set(C10_USE_GFLAGS ${USE_GFLAGS}) # used in cmake_macros.h.in
set(C10_USE_GLOG ${USE_GLOG}) # used in cmake_macros.h.in
set(C10_BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS}) # used in cmake_macros.h.in
+set(C10_USE_NUMA ${USE_NUMA})
+set(C10_DISABLE_NUMA ${CAFFE2_DISABLE_NUMA}) # used in cmake_macros.h.in
configure_file(
${CMAKE_CURRENT_LIST_DIR}/macros/cmake_macros.h.in
${CMAKE_BINARY_DIR}/c10/macros/cmake_macros.h)
target_link_libraries(c10 PUBLIC glog::glog)
endif()
+if (USE_NUMA)
+ if (NOT CAFFE2_DISABLE_NUMA)
+ message(STATUS "NUMA paths:")
+ message(STATUS ${Numa_INCLUDE_DIR})
+ message(STATUS ${Numa_LIBRARIES})
+ include_directories(SYSTEM ${Numa_INCLUDE_DIR})
+ target_link_libraries(c10 PRIVATE ${Numa_LIBRARIES})
+ else()
+ message(STATUS "NUMA is disabled")
+ endif()
+else()
+ message(STATUS "don't use NUMA")
+endif()
+
if (ANDROID)
target_link_libraries(c10 PRIVATE log)
endif()
#cmakedefine C10_BUILD_SHARED_LIBS
#cmakedefine C10_USE_GLOG
#cmakedefine C10_USE_GFLAGS
+#cmakedefine C10_DISABLE_NUMA
#endif // C10_MACROS_CMAKE_MACROS_H_
--- /dev/null
+#include "c10/util/numa.h"
+
+#if defined(__linux__) && !defined(C10_DISABLE_NUMA) && C10_MOBILE == 0
+#include <numa.h>
+#include <numaif.h>
+#include <unistd.h>
+#define C10_ENABLE_NUMA
+#endif
+
+namespace c10 {
+
+#ifdef C10_ENABLE_NUMA
+bool IsNUMAEnabled() {
+ return numa_available() >= 0;
+}
+
+void NUMABind(int numa_node_id) {
+ if (numa_node_id < 0) {
+ return;
+ }
+ if (!IsNUMAEnabled()) {
+ VLOG(1) << "NUMA is not enabled";
+ return;
+ }
+
+ AT_CHECK(
+ numa_node_id <= numa_max_node(),
+ "NUMA node id ", numa_node_id, " is unavailable");
+
+ auto bm = numa_allocate_nodemask();
+ numa_bitmask_setbit(bm, numa_node_id);
+ numa_bind(bm);
+ numa_bitmask_free(bm);
+}
+
+int GetNUMANode(const void* ptr) {
+ if (!IsNUMAEnabled()) {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+ }
+ AT_ASSERT(ptr);
+
+ int numa_node = -1;
+ AT_CHECK(
+ get_mempolicy(
+ &numa_node, NULL, 0, const_cast<void*>(ptr), MPOL_F_NODE | MPOL_F_ADDR) == 0,
+ "Unable to get memory policy, errno:", errno);
+ return numa_node;
+}
+
+int GetNumNUMANodes() {
+ if (!IsNUMAEnabled()) {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+ }
+
+ return numa_num_configured_nodes();
+}
+
+void NUMAMove(void* ptr, size_t size, int numa_node_id) {
+ if (numa_node_id < 0) {
+ return;
+ }
+ if (!IsNUMAEnabled()) {
+ VLOG(1) << "NUMA is not enabled";
+ return;
+ }
+ AT_ASSERT(ptr);
+
+ uintptr_t page_start_ptr = ((reinterpret_cast<uintptr_t>(ptr)) & ~(getpagesize() - 1));
+ ptrdiff_t offset = reinterpret_cast<uintptr_t>(ptr) - page_start_ptr;
+ // Avoid extra dynamic allocation and NUMA api calls
+ AT_ASSERT(numa_node_id >= 0 && static_cast<unsigned>(numa_node_id) < sizeof(unsigned long) * 8);
+ unsigned long mask = 1UL << numa_node_id;
+ AT_CHECK(
+ mbind(
+ reinterpret_cast<void*>(page_start_ptr),
