[UR] Add device-wide synchronization extension

unified-runtime/scripts/core/EXP-DEVICE-WAIT.rst

@@ -0,0 +1,64 @@
+<%
+    OneApi=tags['$OneApi']
+    x=tags['$x']
+    X=x.upper()
+%>
+
+.. _experimental-device-wait:
+
+================================================================================
+Device Wait
+================================================================================
+
+.. warning::
+
+    Experimental features:
+
+    *   May be replaced, updated, or removed at any time.
+    *   Do not require maintaining API/ABI stability of their own additions over
+        time.
+    *   Do not require conformance testing of their own additions.
+
+
+
+Motivation
+--------------------------------------------------------------------------------
+
+This extension adds the ability to do device-wide synchronization, instead of at
+queue or event level.
+
+API
+--------------------------------------------------------------------------------
+
+Enums
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+* ${x}_device_info_t
+    * ${X}_DEVICE_INFO_DEVICE_WAIT_SUPPORT_EXP
+
+Functions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* ${x}DeviceWaitExp
+
+Changelog
+--------------------------------------------------------------------------------
+
++-----------+---------------------------------------------+
+| Revision  | Changes                                     |
++===========+=============================================+
+| 1.0       | Initial Draft                               |
++-----------+---------------------------------------------+
+
+Support
+--------------------------------------------------------------------------------
+
+Adapters which support this experimental feature *must* return ``true`` when
+queried for ${X}_DEVICE_INFO_DEVICE_WAIT_SUPPORT_EXP via
+${x}DeviceGetInfo. Conversely, before using any of the functionality defined
+in this experimental feature the user *must* use the device query to determine
+if the adapter supports this feature.
+
+Contributors
+--------------------------------------------------------------------------------
+
+* Steffen Larsen `steffen.larsen@intel.com <steffen.larsen@intel.com>`_

unified-runtime/scripts/core/exp-device-wait.yml

@@ -0,0 +1,38 @@
+#
+# Copyright (C) 2023 Intel Corporation
+#
+# Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
+# See LICENSE.TXT
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+#
+# See YaML.md for syntax definition
+#
+--- #--------------------------------------------------------------------------
+type: header
+desc: "Intel $OneApi Unified Runtime Experimental APIs for device-wide synchronization"
+ordinal: "99"
+--- #--------------------------------------------------------------------------
+type: enum
+extend: true
+typed_etors: true
+desc: "Extension enums for $x_device_info_t to support device-wide synchronization."
+name: $x_device_info_t
+etors:
+    - name: DEVICE_WAIT_SUPPORT_EXP
+      value: "0x6002"
+      desc: "[$x_bool_t] Returns true if the device supports the device-wide synchronization experimental feature."
+--- #--------------------------------------------------------------------------
+type: function
+desc: "Synchronizes with all queues on the device."
+class: $xDevice
+name: WaitExp
+decl: static
+ordinal: "1"
+params:
+    - type: $x_device_handle_t
+      name: hDevice
+      desc: "[in] handle of the device instance."
+returns:
+    - $X_RESULT_ERROR_INVALID_DEVICE
+    - $X_RESULT_ERROR_INVALID_NULL_HANDLE:
+      - "`hDevice == nullptr`"

unified-runtime/scripts/core/registry.yml

@@ -682,7 +682,10 @@ etors:
 - name: IPC_CLOSE_MEM_HANDLE_EXP
   desc: Enumerator for $xIPCCloseMemHandleExp
   value: '292'
-max_id: '292'
+- name: DEVICE_WAIT_EXP
+  desc: Enumerator for $xDeviceWaitExp
+  value: '293'
+max_id: '293'
 ---
 type: enum
 desc: Defines structure types

unified-runtime/source/adapters/cuda/device.cpp

@@ -1178,6 +1178,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(true);
   case UR_DEVICE_INFO_MULTI_DEVICE_COMPILE_SUPPORT_EXP:
     return ReturnValue(false);
+  case UR_DEVICE_INFO_DEVICE_WAIT_SUPPORT_EXP:
+    return ReturnValue(true);
   case UR_DEVICE_INFO_ASYNC_USM_ALLOCATIONS_SUPPORT_EXP:
     return ReturnValue(true);
   case UR_DEVICE_INFO_KERNEL_LAUNCH_CAPABILITIES: {
@@ -1402,3 +1404,11 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceSelectBinary(
   // No image can be loaded for the given device
   return UR_RESULT_ERROR_INVALID_BINARY;
 }
+
+/// Synchronizes with all queues on the device.
+UR_APIEXPORT ur_result_t UR_APICALL
+urDeviceWaitExp(ur_device_handle_t hDevice) {
+  ScopedContext Active(hDevice);
+  UR_CHECK_ERROR(cuCtxSynchronize());
+  return UR_RESULT_SUCCESS;
+}

unified-runtime/source/adapters/cuda/ur_interface_loader.cpp

@@ -270,6 +270,16 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetDeviceProcAddrTable(
   return UR_RESULT_SUCCESS;
 }
 
+UR_DLLEXPORT ur_result_t UR_APICALL urGetDeviceExpProcAddrTable(
+    ur_api_version_t version, ur_device_exp_dditable_t *pDdiTable) {
+  auto result = validateProcInputs(version, pDdiTable);
+  if (UR_RESULT_SUCCESS != result) {
+    return result;
+  }
+  pDdiTable->pfnWaitExp = urDeviceWaitExp;
+  return UR_RESULT_SUCCESS;
+}
+
 UR_DLLEXPORT ur_result_t UR_APICALL urGetCommandBufferExpProcAddrTable(
     /// [in] API version requested
     ur_api_version_t version,
@@ -510,6 +520,7 @@ UR_DLLEXPORT ur_result_t UR_APICALL urAllAddrTable(ur_api_version_t version,
   urGetUsmP2PExpProcAddrTable(version, &pDdiTable->UsmP2PExp);
   urGetVirtualMemProcAddrTable(version, &pDdiTable->VirtualMem);
   urGetDeviceProcAddrTable(version, &pDdiTable->Device);
+  urGetDeviceExpProcAddrTable(version, &pDdiTable->DeviceExp);
   urGetMemoryExportExpProcAddrTable(version, &pDdiTable->MemoryExportExp);
 
   return UR_RESULT_SUCCESS;

unified-runtime/source/adapters/hip/device.cpp

@@ -1043,6 +1043,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(true);
   case UR_DEVICE_INFO_MULTI_DEVICE_COMPILE_SUPPORT_EXP:
     return ReturnValue(false);
+  case UR_DEVICE_INFO_DEVICE_WAIT_SUPPORT_EXP:
+    return ReturnValue(false);
   case UR_DEVICE_INFO_KERNEL_LAUNCH_CAPABILITIES:
     return ReturnValue(0);
   case UR_DEVICE_INFO_MEMORY_EXPORT_EXPORTABLE_DEVICE_MEM_EXP:
@@ -1241,3 +1243,7 @@ ur_result_t UR_APICALL urDeviceGetGlobalTimestamps(ur_device_handle_t hDevice,
   }
   return UR_RESULT_SUCCESS;
 }
+
+ur_result_t UR_APICALL urDeviceWaitExp(ur_device_handle_t) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}