[SYCL][Graph] Add support for CUDA backend (#11133)

Adds support for enabling CUDA backend with Sycl-Graph:

- Enables Graph e2e tests with CUDA backend.

Companion PR with CUDA UR adapter changes is here:
oneapi-src/unified-runtime#932

---------

Co-authored-by: Ewan Crawford <[email protected]>
Co-authored-by: Julian Miller <[email protected]>
Co-authored-by: Ben Tracy <[email protected]>
Co-authored-by: Kenneth Benzie (Benie) <[email protected]>
Co-authored-by: Maxime France-Pillois <[email protected]>

Author: 5 people

sycl/doc/design/CommandGraph.md

@@ -54,13 +54,13 @@ if(SYCL_PI_UR_USE_FETCH_CONTENT)
   include(FetchContent)
 
   set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
-  # commit 036b9cfcd8fb4be9394449b5406e6402580a63c9
-  # Merge: 2ab0734 86f96f0
+
+  # commit be53fb3ba5bf1ac33051456d72c61b6a01d94a72
   # Author: Kenneth Benzie (Benie) <[email protected]>
-  # Date:   Fri Oct 27 16:36:23 2023 +0100
-  #     Merge pull request #961 from hdelan/change-unions-to-stdvariant
-  #     [HIP][CUDA] Change unions in ur_mem_handle_t_ to stdvariant
-  set(UNIFIED_RUNTIME_TAG 036b9cfcd8fb4be9394449b5406e6402580a63c9)
+  # Date:   Tue Oct 31 13:22:52 2023 +0000
+  # - Merge pull request #932 from Bensuo/cuda-cmd-buffers
+  # - [CUDA][EXP] CUDA adapter support for command buffers
+  set(UNIFIED_RUNTIME_TAG be53fb3ba5bf1ac33051456d72c61b6a01d94a72)
 
   if(SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO)
     set(UNIFIED_RUNTIME_REPO "${SYCL_PI_UR_OVERRIDE_FETCH_CONTENT_REPO}")

sycl/doc/design/images/SYCL-Graph-Architecture.svg

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/plugins/unified_runtime/CMakeLists.txt

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/add_node_while_recording.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/add_nodes_after_finalize.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/assume_buffer_outlives_graph_property.cpp

@@ -1,5 +1,5 @@
-// REQUIRES: level_zero, gpu
-// RUN: %{build_pthread_inc} -o %t.out
+// REQUIRES: cuda || level_zero, gpu
+// RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}

sycl/test-e2e/Graph/Explicit/basic_buffer.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/basic_usm.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/basic_usm_host.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/basic_usm_mixed.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/basic_usm_shared.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/basic_usm_system.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_2d.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_host2target.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_host2target_2d.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_host2target_offset.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_offsets.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_target2host.cpp

@@ -5,6 +5,9 @@
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out 2>&1 | FileCheck %s %}
 //
 // CHECK-NOT: LEAK
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/buffer_copy_target2host_2d.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 

sycl/test-e2e/Graph/Explicit/buffer_copy_target2host_offset.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/cycle_error.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/depends_on.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/dotp_buffer_reduction.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/dotp_usm_reduction.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/double_buffer.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/empty_node.cpp

@@ -3,6 +3,9 @@
 // RUN: %{run} %t.out 2>&1 | FileCheck %s
 //
 // CHECK: complete
+//
+// TODO enable cuda once buffer issue investigated and fixed
+// UNSUPPORTED: cuda
 
 #define GRAPH_E2E_EXPLICIT
 

sycl/test-e2e/Graph/Explicit/enqueue_ordering.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/event_status_querying.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/executable_graph_update.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/executable_graph_update_ordering.cpp

@@ -1,6 +1,7 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
+// RUN: %if ext_oneapi_cuda %{ %{run} %t.out %}
+// RUN: %if ext_oneapi_level_zero %{env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s %}
 
 // Checks the PI call trace to ensure that the bundle kernel of the single task
 // is used.

sycl/test-e2e/Graph/Explicit/host_task.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/kernel_bundle.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/multiple_exec_graphs.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/multiple_kernel_bundles.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/node_ordering.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/queue_shortcuts.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/repeated_exec.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out %GPU_CHECK_PLACEHOLDER
 // RUN: %if ext_oneapi_level_zero %{env ZE_DEBUG=4 %{run} %t.out %GPU_CHECK_PLACEHOLDER 2>&1 | FileCheck %s %}

sycl/test-e2e/Graph/Explicit/single_node.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/stream.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph_execute_without_parent.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph_multiple_submission.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph_nested.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph_reduction.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/sub_graph_two_parent_graphs.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/temp_buffer_reinterpret.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/usm_copy.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG

sycl/test-e2e/Graph/Explicit/usm_fill.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero, gpu
+// REQUIRES: cuda || level_zero, gpu
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using ZE_DEBUG