+ size + offset,
+ MPOL_BIND,
+ &mask,
+ sizeof(mask) * 8,
+ MPOL_MF_MOVE | MPOL_MF_STRICT) == 0,
+ "Could not move memory to a NUMA node");
+}
+
+int GetCurrentNUMANode() {
+ if (!IsNUMAEnabled()) {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+ }
+
+ auto n = numa_node_of_cpu(sched_getcpu());
+ return n;
+}
+
+#else // C10_ENABLE_NUMA
+
+bool IsNUMAEnabled() {
+ return false;
+}
+
+void NUMABind(int numa_node_id) {
+ if (numa_node_id >= 0) {
+ VLOG(1) << "NUMA is not enabled";
+ }
+}
+
+int GetNUMANode(const void* ptr) {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+}
+
+int GetNumNUMANodes() {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+}
+
+void NUMAMove(void* ptr, size_t size, int numa_node_id) {
+ if (numa_node_id >= 0) {
+ VLOG(1) << "NUMA is not enabled";
+ }
+}
+
+int GetCurrentNUMANode() {
+ VLOG(1) << "NUMA is not enabled";
+ return -1;
+}
+
+#endif // C10_NUMA_ENABLED
+
+} // namespace c10
--- /dev/null
+#pragma once
+
+#include <c10/util/Logging.h>
+#include <c10/util/Optional.h>
+
+namespace c10 {
+
+/**
+ * Check whether NUMA is enabled
+ */
+C10_API bool IsNUMAEnabled();
+
+/**
+ * Bind to a given NUMA node
+ */
+C10_API void NUMABind(int numa_node_id);
+
+/**
+ * Get the NUMA id for a given pointer `ptr`
+ */
+C10_API int GetNUMANode(const void* ptr);
+
+/**
+ * Get number of NUMA nodes
+ */
+C10_API int GetNumNUMANodes();
+
+/**
+ * Move the memory pointed to by `ptr` of a given size to another NUMA node
+ */
+C10_API void NUMAMove(void* ptr, size_t size, int numa_node_id);
+
+/**
+ * Get the current NUMA node id
+ */
+C10_API int GetCurrentNUMANode();
+
+} // namespace c10
#cmakedefine CAFFE2_USE_MKLDNN
#cmakedefine CAFFE2_USE_NVTX
#cmakedefine CAFFE2_USE_TRT
-#cmakedefine CAFFE2_DISABLE_NUMA
#ifndef USE_NUMPY
#cmakedefine USE_NUMPY
{"USE_MKLDNN", "${CAFFE2_USE_MKLDNN}"}, \
{"USE_NVTX", "${CAFFE2_USE_NVTX}"}, \
{"USE_TRT", "${CAFFE2_USE_TRT}"}, \
- {"DISABLE_NUMA", "${CAFFE2_DISABLE_NUMA}"}, \
+ {"DISABLE_NUMA", "${CAFFE2_DISABLE_NUMA}"}, \
}
#include "caffe2/core/numa.h"
-
-C10_DEFINE_bool(caffe2_cpu_numa_enabled, false, "Use NUMA whenever possible.");
-
-#if defined(__linux__) && !defined(CAFFE2_DISABLE_NUMA) && CAFFE2_MOBILE == 0
-#include <numa.h>
-#include <numaif.h>
-#define CAFFE2_NUMA_ENABLED
-#endif
-
-namespace caffe2 {
-
-#ifdef CAFFE2_NUMA_ENABLED
-bool IsNUMAEnabled() {
- return FLAGS_caffe2_cpu_numa_enabled && numa_available() >= 0;
-}
-
-void NUMABind(int numa_node_id) {
- if (numa_node_id < 0) {
- return;
- }
- if (!IsNUMAEnabled()) {
- VLOG(1) << "NUMA is not enabled";
- return;
- }
-
- CAFFE_ENFORCE(
- numa_node_id <= numa_max_node(),
- "NUMA node id " + c10::to_string(numa_node_id) + " is unavailable");
-
- auto bm = numa_allocate_nodemask();
- numa_bitmask_clearall(bm);
- numa_bitmask_setbit(bm, numa_node_id);
- numa_bind(bm);
