_t os_finalize(void *provider) {
     if (os_provider->nodeset_str_buf) {
         umf_ba_global_free(os_provider->nodeset_str_buf);
     }
-    hwloc_topology_destroy(os_provider->topo);
+
     umf_ba_global_free(os_provider);
     return UMF_RESULT_SUCCESS;
 }

src/topology.c

@@ -1,6 +1,6 @@
 /*
  *
- * Copyright (C) 2024 Intel Corporation
+ * Copyright (C) 2024-2025 Intel Corporation
  *
  * Under the Apache License v2.0 with LLVM Exceptions. See LICENSE.TXT.
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
@@ -13,7 +13,9 @@
 #include "utils_log.h"
 
 static hwloc_topology_t topology = NULL;
+static hwloc_topology_t topology_reduced = NULL;
 static UTIL_ONCE_FLAG topology_initialized = UTIL_ONCE_FLAG_INIT;
+static UTIL_ONCE_FLAG topology_reduced_initialized = UTIL_ONCE_FLAG_INIT;
 
 void umfDestroyTopology(void) {
     if (topology) {
@@ -24,22 +26,55 @@ void umfDestroyTopology(void) {
         memcpy(&topology_initialized, &is_initialized,
                sizeof(topology_initialized));
     }
+    if (topology_reduced) {
+        hwloc_topology_destroy(topology_reduced);
+
+        // portable version of "topology_initialized = UTIL_ONCE_FLAG_INIT;"
+        static UTIL_ONCE_FLAG is_initialized = UTIL_ONCE_FLAG_INIT;
+        memcpy(&topology_reduced_initialized, &is_initialized,
+               sizeof(topology_reduced_initialized));
+    }
 }
 
-static void umfCreateTopology(void) {
-    if (hwloc_topology_init(&topology)) {
+static void umfCreateTopologyHelper(bool reduced,
+                                    hwloc_topology_t *topology_ptr) {
+    if (hwloc_topology_init(topology_ptr)) {
         LOG_ERR("Failed to initialize topology");
-        topology = NULL;
+        *topology_ptr = NULL;
         return;
     }
 
-    if (hwloc_topology_load(topology)) {
+    if (reduced) {
+        // Set the topology to only include NUMA nodes and memory
+        // to improve performance of the topology load on large systems
+        if (hwloc_topology_set_all_types_filter(*topology_ptr,
+                                                HWLOC_TYPE_FILTER_KEEP_NONE)) {
+            LOG_ERR("Failed to set topology filter");
+            hwloc_topology_destroy(*topology_ptr);
+            *topology_ptr = NULL;
+            return;
+        }
+    }
+    if (hwloc_topology_load(*topology_ptr)) {
         LOG_ERR("Failed to initialize topology");
-        hwloc_topology_destroy(topology);
-        topology = NULL;
+        hwloc_topology_destroy(*topology_ptr);
+        *topology_ptr = NULL;
     }
 }
 
+static void umfCreateTopology(void) {
+    umfCreateTopologyHelper(false, &topology);
+}
+
+static void umfCreateTopologyReduced(void) {
+    umfCreateTopologyHelper(true, &topology_reduced);
+}
+
+hwloc_topology_t umfGetTopologyReduced(void) {
+    utils_init_once(&topology_reduced_initialized, umfCreateTopologyReduced);
+    return topology_reduced;
+}
+
 hwloc_topology_t umfGetTopology(void) {
     utils_init_once(&topology_initialized, umfCreateTopology);
     return topology;

src/topology.h

@@ -1,6 +1,6 @@
 /*
  *
- * Copyright (C) 2024 Intel Corporation
+ * Copyright (C) 2024-2025 Intel Corporation
  *
  * Under the Apache License v2.0 with LLVM Exceptions. See LICENSE.TXT.
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
@@ -17,6 +17,7 @@ extern "C" {
 #endif
 
 hwloc_topology_t umfGetTopology(void);
+hwloc_topology_t umfGetTopologyReduced(void);
 void umfDestroyTopology(void);
 
 #ifdef __cplusplus