Move numa.{h, cc} to c10/util (#15024)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15024

Pull Request resolved: https://github.com/pytorch/pytorch/pull/14393

att

Reviewed By: dzhulgakov

Differential Revision: D13380559

fbshipit-source-id: abc3fc7321cf37323f756dfd614c7b41978734e4

4 files changed, 200 insertions, 0 deletions

diff --git a/c10/CMakeLists.txt b/c10/CMakeLists.txt
index 8e5ddc4042..e80dd795c2 100644
--- a/c10/CMakeLists.txt
+++ b/c10/CMakeLists.txt
@@ -16,6 +16,8 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 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
     ${CMAKE_BINARY_DIR}/c10/macros/cmake_macros.h)
@@ -50,6 +52,20 @@ if (${USE_GLOG})
     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()
diff --git a/c10/macros/cmake_macros.h.in b/c10/macros/cmake_macros.h.in
index 29f8705fa1..5b20360af5 100644
--- a/c10/macros/cmake_macros.h.in
+++ b/c10/macros/cmake_macros.h.in
@@ -7,5 +7,6 @@
 #cmakedefine C10_BUILD_SHARED_LIBS
 #cmakedefine C10_USE_GLOG
 #cmakedefine C10_USE_GFLAGS
+#cmakedefine C10_DISABLE_NUMA
 
 #endif // C10_MACROS_CMAKE_MACROS_H_
diff --git a/c10/util/numa.cpp b/c10/util/numa.cpp
new file mode 100644
index 0000000000..4a5086ec7a
--- /dev/null
+++ b/c10/util/numa.cpp
@@ -0,0 +1,143 @@
+#include "c10/util/numa.h"
+
+C10_DEFINE_bool(caffe2_cpu_numa_enabled, false, "Use NUMA whenever possible.");
+
+#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 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;
+  }
+
+  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
diff --git a/c10/util/numa.h b/c10/util/numa.h
new file mode 100644
index 0000000000..aa5ae52332
--- /dev/null
+++ b/c10/util/numa.h
@@ -0,0 +1,40 @@
+#pragma once
+
+#include <c10/util/Logging.h>
+#include <c10/util/Optional.h>
+
+C10_DECLARE_bool(caffe2_cpu_numa_enabled);
+
+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