[UR][Graph] in-order USM Memcpy regression CTS test (#18241)

Create a UR CTS test based on the E2E test
[Graph/RecordReplay/usm_copy_in_order.cpp](https://github.com/intel/llvm/blob/sycl/sycl/test-e2e/Graph/RecordReplay/usm_copy_in_order.cpp)
to help debug #18169

Disabled for Level-Zero v1 adapter (passes on v2) as [test fails
with](https://github.com/intel/llvm/actions/runs/14733668954/job/41355110585?pr=18241):
```
[ RUN      ] urCommandBufferUSMCopyInOrderTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero__Intel_R__Data_Center_GPU_Max_1100_ID0ID____________________
/home/test-user/actions-runner/sycl-ur-01/_work/llvm/llvm/unified-runtime/test/conformance/exp_command_buffer/regression/usm_copy.cpp:140: Failure
Expected equality of these values:
  result3
    Which is: 3656
  output[i]
    Which is: 170
```
This fails both with an without driver in-order command-lists used,
indicating a separate problem in the command-buffer L0 v1 adapter code
that needs investigated and debugger to re-enable this test.

Author: EwanC

unified-runtime/test/conformance/exp_command_buffer/CMakeLists.txt

@@ -21,6 +21,7 @@ add_conformance_test_with_kernels_environment(exp_command_buffer
   update/event_sync.cpp
   update/kernel_event_sync.cpp
   update/local_memory_update.cpp
+  regression/usm_copy.cpp
 )
 
 add_subdirectory(native-command)

