Cycles: Remove usage of libnumaapi

No need for it now since all the threading queries and
scheduling is done via TBB.

Should be no functional changes as all the removed code
is supposed to be unused.

intern/cycles/cmake/macros.cmake

@@ -168,12 +168,6 @@ macro(cycles_target_link_libraries target)
     target_link_libraries(${target} extern_hipew)
   endif()
 
-  if(CYCLES_STANDALONE_REPOSITORY)
-    target_link_libraries(${target} extern_numaapi)
-  else()
-    target_link_libraries(${target} bf_intern_numaapi)
-  endif()
-
   if(UNIX AND NOT APPLE)
     if(CYCLES_STANDALONE_REPOSITORY)
       target_link_libraries(${target} extern_libc_compat)

intern/cycles/util/CMakeLists.txt

@@ -53,16 +53,6 @@ if(WITH_CYCLES_STANDALONE)
   endif()
 endif()
 
-if(CYCLES_STANDALONE_REPOSITORY)
-  list(APPEND INC_SYS
-    ../../third_party/numaapi/include
-  )
-else()
-  list(APPEND INC_SYS
-    ../../numaapi/include
-  )
-endif()
-
 set(SRC_HEADERS
   algorithm.h
   aligned_malloc.h

intern/cycles/util/system.cpp

@@ -20,9 +20,8 @@
 #include "util/string.h"
 #include "util/types.h"
 
-#include <numaapi.h>
-
 #include <OpenImageIO/sysutil.h>
+
 OIIO_NAMESPACE_USING
 
 #ifdef _WIN32
@@ -41,83 +40,6 @@ OIIO_NAMESPACE_USING
 
 CCL_NAMESPACE_BEGIN
 
-bool system_cpu_ensure_initialized()
-{
-  static bool is_initialized = false;
-  static bool result = false;
-  if (is_initialized) {
-    return result;
-  }
-  is_initialized = true;
-  const NUMAAPI_Result numa_result = numaAPI_Initialize();
-  result = (numa_result == NUMAAPI_SUCCESS);
-  return result;
-}
-
-/* Fallback solution, which doesn't use NUMA/CPU groups. */
-static int system_cpu_thread_count_fallback()
-{
-#ifdef _WIN32
-  SYSTEM_INFO info;
-  GetSystemInfo(&info);
-  return info.dwNumberOfProcessors;
-#elif defined(__APPLE__)
-  int count;
-  size_t len = sizeof(count);
-  int mib[2] = {CTL_HW, HW_NCPU};
-  sysctl(mib, 2, &count, &len, NULL, 0);
-  return count;
-#else
-  return sysconf(_SC_NPROCESSORS_ONLN);
-#endif
-}
-
-int system_cpu_thread_count()
-{
-  const int num_nodes = system_cpu_num_numa_nodes();
-  int num_threads = 0;
-  for (int node = 0; node < num_nodes; ++node) {
-    if (!system_cpu_is_numa_node_available(node)) {
-      continue;
-    }
-    num_threads += system_cpu_num_numa_node_processors(node);
-  }
-  return num_threads;
-}
-
-int system_cpu_num_numa_nodes()
-{
-  if (!system_cpu_ensure_initialized()) {
-    /* Fallback to a single node with all the threads. */
-    return 1;
-  }
-  return numaAPI_GetNumNodes();
-}
-
-bool system_cpu_is_numa_node_available(int node)
-{
-  if (!system_cpu_ensure_initialized()) {
-    return true;
-  }
-  return numaAPI_IsNodeAvailable(node);
-}
-
-int system_cpu_num_numa_node_processors(int node)
-{
-  if (!system_cpu_ensure_initialized()) {
-    return system_cpu_thread_count_fallback();
-  }
-  return numaAPI_GetNumNodeProcessors(node);
-}
-
-bool system_cpu_run_thread_on_node(int node)
-{
-  if (!system_cpu_ensure_initialized()) {
-    return true;
-  }
-  return numaAPI_RunThreadOnNode(node);
-}
-
 int system_console_width()
 {
   int columns = 0;
@@ -137,14 +59,6 @@ int system_console_width()
   return (columns > 0) ? columns : 80;
 }
 
-int system_cpu_num_active_group_processors()
-{
-  if (!system_cpu_ensure_initialized()) {
-    return system_cpu_thread_count_fallback();
-  }
-  return numaAPI_GetNumCurrentNodesProcessors();
-}
-
 /* Equivalent of Windows __cpuid for x86 processors on other platforms. */
 #if (!defined(_WIN32) || defined(FREE_WINDOWS)) && (defined(__x86_64__) || defined(__i386__))
 static void __cpuid(int data[4], int selector)

intern/cycles/util/system.h

@@ -22,36 +22,9 @@
 
 CCL_NAMESPACE_BEGIN
 
-/* Make sure CPU groups / NUMA API is initialized. */
-bool system_cpu_ensure_initialized();
-
-/* Get total number of threads in all NUMA nodes / CPU groups. */
-int system_cpu_thread_count();
-
 /* Get width in characters of the current console output. */
 int system_console_width();
 
-/* Get number of available nodes.
- *
- * This is in fact an index of last node plus one and it's not guaranteed
- * that all nodes up to this one are available. */
-int system_cpu_num_numa_nodes();
-
-/* Returns truth if the given node is available for compute. */
-bool system_cpu_is_numa_node_available(int node);
-
-/* Get number of available processors on a given node. */
-int system_cpu_num_numa_node_processors(int node);
-
-/* Runs the current thread and its children on a specific node.
- *
- * Returns truth if affinity has successfully changed. */
-bool system_cpu_run_thread_on_node(int node);
-
-/* Number of processors within the current CPU group (or within active thread
- * thread affinity). */
-int system_cpu_num_active_group_processors();
-
 string system_cpu_brand_string();
 int system_cpu_bits();
 bool system_cpu_support_sse2();

intern/cycles/util/thread.cpp

@@ -21,7 +21,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-thread::thread(function<void()> run_cb, int node) : run_cb_(run_cb), joined_(false), node_(node)
+thread::thread(function<void()> run_cb) : run_cb_(run_cb), joined_(false)
 {
 #ifdef __APPLE__
   /* Set the stack size to 2MB to match Linux. The default 512KB on macOS is
@@ -46,9 +46,6 @@ thread::~thread()
 void *thread::run(void *arg)
 {
   thread *self = (thread *)(arg);
-  if (self->node_ != -1) {
-    system_cpu_run_thread_on_node(self->node_);
-  }
   self->run_cb_();
   return NULL;
 }

intern/cycles/util/thread.h

@@ -46,9 +46,7 @@ typedef std::condition_variable thread_condition_variable;
 
 class thread {
  public:
-  /* NOTE: Node index of -1 means that affinity will be inherited from the
-   * parent thread and no override on top of that will happen. */
-  thread(function<void()> run_cb, int node = -1);
+  thread(function<void()> run_cb);
   ~thread();
 
   static void *run(void *arg);
@@ -62,7 +60,6 @@ class thread {
   std::thread std_thread;
 #endif
   bool joined_;
-  int node_;
 };
 
 using thread_spin_lock = tbb::spin_mutex;