[UR][CUDA][HIP] Cleanup adapter code (#18370)

- Inline createProgram function in CUDA adapter, and catch ur result
exceptions in both adapters.
- Fix pattern of unnecessarily storing error code in a variable instead
of just returning it.
- Fix incorrect setErrorMessage uses, the adapter specific code signals
that there is a specific error code and message in get last error, it
doesn't make sense to use it as the code in the last error.
- Remove now unrelated comment on now removed primary context extension.
- Don't use `unique_ptr` when `delete` is enough.

Author: npmiller

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

@@ -36,13 +36,6 @@ ur_context_handle_t_::getOwningURPool(umf_memory_pool_t *UMFPool) {
 }
 
 /// Create a UR CUDA context.
-///
-/// By default creates a scoped context and keeps the last active CUDA context
-/// on top of the CUDA context stack.
-/// With the __SYCL_PI_CONTEXT_PROPERTIES_CUDA_PRIMARY key/id and a value of
-/// PI_TRUE creates a primary CUDA context and activates it on the CUDA context
-/// stack.
-///
 UR_APIEXPORT ur_result_t UR_APICALL
 urContextCreate(uint32_t DeviceCount, const ur_device_handle_t *phDevices,
                 const ur_context_properties_t * /*pProperties*/,
@@ -94,8 +87,7 @@ urContextRelease(ur_context_handle_t hContext) {
     return UR_RESULT_SUCCESS;
   }
   hContext->invokeExtendedDeleters();
-
-  std::unique_ptr<ur_context_handle_t_> Context{hContext};
+  delete hContext;
 
   return UR_RESULT_SUCCESS;
 }

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

@@ -1216,7 +1216,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGet(ur_platform_handle_t hPlatform,
                                                 uint32_t NumEntries,
                                                 ur_device_handle_t *phDevices,
                                                 uint32_t *pNumDevices) {
-  ur_result_t Result = UR_RESULT_SUCCESS;
   const bool AskingForAll = DeviceType == UR_DEVICE_TYPE_ALL;
   const bool AskingForDefault = DeviceType == UR_DEVICE_TYPE_DEFAULT;
   const bool AskingForGPU = DeviceType == UR_DEVICE_TYPE_GPU;
@@ -1234,13 +1233,13 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGet(ur_platform_handle_t hPlatform,
         phDevices[i] = hPlatform->Devices[i].get();
       }
     }
-
-    return Result;
   } catch (ur_result_t Err) {
     return Err;
   } catch (...) {
     return UR_RESULT_ERROR_OUT_OF_RESOURCES;
   }
+
+  return UR_RESULT_SUCCESS;
 }
 
 /// Gets the native CUDA handle of a UR device object