unified-runtime/test/conformance/exp_command_buffer/regression/usm_copy.cpp

@@ -0,0 +1,145 @@
+// 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
+
+#include "../fixtures.h"
+
+// UR reproducer for SYCL-Graph E2E test "RecordReplay/usm_copy_in_order.cpp"
+// Note that the kernel code is different, in that this test uses the
+// saxpy_usm kernel, but the sequence of operations is the same.
+struct urCommandBufferUSMCopyInOrderTest
+    : uur::command_buffer::urCommandBufferExpExecutionTest {
+  virtual void SetUp() override {
+    program_name = "saxpy_usm";
+    UUR_RETURN_ON_FATAL_FAILURE(urCommandBufferExpExecutionTest::SetUp());
+
+    // See URLZA-521
+    UUR_KNOWN_FAILURE_ON(uur::LevelZero{});
+
+    // Create in-order command-buffer
+    ur_exp_command_buffer_desc_t desc{
+        UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_DESC, // stype
+        nullptr,                                   // pNext
+        false,                                     // isUpdatable
+        true,                                      // isInOrder
+        false                                      // enableProfiling
+    };
+    ASSERT_SUCCESS(
+        urCommandBufferCreateExp(context, device, &desc, &in_order_cmd_buf));
+    ASSERT_NE(in_order_cmd_buf, nullptr);
+
+    // Create 4 device USM allocations and initialize elements to list index
+    for (unsigned i = 0; i < device_ptrs.size(); i++) {
+      auto &device_ptr = device_ptrs[i];
+      ASSERT_SUCCESS(urUSMDeviceAlloc(context, device, nullptr, nullptr,
+                                      allocation_size, &device_ptr));
+      ASSERT_NE(device_ptr, nullptr);
+
+      uint32_t pattern = i;
+      ASSERT_SUCCESS(urEnqueueUSMFill(queue, device_ptr, sizeof(pattern),
+                                      &pattern, allocation_size, 0, nullptr,
+                                      nullptr));
+    }
+    ASSERT_SUCCESS(urQueueFinish(queue));
+
+    // Index 0 is output
+    ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 0, nullptr, device_ptrs[0]));
+    // Index 1 is A
+    ASSERT_SUCCESS(urKernelSetArgValue(kernel, 1, sizeof(A), nullptr, &A));
+    // Index 2 is X
+    ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 2, nullptr, device_ptrs[1]));
+    // Index 3 is Y
+    ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 3, nullptr, device_ptrs[2]));
+  }
+
+  virtual void TearDown() override {
+    for (auto &device_ptr : device_ptrs) {
+      if (device_ptr) {
+        EXPECT_SUCCESS(urUSMFree(context, device_ptr));
+      }
+    }
+    if (in_order_cmd_buf) {
+      EXPECT_SUCCESS(urCommandBufferReleaseExp(in_order_cmd_buf));
+    }
+
+    UUR_RETURN_ON_FATAL_FAILURE(urCommandBufferExpExecutionTest::TearDown());
+  }
+
+  ur_exp_command_buffer_handle_t in_order_cmd_buf = nullptr;
+  static constexpr size_t global_size = 10;
+  static constexpr size_t global_offset = 0;
+  static constexpr size_t n_dimensions = 1;
+  static constexpr size_t allocation_size = sizeof(uint32_t) * global_size;
+  static constexpr uint32_t A = 42;
+  std::array<void *, 4> device_ptrs = {nullptr, nullptr, nullptr, nullptr};
+};
+
+UUR_INSTANTIATE_DEVICE_TEST_SUITE(urCommandBufferUSMCopyInOrderTest);
+TEST_P(urCommandBufferUSMCopyInOrderTest, Success) {
+  // Do an eager kernel enqueue without wait on completion
+  // D[0] = A * D[1] + D[2]
+  // D[0] = 42 * 1 + 2
+  // D[0] = 44
+  ASSERT_SUCCESS(urEnqueueKernelLaunch(queue, kernel, n_dimensions,
+                                       &global_offset, &global_size, nullptr, 0,
+                                       nullptr, nullptr));
+
+  // command-buffer sync point used to enforce linear dependencies when
+  // appending commands to the command-buffer.
+  ur_exp_command_buffer_sync_point_t sync_point;
+
+  // Add SAXPY kernel node to command-buffer
+  // D[3] = A * D[1] + D[0]
+  // D[3] = 42 * 1 + 44
+  // D[3] = 86
+  ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 3, nullptr, device_ptrs[0]));
+  ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 0, nullptr, device_ptrs[3]));
+  ASSERT_SUCCESS(urCommandBufferAppendKernelLaunchExp(
+      in_order_cmd_buf, kernel, n_dimensions, &global_offset, &global_size,
+      nullptr, 0, nullptr, 0, nullptr, 0, nullptr, &sync_point, nullptr,
+      nullptr));
+
+  // Add device-to-device memcpy node from output of previous command to
+  // the X component of the expression.
+  // D[1] = 86
+  ASSERT_SUCCESS(urCommandBufferAppendUSMMemcpyExp(
+      in_order_cmd_buf, device_ptrs[1], device_ptrs[3], allocation_size, 0,
+      nullptr, 0, nullptr, &sync_point, nullptr, nullptr));
+
+  // Add SAXPY kernel node
+  // D[3] = A * [1] + [0]
+  // D[3] = 42 * 86 + 44
+  // D[3] = 3656
+  ASSERT_SUCCESS(urCommandBufferAppendKernelLaunchExp(
+      in_order_cmd_buf, kernel, n_dimensions, &global_offset, &global_size,
+      nullptr, 0, nullptr, 1, &sync_point, 0, nullptr, &sync_point, nullptr,
+      nullptr));
+
+  // Add device-to-device memcpy node from output of previous command to
+  // currently unused USM allocation.
+  // D[2] = 3656
+  ASSERT_SUCCESS(urCommandBufferAppendUSMMemcpyExp(
+      in_order_cmd_buf, device_ptrs[2], device_ptrs[3], allocation_size, 1,
+      &sync_point, 0, nullptr, &sync_point, nullptr, nullptr));
+
+  // Add device-to-host memcpy node
+  std::vector<uint32_t> output(global_size);
+  ASSERT_SUCCESS(urCommandBufferAppendUSMMemcpyExp(
+      in_order_cmd_buf, output.data(), device_ptrs[2], allocation_size, 1,
+      &sync_point, 0, nullptr, &sync_point, nullptr, nullptr));
+  ASSERT_SUCCESS(urCommandBufferFinalizeExp(in_order_cmd_buf));
+
+  ASSERT_SUCCESS(
+      urEnqueueCommandBufferExp(queue, in_order_cmd_buf, 0, nullptr, nullptr));
+  ASSERT_SUCCESS(urQueueFinish(queue));
+
+  // Verify
+  constexpr uint32_t result1 = A * 1 + 2;             // eager kernel submission
+  constexpr uint32_t result2 = A * 1 + result1;       // first kernel command
+  constexpr uint32_t result3 = A * result2 + result1; // second kernel command
+  for (size_t i = 0; i < global_size; i++) {
+    ASSERT_EQ(result3, output[i]);
+  }
+}