Unified Runtime v0.8.3 changes

[kbenzie]

This branch is based upon #12192 with
additional bug fixes.

• [UR][HIP] Fix include for AMD COMGR oneapi-src/unified-runtime#1222
• [SYCL][Graph] Add new 'are_graphs_supported' helper function #11877
• [UR] call urGetAdapter before creating a platform from native handle #12432
• [L0][UR] Only Override max allocation limits given env #12375
• [L0] fix waiting on non-owned native handle events #12425
• [L0] do not ignore returned values from zeHostSynchronize #12427
• Set UR tag to v0.8.3

[github-actions]

⚠️ Python code formatter, darker found issues in your code. ⚠️

You can test this locally with the following command:

darker --check --diff -r 6a0692cd91b3f1f4c1797a2644422e5f7c1c2c62..dcfe7f8048895c5fae66df10679b375200060d39 sycl/test-e2e/format.py sycl/test-e2e/lit.cfg.py

View the diff from darker here.

--- format.py	2023-12-18 08:57:41.000000 +0000
+++ format.py	2024-01-22 15:00:41.637048 +0000
@@ -162,15 +162,21 @@
             # current llvm-lit invocation isn't configured to include it. We
             # don't use ONEAPI_DEVICE_SELECTOR for `%{run-unfiltered-devices}`
             # so that device might still be accessible to some of the tests yet
             # we won't set the environment variable below for such scenario.
             extra_env = []
-            if 'ext_oneapi_level_zero:gpu' in sycl_devices and litConfig.params.get('ur_l0_debug'):
-                extra_env.append('UR_L0_DEBUG={}'.format(test.config.ur_l0_debug))
-
-            if 'ext_oneapi_level_zero:gpu' in sycl_devices and litConfig.params.get('ur_l0_leaks_debug'):
-                extra_env.append('UR_L0_LEAKS_DEBUG={}'.format(test.config.ur_l0_leaks_debug))
+            if "ext_oneapi_level_zero:gpu" in sycl_devices and litConfig.params.get(
+                "ur_l0_debug"
+            ):
+                extra_env.append("UR_L0_DEBUG={}".format(test.config.ur_l0_debug))
+
+            if "ext_oneapi_level_zero:gpu" in sycl_devices and litConfig.params.get(
+                "ur_l0_leaks_debug"
+            ):
+                extra_env.append(
+                    "UR_L0_LEAKS_DEBUG={}".format(test.config.ur_l0_leaks_debug)
+                )
 
             if 'ext_oneapi_cuda:gpu' in sycl_devices:
                 extra_env.append('SYCL_PI_CUDA_ENABLE_IMAGE_SUPPORT=1')
 
             return extra_env
--- lit.cfg.py	2023-12-18 08:57:41.000000 +0000
+++ lit.cfg.py	2024-01-22 15:00:41.849029 +0000
@@ -133,19 +133,19 @@
     config.available_features.add('matrix-fp16') # XMX implies the support of FP16 matrix
 
 if lit_config.params.get('matrix-fp16', False):
     config.available_features.add('matrix-fp16')
 
-#support for LIT parameter ur_l0_debug<num>
-if lit_config.params.get('ur_l0_debug'):
-    config.ur_l0_debug = lit_config.params.get('ur_l0_debug')
-    lit_config.note("UR_L0_DEBUG: "+config.ur_l0_debug)
-
-#support for LIT parameter ur_l0_leaks_debug
-if lit_config.params.get('ur_l0_leaks_debug'):
-    config.ur_l0_leaks_debug = lit_config.params.get('ur_l0_leaks_debug')
-    lit_config.note("UR_L0_LEAKS_DEBUG: "+config.ur_l0_leaks_debug)
+# support for LIT parameter ur_l0_debug<num>
+if lit_config.params.get("ur_l0_debug"):
+    config.ur_l0_debug = lit_config.params.get("ur_l0_debug")
+    lit_config.note("UR_L0_DEBUG: " + config.ur_l0_debug)
+
+# support for LIT parameter ur_l0_leaks_debug
+if lit_config.params.get("ur_l0_leaks_debug"):
+    config.ur_l0_leaks_debug = lit_config.params.get("ur_l0_leaks_debug")
+    lit_config.note("UR_L0_LEAKS_DEBUG: " + config.ur_l0_leaks_debug)
 
 # Make sure that any dynamic checks below are done in the build directory and
 # not where the sources are located. This is important for the in-tree
 # configuration (as opposite to the standalone one).
 os.chdir(config.sycl_obj_root)

[JackAKirk]

Hi @steffenlarsen,

The release branch is failing with the following tests (I think on all backends, but definitely cuda and hip)

SYCL :: Basic/built-ins/vec_common.cpp
SYCL :: Basic/built-ins/vec_geometric.cpp
SYCL :: Basic/built-ins/vec_math.cpp
SYCL :: Basic/built-ins/vec_relational.cpp

These tests can be fixed by using the latest version of these tests in sycl tip, which could be achieved by cherry picking the following into the release branch (via this PR):

#11774
#11919
#11929

However, in particular since #11774 also includes runtime changes, I wanted to check with you whether you think these PRs should be cherry picked?

Thanks

[steffenlarsen]

@JackAKirk - All the mentioned patches have been picked. I have not heard of any failures in those since that, so it may be CUDA/HIP specific.

[JackAKirk]

@JackAKirk - All the mentioned patches have been picked. I have not heard of any failures in those since that, so it may be CUDA/HIP specific.

OK that's fine then thanks. They should make their way to sycl-rel-5_1_0 eventually I guess.

[kbenzie]

This CMake error is I believe due to changes in HIP 6 but this branch doesn't support that HIP version.