- numa_bitmask_free(bm);
-}
-
-int GetNUMANode(const void* ptr) {
- if (!IsNUMAEnabled()) {
- VLOG(1) << "NUMA is not enabled";
- return -1;
- }
- CAFFE_ENFORCE(ptr);
-
- int numa_node = -1;
- CAFFE_ENFORCE(
- get_mempolicy(
- &numa_node, NULL, 0, (void*)ptr, MPOL_F_NODE | MPOL_F_ADDR) == 0,
- "Unable to get memory policy");
- return numa_node;
-}
-
-int GetNumNUMANodes() {
- if (!IsNUMAEnabled()) {
- VLOG(1) << "NUMA is not enabled";
- return -1;
- }
-
- return numa_num_configured_nodes();
-}
-
-void NUMAMove(void* ptr, size_t size, int numa_node_id) {
- if (numa_node_id < 0) {
- return;
- }
- if (!IsNUMAEnabled()) {
- VLOG(1) << "NUMA is not enabled";
- return;
- }
- CAFFE_ENFORCE(ptr);
-
- size_t page_start_ptr = (((size_t)ptr) & ~(getpagesize() - 1));
- size_t offset = ((size_t)ptr) - page_start_ptr;
- // Avoid extra dynamic allocation and NUMA api calls
- CAFFE_ENFORCE(numa_node_id >= 0 && (unsigned)numa_node_id < sizeof(unsigned long) * 8);
- unsigned long mask = 1UL << numa_node_id;
- CAFFE_ENFORCE(
- mbind(
- (void*)page_start_ptr,
- size + offset,
- MPOL_BIND,
- &mask,
- sizeof(mask) * 8,
- MPOL_MF_MOVE | MPOL_MF_STRICT) == 0,
- "Could not move memory to a NUMA node");
-}
-
-int GetCurrentNUMANode() {
- if (!IsNUMAEnabled()) {
- VLOG(1) << "NUMA is not enabled";
- return -1;
- }
-
- return numa_node_of_cpu(sched_getcpu());
-}
-
-#else // CAFFE2_NUMA_ENABLED
-
-bool IsNUMAEnabled() {
- return false;
-}
-
-void NUMABind(int numa_node_id) {
- if (numa_node_id >= 0) {
- VLOG(1) << "NUMA is not enabled";
- }
-}
-
-int GetNUMANode(const void* ptr) {
- VLOG(1) << "NUMA is not enabled";
- return -1;
-}
-
-int GetNumNUMANodes() {
- VLOG(1) << "NUMA is not enabled";
- return -1;
-}
-
-void NUMAMove(void* ptr, size_t size, int numa_node_id) {
- if (numa_node_id >= 0) {
- VLOG(1) << "NUMA is not enabled";
- }
-}
-
-int GetCurrentNUMANode() {
- VLOG(1) << "NUMA is not enabled";
- return -1;
-}
-
-#endif // CAFFE2_NUMA_ENABLED
-
-} // namespace caffe2
-#ifndef CAFFE2_CORE_NUMA_H_
-#define CAFFE2_CORE_NUMA_H_
-
-#include "caffe2/core/logging.h"
-
-C10_DECLARE_bool(caffe2_cpu_numa_enabled);
-
-namespace caffe2 {
-
-CAFFE2_API bool IsNUMAEnabled();
-
-CAFFE2_API void NUMABind(int numa_node_id);
-
-CAFFE2_API int GetNUMANode(const void* ptr);
-
-CAFFE2_API int GetNumNUMANodes();
-
-CAFFE2_API void NUMAMove(void* ptr, size_t size, int numa_node_id);
-
-CAFFE2_API int GetCurrentNUMANode();
-
-} // namespace caffe2
-
-#endif // CAFFE2_CORE_NUMA_H_
+#pragma once
+#include "c10/util/numa.h"
+#include "caffe2/core/common.h"
if __name__ == '__main__':
- core.GlobalInit(["caffe2", "--caffe2_cpu_numa_enabled=1"])
main()
from caffe2.python.test_util import TestCase
import unittest
-core.GlobalInit(["caffe2", "--caffe2_cpu_numa_enabled=1"])
-
-
def build_test_net(net_name):
net = core.Net(net_name)
net.Proto().type = "async_scheduling"