Implement efficient, asynchronous fill method

Leverages dpctl's strided fill kernel or and zeros kernel

Author: ndgrigorian

dpnp/dpnp_algo/dpnp_fill.py

@@ -0,0 +1,84 @@
+# -*- coding: utf-8 -*-
+# *****************************************************************************
+# Copyright (c) 2016-2024, Intel Corporation
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+# - Redistributions of source code must retain the above copyright notice,
+#   this list of conditions and the following disclaimer.
+# - Redistributions in binary form must reproduce the above copyright notice,
+#   this list of conditions and the following disclaimer in the documentation
+#   and/or other materials provided with the distribution.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+# THE POSSIBILITY OF SUCH DAMAGE.
+# *****************************************************************************
+
+import dpctl.tensor as dpt
+import numpy as np
+from dpctl.tensor._tensor_impl import (
+    _copy_usm_ndarray_into_usm_ndarray,
+    _full_usm_ndarray,
+    _zeros_usm_ndarray,
+)
+from dpctl.utils import SequentialOrderManager
+
+import dpnp
+
+
+def dpnp_fill(arr, val):
+    dpnp.check_supported_arrays_type(arr)
+    arr = dpnp.get_usm_ndarray(arr)
+    exec_q = arr.sycl_queue
+
+    dpnp.check_supported_arrays_type(val, scalar_type=True, all_scalars=True)
+    # if val is an array, process it
+    if isinstance(val, (dpnp.dpnp_array, dpt.usm_ndarray)):
+        val = dpnp.get_usm_ndarray(val)
+        if val.shape != ():
+            raise ValueError("`val` must be a scalar")
+        # asarray moves scalar to the correct device
+        # and casts to the expected dtype
+        a_val = dpt.asarray(
+            val,
+            dtype=arr.dtype,
+            usm_type=arr.usm_type,
+            sycl_queue=exec_q,
+        )
+        a_val = dpt.broadcast_to(a_val, arr.shape)
+        _manager = SequentialOrderManager[exec_q]
+        dep_evs = _manager.submitted_events
+        h_ev, c_ev = _copy_usm_ndarray_into_usm_ndarray(
+            src=a_val, dst=arr, sycl_queue=exec_q, depends=dep_evs
+        )
+        _manager.add_event_pair(h_ev, c_ev)
+        return
+
+    dt = arr.dtype
+    val_type = type(val)
+    if val_type in [float, complex] and dpnp.issubdtype(dt, dpnp.integer):
+        val = int(val.real)
+    elif val_type is complex and dpnp.issubdtype(dt, dpnp.floating):
+        val = val.real
+    elif val_type is int and dpnp.issubdtype(dt, dpnp.integer):
+        val = np.asarray(val, dtype=dt)[()]
+
+    _manager = SequentialOrderManager[exec_q]
+    dep_evs = _manager.submitted_events
+    # can leverage efficient memset when val is 0
+    if arr.flags["FORC"] and val == 0:
+        h_ev, zeros_ev = _zeros_usm_ndarray(arr, exec_q)
+        _manager.add_event_pair(h_ev, zeros_ev)
+    else:
+        h_ev, fill_ev = _full_usm_ndarray(val, arr, exec_q)
+        _manager.add_event_pair(h_ev, fill_ev)

dpnp/dpnp_array.py

@@ -903,8 +903,10 @@ def fill(self, value):
 
         """
 
-        for i in range(self.size):
-            self.flat[i] = value
+        # lazy import avoids circular imports
+        from .dpnp_algo.dpnp_fill import dpnp_fill
+
+        dpnp_fill(self, value)
 
     @property
     def flags(self):