[UR][L0] Add support for passing device list to urProgramBuild/Link/Compile (#11464)

piProgramBuild receives a list of devices, while urProgramBuild does
not. This produces a series of issues when a UR program needs to be
created for a specific device.

So define a new API, called urProgramBuildExp to pass this list.

Requires related patch in Unified Runtime Adapters here:
oneapi-src/unified-runtime#934

---------

Signed-off-by: Spruit, Neil R <[email protected]>
Co-authored-by: Kenneth Benzie (Benie) <[email protected]>

Author: nrspruit

sycl/plugins/unified_runtime/CMakeLists.txt

@@ -56,13 +56,13 @@ if(SYCL_PI_UR_USE_FETCH_CONTENT)
   include(FetchContent)
 
   set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
-  # commit 04799e735a7ddd0839568b25bf03fa8f77160b7d
-  # Merge: 534071e5 4cf02cb9
+  # commit 71957e849435169b43e51a42025afeeb63d7508b
+  # Merge: ce152a64 e001b98d
   # Author: Kenneth Benzie (Benie) <[email protected]>
-  # Date:   Mon Nov 20 10:24:50 2023 +0000
-  #     Merge pull request #1048 from callumfare/opencl_fix_urMemBufferCreate_leak
-  #     [OpenCL] Fix memory leak and coverity issue with struct-to-array casts
-  set(UNIFIED_RUNTIME_TAG 04799e735a7ddd0839568b25bf03fa8f77160b7d)
+  # Date:   Tue Nov 21 16:41:08 2023 +0000
+  #     Merge pull request #1100 from kbenzie/benie/fixup-multi-device-compile
+  #     [CUDA][HIP][OpenCL][NATIVECPU] Fix multi-device compile
+  set(UNIFIED_RUNTIME_TAG 71957e849435169b43e51a42025afeeb63d7508b)
 
   if(SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO)
     set(UNIFIED_RUNTIME_REPO "${SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO}")

sycl/plugins/unified_runtime/pi2ur.hpp

@@ -1985,10 +1985,17 @@ piProgramLink(pi_context Context, pi_uint32 NumDevices,
   ur_program_handle_t *UrProgram =
       reinterpret_cast<ur_program_handle_t *>(RetProgram);
 
-  HANDLE_ERRORS(urProgramLink(UrContext, NumInputPrograms, UrInputPrograms,
-                              Options, UrProgram));
+  auto UrDevices = reinterpret_cast<ur_device_handle_t *>(
+      const_cast<pi_device *>(DeviceList));
 
-  return PI_SUCCESS;
+  auto urResult =
+      urProgramLinkExp(UrContext, NumDevices, UrDevices, NumInputPrograms,
+                       UrInputPrograms, Options, UrProgram);
+  if (urResult == UR_RESULT_ERROR_UNSUPPORTED_FEATURE) {
+    urResult = urProgramLink(UrContext, NumInputPrograms, UrInputPrograms,
+                             Options, UrProgram);
+  }
+  return ur2piResult(urResult);
 }
 
 inline pi_result piProgramCompile(
@@ -2017,9 +2024,15 @@ inline pi_result piProgramCompile(
   HANDLE_ERRORS(urProgramGetInfo(UrProgram, PropName, sizeof(&UrContext),
                                  &UrContext, nullptr));
 
-  HANDLE_ERRORS(urProgramCompile(UrContext, UrProgram, Options));
+  auto UrDevices = reinterpret_cast<ur_device_handle_t *>(
+      const_cast<pi_device *>(DeviceList));
 
-  return PI_SUCCESS;
+  auto urResult =
+      urProgramCompileExp(UrProgram, NumDevices, UrDevices, Options);
+  if (urResult == UR_RESULT_ERROR_UNSUPPORTED_FEATURE) {
+    urResult = urProgramCompile(UrContext, UrProgram, Options);
+  }
+  return ur2piResult(urResult);
 }
 
 inline pi_result
@@ -2050,9 +2063,14 @@ piProgramBuild(pi_program Program, pi_uint32 NumDevices,
   HANDLE_ERRORS(urProgramGetInfo(UrProgram, PropName, sizeof(&UrContext),
                                  &UrContext, nullptr));
 
-  HANDLE_ERRORS(urProgramBuild(UrContext, UrProgram, Options));
+  auto UrDevices = reinterpret_cast<ur_device_handle_t *>(
+      const_cast<pi_device *>(DeviceList));
 
-  return PI_SUCCESS;
+  auto urResult = urProgramBuildExp(UrProgram, NumDevices, UrDevices, Options);
+  if (urResult == UR_RESULT_ERROR_UNSUPPORTED_FEATURE) {
+    urResult = urProgramBuild(UrContext, UrProgram, Options);
+  }
+  return ur2piResult(urResult);
 }
 
 inline pi_result piextProgramSetSpecializationConstant(pi_program Program,