[UR][SYCL] Add urUSMContextMemcpyExp API to enable device global support.

unified-runtime/scripts/core/EXP-USM-CONTEXT-MEMCPY.rst

@@ -0,0 +1,63 @@
+<%
+    OneApi=tags['$OneApi']
+    x=tags['$x']
+    X=x.upper()
+%>
+
+.. _experimental-usm-context-memcpy:
+
+================================================================================
+USM Context Memcpy
+================================================================================
+
+.. 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
+--------------------------------------------------------------------------------
+
+In order to support device globals there's a need for a blocking USM write
+operation that doesn't need a queue. This is to facilitate fast initialization
+of the device global memory via native APIs that enable this kind of operation.
+
+API
+--------------------------------------------------------------------------------
+
+Enums
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* ${x}_device_info_t
+    * ${X}_DEVICE_INFO_USM_CONTEXT_MEMCPY_SUPPORT_EXP
+
+Functions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+* ${x}USMContextMemcpyExp
+
+Changelog
+--------------------------------------------------------------------------------
+
++-----------+---------------------------+
+| Revision  | Changes                   |
++===========+===========================+
+| 1.0       | Initial Draft             |
++-----------+---------------------------+
+
+
+Support
+--------------------------------------------------------------------------------
+
+Adapters which support this experimental feature *must* return true for the new
+``${X}_DEVICE_INFO_USM_CONTEXT_MEMCPY_SUPPORT_EXP`` device info query.
+
+
+Contributors
+--------------------------------------------------------------------------------
+
+* Aaron Greig `[email protected] <[email protected]>`

unified-runtime/scripts/core/exp-usm-context-memcpy.yml

@@ -0,0 +1,48 @@
+#
+# Copyright (C) 2025 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 USM Context Memcpy"
+ordinal: "99"
+--- #--------------------------------------------------------------------------
+type: enum
+extend: true
+typed_etors: true
+desc: "Extension enums to $x_device_info_t to support $xUSMContextMemcpy"
+name: $x_device_info_t
+etors:
+    - name: USM_CONTEXT_MEMCPY_SUPPORT_EXP
+      value: "0x7000"
+      desc: "[$x_bool_t] returns true if the device supports $xUSMContextMemcpyExp"
+--- #--------------------------------------------------------------------------
+type: function
+desc: "Perform a synchronous, blocking memcpy operation between USM allocations."
+class: $xUSM
+name: ContextMemcpyExp
+ordinal: "0"
+params:
+    - type: $x_context_handle_t
+      name: hContext
+      desc: "[in] Context associated with the device(s) that own the allocations `pSrc` and `pDst`."
+    - type: void*
+      name: pDst
+      desc: "[in] Destination pointer to copy to."
+    - type: const void*
+      name: pSrc
+      desc: "[in] Source pointer to copy from."
+    - type: size_t
+      name: size
+      desc: "[in] Size in bytes to be copied."
+returns:
+    - $X_RESULT_SUCCESS
+    - $X_RESULT_ERROR_ADAPTER_SPECIFIC
+    - $X_RESULT_ERROR_INVALID_SIZE:
+        - "`size == 0`"
+        - "If `size` is higher than the allocation size of `pSrc` or `pDst`"

unified-runtime/scripts/core/registry.yml

@@ -637,6 +637,9 @@ etors:
 - name: USM_POOL_GET_INFO_EXP
   desc: Enumerator for $xUSMPoolGetInfoExp
   value: '262'
+- name: USM_CONTEXT_MEMCPY_EXP
+  desc: Enumerator for $xUSMContextMemcpyExp
+  value: '264'
 ---
 type: enum
 desc: Defines structure types

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

@@ -1113,6 +1113,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   }
   case UR_DEVICE_INFO_LOW_POWER_EVENTS_EXP:
     return ReturnValue(false);
+  case UR_DEVICE_INFO_USM_CONTEXT_MEMCPY_SUPPORT_EXP:
+    return ReturnValue(false);
   case UR_DEVICE_INFO_USM_P2P_SUPPORT_EXP:
     return ReturnValue(true);
   case UR_DEVICE_INFO_LAUNCH_PROPERTIES_SUPPORT_EXP:

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

@@ -368,6 +368,7 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetUSMExpProcAddrTable(
   pDdiTable->pfnPoolSetDevicePoolExp = urUSMPoolSetDevicePoolExp;
   pDdiTable->pfnPoolGetDevicePoolExp = urUSMPoolGetDevicePoolExp;
   pDdiTable->pfnPoolTrimToExp = urUSMPoolTrimToExp;
+  pDdiTable->pfnContextMemcpyExp = urUSMContextMemcpyExp;
   return UR_RESULT_SUCCESS;
 }
 

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

@@ -459,3 +459,9 @@ UR_APIEXPORT ur_result_t UR_APICALL urUSMPoolTrimToExp(ur_context_handle_t,
                                                        size_t) {
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
+
+UR_APIEXPORT ur_result_t UR_APICALL urUSMContextMemcpyExp(ur_context_handle_t,
+                                                          void *, const void *,
+                                                          size_t) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}

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

@@ -1077,9 +1077,10 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   }
   case UR_DEVICE_INFO_COMMAND_BUFFER_EVENT_SUPPORT_EXP:
     return ReturnValue(false);
-  case UR_DEVICE_INFO_LOW_POWER_EVENTS_EXP: {
+  case UR_DEVICE_INFO_LOW_POWER_EVENTS_EXP:
+    return ReturnValue(false);
+  case UR_DEVICE_INFO_USM_CONTEXT_MEMCPY_SUPPORT_EXP:
     return ReturnValue(false);
-  }
   case UR_DEVICE_INFO_USM_P2P_SUPPORT_EXP:
     return ReturnValue(true);
   case UR_DEVICE_INFO_LAUNCH_PROPERTIES_SUPPORT_EXP:

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

@@ -366,6 +366,7 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetUSMExpProcAddrTable(
   pDdiTable->pfnPoolSetDevicePoolExp = urUSMPoolSetDevicePoolExp;
   pDdiTable->pfnPoolGetDevicePoolExp = urUSMPoolGetDevicePoolExp;
   pDdiTable->pfnPoolTrimToExp = urUSMPoolTrimToExp;
+  pDdiTable->pfnContextMemcpyExp = urUSMContextMemcpyExp;
   return UR_RESULT_SUCCESS;
 }