FIX: update functional support fallback logic for a DPNP/DPCTL ndarray inputs (uxlfoundation#2113)

* FIX: update functional support fallback logic a little bit

* host numpy copies of the inputs data will be used for the fallback cases, since stock scikit-learn doesn't support DPCTL usm_ndarray and DPNP ndarray

* Added a clarifying comment

* Enhanced patch message for data transfer

Author: samir-nasibli

onedal/_device_offload.py

@@ -133,7 +133,7 @@ def _transfer_to_host(queue, *data):
             raise RuntimeError("Input data shall be located on single target device")
 
         host_data.append(item)
-    return queue, host_data
+    return has_usm_data, queue, host_data
 
 
 def _get_global_queue():
@@ -150,8 +150,8 @@ def _get_global_queue():
 
 def _get_host_inputs(*args, **kwargs):
     q = _get_global_queue()
-    q, hostargs = _transfer_to_host(q, *args)
-    q, hostvalues = _transfer_to_host(q, *kwargs.values())
+    _, q, hostargs = _transfer_to_host(q, *args)
+    _, q, hostvalues = _transfer_to_host(q, *kwargs.values())
     hostkwargs = dict(zip(kwargs.keys(), hostvalues))
     return q, hostargs, hostkwargs
 

sklearnex/_device_offload.py

@@ -63,25 +63,34 @@ def _get_backend(obj, queue, method_name, *data):
 
 def dispatch(obj, method_name, branches, *args, **kwargs):
     q = _get_global_queue()
-    q, hostargs = _transfer_to_host(q, *args)
-    q, hostvalues = _transfer_to_host(q, *kwargs.values())
+    has_usm_data_for_args, q, hostargs = _transfer_to_host(q, *args)
+    has_usm_data_for_kwargs, q, hostvalues = _transfer_to_host(q, *kwargs.values())
     hostkwargs = dict(zip(kwargs.keys(), hostvalues))
 
     backend, q, patching_status = _get_backend(obj, q, method_name, *hostargs)
-
+    has_usm_data = has_usm_data_for_args or has_usm_data_for_kwargs
     if backend == "onedal":
-        patching_status.write_log(queue=q)
+        # Host args only used before onedal backend call.
+        # Device will be offloaded when onedal backend will be called.
+        patching_status.write_log(queue=q, transferred_to_host=False)
         return branches[backend](obj, *hostargs, **hostkwargs, queue=q)
     if backend == "sklearn":
         if (
             "array_api_dispatch" in get_config()
             and get_config()["array_api_dispatch"]
             and "array_api_support" in obj._get_tags()
             and obj._get_tags()["array_api_support"]
+            and not has_usm_data
         ):
+            # USM ndarrays are also excluded for the fallback Array API. Currently, DPNP.ndarray is
+            # not compliant with the Array API standard, and DPCTL usm_ndarray Array API is compliant,
+            # except for the linalg module. There is no guarantee that stock scikit-learn will
+            # work with such input data. The condition will be updated after DPNP.ndarray and
+            # DPCTL usm_ndarray enabling for conformance testing and these arrays supportance
+            # of the fallback cases.
             # If `array_api_dispatch` enabled and array api is supported for the stock scikit-learn,
             # then raw inputs are used for the fallback.
-            patching_status.write_log()
+            patching_status.write_log(transferred_to_host=False)
             return branches[backend](obj, *args, **kwargs)
         else:
             patching_status.write_log()

sklearnex/_utils.py

@@ -29,10 +29,10 @@ class PatchingConditionsChain(daal4py_PatchingConditionsChain):
     def get_status(self):
         return self.patching_is_enabled
 
-    def write_log(self, queue=None):
+    def write_log(self, queue=None, transferred_to_host=True):
         if self.patching_is_enabled:
             self.logger.info(
-                f"{self.scope_name}: {get_patch_message('onedal', queue=queue)}"
+                f"{self.scope_name}: {get_patch_message('onedal', queue=queue, transferred_to_host=transferred_to_host)}"
             )
         else:
             self.logger.debug(
@@ -43,7 +43,9 @@ def write_log(self, queue=None):
                 self.logger.debug(
                     f"{self.scope_name}: patching failed with cause - {message}"
                 )
-            self.logger.info(f"{self.scope_name}: {get_patch_message('sklearn')}")
+            self.logger.info(
+                f"{self.scope_name}: {get_patch_message('sklearn', transferred_to_host=transferred_to_host)}"
+            )
 
 
 def set_sklearn_ex_verbose():
@@ -66,7 +68,7 @@ def set_sklearn_ex_verbose():
         )
 
 
-def get_patch_message(s, queue=None):
+def get_patch_message(s, queue=None, transferred_to_host=True):
     if s == "onedal":
         message = "running accelerated version on "
         if queue is not None:
@@ -87,6 +89,10 @@ def get_patch_message(s, queue=None):
             f"Invalid input - expected one of 'onedal','sklearn',"
             f" 'sklearn_after_onedal', got {s}"
         )
+    if transferred_to_host:
+        message += (
+            ". All input data transferred to host for further backend computations."
+        )
     return message