Dandandan commented on code in PR #8721:
URL: https://github.com/apache/arrow-datafusion/pull/8721#discussion_r1441586235
##########
datafusion/physical-expr/src/aggregate/count_distinct.rs:
##########
@@ -83,10 +106,57 @@ impl AggregateExpr for DistinctCount {
}
fn create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
- Ok(Box::new(DistinctCountAccumulator {
- values: HashSet::default(),
- state_data_type: self.state_data_type.clone(),
- }))
+ use DataType::*;
+ use TimeUnit::*;
+
+ match &self.state_data_type {
+ Int8 => native_distinct_count_accumulator!(Int8Type),
+ Int16 => native_distinct_count_accumulator!(Int16Type),
+ Int32 => native_distinct_count_accumulator!(Int32Type),
+ Int64 => native_distinct_count_accumulator!(Int64Type),
+ UInt8 => native_distinct_count_accumulator!(UInt8Type),
+ UInt16 => native_distinct_count_accumulator!(UInt16Type),
+ UInt32 => native_distinct_count_accumulator!(UInt32Type),
+ UInt64 => native_distinct_count_accumulator!(UInt64Type),
+ Decimal128(_, _) =>
native_distinct_count_accumulator!(Decimal128Type),
+ Decimal256(_, _) =>
native_distinct_count_accumulator!(Decimal256Type),
+
+ Date32 => native_distinct_count_accumulator!(Date32Type),
+ Date64 => native_distinct_count_accumulator!(Date64Type),
+ Time32(Millisecond) => {
+ native_distinct_count_accumulator!(Time32MillisecondType)
+ }
+ Time32(Second) => {
+ native_distinct_count_accumulator!(Time32SecondType)
+ }
+ Time64(Microsecond) => {
+ native_distinct_count_accumulator!(Time64MicrosecondType)
+ }
+ Time64(Nanosecond) => {
+ native_distinct_count_accumulator!(Time64NanosecondType)
+ }
+ Timestamp(Microsecond, _) => {
+ native_distinct_count_accumulator!(TimestampMicrosecondType)
+ }
+ Timestamp(Millisecond, _) => {
+ native_distinct_count_accumulator!(TimestampMillisecondType)
+ }
+ Timestamp(Nanosecond, _) => {
+ native_distinct_count_accumulator!(TimestampNanosecondType)
+ }
+ Timestamp(Second, _) => {
+ native_distinct_count_accumulator!(TimestampSecondType)
+ }
+
+ Float16 => float_distinct_count_accumulator!(Float16Type),
+ Float32 => float_distinct_count_accumulator!(Float32Type),
+ Float64 => float_distinct_count_accumulator!(Float64Type),
Review Comment:
Not for now, but if we would like to do it for strings / bytes, we could do
use a datastructure like this to get maximal performance:
```rust
/// Contains hashes + first offset, could be HashMap<u64, SmallVec<u64>> as
well, or use `RawTable` for extra speed
uniques: HashMap<u64, u64>,
/// optional: contains unique offsets to look for
next: Vec<usize>,
/// actual string/byte data, can be emitted cheaply / free
values: BufferBuilder<u8>,
```
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
