Merge pull request #2254 from PietroGhg/pietro/events_rr

[NATIVECPU] Implement events on Native CPU

source/adapters/native_cpu/common.hpp

@@ -12,6 +12,7 @@
 
 #include "logger/ur_logger.hpp"
 #include "ur/ur.hpp"
+#include <chrono>
 
 constexpr size_t MaxMessageSize = 256;
 
@@ -70,3 +71,31 @@ template <typename T> inline void decrementOrDelete(T *refC) {
   if (refC->decrementReferenceCount() == 0)
     delete refC;
 }
+
+inline uint64_t get_timestamp() {
+  return std::chrono::duration_cast<std::chrono::nanoseconds>(
+             std::chrono::high_resolution_clock::now().time_since_epoch())
+      .count();
+}
+
+namespace native_cpu {
+
+inline void *aligned_malloc(size_t alignment, size_t size) {
+  void *ptr = nullptr;
+#ifdef _MSC_VER
+  ptr = _aligned_malloc(size, alignment);
+#else
+  ptr = std::aligned_alloc(alignment, size);
+#endif
+  return ptr;
+}
+
+inline void aligned_free(void *ptr) {
+#ifdef _MSC_VER
+  _aligned_free(ptr);
+#else
+  free(ptr);
+#endif
+}
+
+} // namespace native_cpu

source/adapters/native_cpu/context.hpp

@@ -64,17 +64,10 @@ static size_t get_padding(uint32_t alignment) {
 // allocation so that the pointer returned to the user
 // always satisfies (ptr % align) == 0.
 static inline void *malloc_impl(uint32_t alignment, size_t size) {
-  void *ptr = nullptr;
   assert(alignment >= alignof(usm_alloc_info) &&
          "memory not aligned to usm_alloc_info");
-#ifdef _MSC_VER
-  ptr = _aligned_malloc(alloc_header_size + get_padding(alignment) + size,
-                        alignment);
-
-#else
-  ptr = std::aligned_alloc(alignment,
-                           alloc_header_size + get_padding(alignment) + size);
-#endif
+  void *ptr = native_cpu::aligned_malloc(
+      alignment, alloc_header_size + get_padding(alignment) + size);
   return ptr;
 }
 
@@ -100,11 +93,8 @@ struct ur_context_handle_t_ : RefCounted {
     const native_cpu::usm_alloc_info &info = native_cpu::get_alloc_info(ptr);
     UR_ASSERT(info.type != UR_USM_TYPE_UNKNOWN,
               UR_RESULT_ERROR_INVALID_MEM_OBJECT);
-#ifdef _MSC_VER
-    _aligned_free(info.base_alloc_ptr);
-#else
-    free(info.base_alloc_ptr);
-#endif
+
+    native_cpu::aligned_free(info.base_alloc_ptr);
     allocations.erase(ptr);
     return UR_RESULT_SUCCESS;
   }

source/adapters/native_cpu/device.cpp

@@ -10,6 +10,7 @@
 
 #include <ur_api.h>
 
+#include "common.hpp"
 #include "platform.hpp"
 
 #if defined(_MSC_VER) || defined(__MINGW32__) || defined(__MINGW64__)
@@ -247,7 +248,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(uint32_t{4});
   case UR_DEVICE_INFO_NATIVE_VECTOR_WIDTH_HALF:
     return ReturnValue(uint32_t{16});
-  // Imported from level_zero
   case UR_DEVICE_INFO_USM_HOST_SUPPORT:
   case UR_DEVICE_INFO_USM_DEVICE_SUPPORT:
   case UR_DEVICE_INFO_USM_SINGLE_SHARED_SUPPORT:
@@ -472,19 +472,12 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceCreateWithNativeHandle(
 UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetGlobalTimestamps(
     ur_device_handle_t hDevice, uint64_t *pDeviceTimestamp,
     uint64_t *pHostTimestamp) {
-  std::ignore = hDevice; // todo
+  std::ignore = hDevice;
   if (pHostTimestamp) {
-    using namespace std::chrono;
-    *pHostTimestamp =
-        duration_cast<nanoseconds>(steady_clock::now().time_since_epoch())
-            .count();
+    *pHostTimestamp = get_timestamp();
   }
   if (pDeviceTimestamp) {
-    // todo: calculate elapsed time properly
-    using namespace std::chrono;
-    *pDeviceTimestamp =
-        duration_cast<nanoseconds>(steady_clock::now().time_since_epoch())
-            .count();
+    *pDeviceTimestamp = get_timestamp();
   }
   return UR_RESULT_SUCCESS;
 }