[GitHub] [arrow-rs] tustvold commented on a diff in pull request #4406: Faster unpacking of Int32Type dictionary

["tustvold (via GitHub)" <gi...@apache.org>]

tustvold commented on code in PR #4406:
URL: https://github.com/apache/arrow-rs/pull/4406#discussion_r1228184372


##########
arrow-cast/src/cast.rs:
##########
@@ -3466,34 +3466,21 @@ fn unpack_dictionary<K>(
 where
     K: ArrowDictionaryKeyType,
 {
-    let dict_array = array
-        .as_any()
-        .downcast_ref::<DictionaryArray<K>>()
-        .ok_or_else(|| {
-            ArrowError::ComputeError(
-                "Internal Error: Cannot cast dictionary to DictionaryArray of expected type".to_string(),
-            )
-        })?;
-
-    // attempt to cast the dict values to the target type
-    // use the take kernel to expand out the dictionary
+    let dict_array = array.as_dictionary::<K>();
     let cast_dict_values = cast_with_options(dict_array.values(), to_type, cast_options)?;
-
-    // Note take requires first casting the indices to u32
-    let keys_array: ArrayRef =
-        Arc::new(PrimitiveArray::<K>::from(dict_array.keys().to_data()));
-    let indices = cast_with_options(&keys_array, &DataType::UInt32, cast_options)?;
-    let u32_indices =
-        indices
-            .as_any()
-            .downcast_ref::<UInt32Array>()
-            .ok_or_else(|| {
-                ArrowError::ComputeError(
-                    "Internal Error: Cannot cast dict indices to UInt32".to_string(),
-                )
-            })?;
-
-    take(cast_dict_values.as_ref(), u32_indices, None)
+    let keys = dict_array.keys();
+    match K::DATA_TYPE {
+        DataType::Int32 => {

Review Comment:
   Theoretically we could pass the keys directly to take, this approach avoids additional codegen, and also appears to be faster for reasons I don't understand :sweat_smile: 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org