[OpenVINO Backend] support ops.slice_update

keras/src/backend/openvino/core.py

@@ -657,9 +657,83 @@ def slice(inputs, start_indices, shape):
 
 
 def slice_update(inputs, start_indices, updates):
-    raise NotImplementedError(
-        "`slice_update` is not supported with openvino backend"
+    inputs = get_ov_output(inputs)
+    if isinstance(start_indices, (list, np.ndarray)):
+        start_indices = tuple(start_indices)
+    assert isinstance(start_indices, tuple), (
+        "`slice_update` is not supported by openvino backend"
+        " for `start_indices` of type {}".format(type(start_indices))
     )
+    processed_start_indices = []
+    for idx in start_indices:
+        val = get_ov_output(idx)
+        val_type = val.get_element_type()
+        if not val_type.is_integral():
+            raise ValueError(
+                "`slice` is not supported by OpenVINO backend "
+                "for `start_indices` or `shape` with non-integer types"
+            )
+        if val_type != Type.i32:
+            val = ov_opset.convert(val, Type.i32).output(0)
+        if len(val.get_partial_shape()) == 0:
+            val = ov_opset.unsqueeze(
+                val, ov_opset.constant(0, Type.i32)
+            ).output(0)
+        processed_start_indices.append(val)
+    start_indices_tensor = ov_opset.concat(processed_start_indices, axis=0)
+
+    rank = len(updates.shape)
+    ranges = []
+    for dim in updates.shape:
+        r = ov_opset.range(
+            ov_opset.constant(0, Type.i32),
+            ov_opset.constant(dim, Type.i32),
+            ov_opset.constant(1, Type.i32),
+            output_type=Type.i32,
+        )
+        ranges.append(r)
+
+    broadcasted_ranges = []
+    for i, r in enumerate(ranges):
+        shape = [1] * rank
+        shape[i] = updates.shape[i]
+        r_reshaped = ov_opset.reshape(
+            r, ov_opset.constant(shape, Type.i32), special_zero=False
+        ).output(0)
+        target_shape = ov_opset.constant(list(updates.shape), Type.i32)
+        r_broadcasted = ov_opset.broadcast(r_reshaped, target_shape).output(0)
+        broadcasted_ranges.append(r_broadcasted)
+
+    indices_stack = ov_opset.concat(broadcasted_ranges, axis=0).output(0)
+
+    num_updates = 1
+    for dim in updates.shape:
+        num_updates *= dim
+    new_shape = ov_opset.constant([rank, num_updates], Type.i32)
+    indices_reshaped = ov_opset.reshape(
+        indices_stack, new_shape, special_zero=False
+    ).output(0)
+    absolute_indices = ov_opset.transpose(
+        indices_reshaped, ov_opset.constant([1, 0], Type.i32)
+    ).output(0)
+
+    start_indices_expanded = ov_opset.broadcast(
+        start_indices_tensor, ov_opset.constant([num_updates, rank], Type.i32)
+    ).output(0)
+    absolute_indices = ov_opset.add(
+        absolute_indices, start_indices_expanded
+    ).output(0)
+
+    updates_tensor = get_ov_output(updates)
+    updates_flat = ov_opset.reshape(
+        updates_tensor,
+        ov_opset.constant([num_updates], Type.i32),
+        special_zero=False,
+    ).output(0)
+    updated = ov_opset.scatter_nd_update(
+        inputs, absolute_indices, updates_flat
+    ).output(0)
+    return OpenVINOKerasTensor(updated)
 
 
 def while_loop(