perf: Implement groups accumulator count distinct primitive types

datafusion/functions-aggregate-common/src/aggregate/count_distinct.rs

@@ -17,11 +17,13 @@
 
 mod bytes;
 mod dict;
+mod groups;
 mod native;
 
 pub use bytes::BytesDistinctCountAccumulator;
 pub use bytes::BytesViewDistinctCountAccumulator;
 pub use dict::DictionaryCountAccumulator;
+pub use groups::PrimitiveDistinctCountGroupsAccumulator;
 pub use native::Bitmap65536DistinctCountAccumulator;
 pub use native::Bitmap65536DistinctCountAccumulatorI16;
 pub use native::BoolArray256DistinctCountAccumulator;

datafusion/functions-aggregate-common/src/aggregate/count_distinct/groups.rs

@@ -0,0 +1,178 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+use arrow::array::{
+    ArrayRef, AsArray, BooleanArray, Int64Array, ListArray, PrimitiveArray,
+};
+use arrow::buffer::OffsetBuffer;
+use arrow::datatypes::{ArrowPrimitiveType, Field};
+use datafusion_common::HashSet;
+use datafusion_common::hash_utils::RandomState;
+use datafusion_expr_common::groups_accumulator::{EmitTo, GroupsAccumulator};
+use std::hash::Hash;
+use std::mem::size_of;
+use std::sync::Arc;
+
+use crate::aggregate::groups_accumulator::accumulate::accumulate;
+
+pub struct PrimitiveDistinctCountGroupsAccumulator<T: ArrowPrimitiveType>
+where
+    T::Native: Eq + Hash,
+{
+    seen: HashSet<(usize, T::Native), RandomState>,
+    counts: Vec<i64>,
+}
+
+impl<T: ArrowPrimitiveType> PrimitiveDistinctCountGroupsAccumulator<T>
+where
+    T::Native: Eq + Hash,
+{
+    pub fn new() -> Self {
+        Self {
+            seen: HashSet::default(),
+            counts: Vec::new(),
+        }
+    }
+}
+
+impl<T: ArrowPrimitiveType> Default for PrimitiveDistinctCountGroupsAccumulator<T>
+where
+    T::Native: Eq + Hash,
+{
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl<T: ArrowPrimitiveType + Send + std::fmt::Debug> GroupsAccumulator
+    for PrimitiveDistinctCountGroupsAccumulator<T>
+where
+    T::Native: Eq + Hash,
+{
+    fn update_batch(
+        &mut self,
+        values: &[ArrayRef],
+        group_indices: &[usize],
+        opt_filter: Option<&BooleanArray>,
+        total_num_groups: usize,
+    ) -> datafusion_common::Result<()> {
+        debug_assert_eq!(values.len(), 1);
+        self.counts.resize(total_num_groups, 0);
+        let arr = values[0].as_primitive::<T>();
+        accumulate(group_indices, arr, opt_filter, |group_idx, value| {
+            if self.seen.insert((group_idx, value)) {
+                self.counts[group_idx] += 1;
+            }
+        });
+        Ok(())
+    }
+
+    fn evaluate(&mut self, emit_to: EmitTo) -> datafusion_common::Result<ArrayRef> {
+        let counts = emit_to.take_needed(&mut self.counts);
+
+        match emit_to {
+            EmitTo::All => {
+                self.seen.clear();
+            }
+            EmitTo::First(n) => {
+                let mut remaining = HashSet::default();
+                for (group_idx, value) in self.seen.drain() {
+                    if group_idx >= n {
+                        remaining.insert((group_idx - n, value));
+                    }
+                }
+                self.seen = remaining;
+            }
+        }
+
+        Ok(Arc::new(Int64Array::from(counts)))
+    }
+
+    fn state(&mut self, emit_to: EmitTo) -> datafusion_common::Result<Vec<ArrayRef>> {
+        let num_emitted = match emit_to {
+            EmitTo::All => self.counts.len(),
+            EmitTo::First(n) => n,
+        };
+
+        let mut group_values: Vec<Vec<T::Native>> = vec![Vec::new(); num_emitted];
+
+        if matches!(emit_to, EmitTo::All) {
+            for (group_idx, value) in self.seen.drain() {
+                group_values[group_idx].push(value);
+            }
+            self.counts.clear();
+        } else {
+            let mut remaining = HashSet::default();
+            for (group_idx, value) in self.seen.drain() {
+                if group_idx < num_emitted {
+                    group_values[group_idx].push(value);
+                } else {
+                    remaining.insert((group_idx - num_emitted, value));
+                }
+            }
+            self.seen = remaining;
+            let _ = emit_to.take_needed(&mut self.counts);
+        }
+
+        let mut offsets = vec![0i32];
+        let mut all_values = Vec::new();
+        for values in &group_values {
+            all_values.extend(values.iter().copied());
+            offsets.push(all_values.len() as i32);
+        }
+
+        let values_array = Arc::new(PrimitiveArray::<T>::from_iter_values(all_values));
+        let list_array = ListArray::new(
+            Arc::new(Field::new_list_field(T::DATA_TYPE, true)),
+            OffsetBuffer::new(offsets.into()),
+            values_array,
+            None,
+        );
+
+        Ok(vec![Arc::new(list_array)])
+    }
+
+    fn merge_batch(
+        &mut self,
+        values: &[ArrayRef],
+        group_indices: &[usize],
+        _opt_filter: Option<&BooleanArray>,
+        total_num_groups: usize,
+    ) -> datafusion_common::Result<()> {
+        debug_assert_eq!(values.len(), 1);
+        self.counts.resize(total_num_groups, 0);
+        let list_array = values[0].as_list::<i32>();
+
+        for (row_idx, &group_idx) in group_indices.iter().enumerate() {
+            let inner = list_array.value(row_idx);
+            let inner_arr = inner.as_primitive::<T>();
+            for &value in inner_arr.values().iter() {
+                if self.seen.insert((group_idx, value)) {
+                    self.counts[group_idx] += 1;
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn size(&self) -> usize {
+        size_of::<Self>()
+            + self.seen.capacity() * (size_of::<(usize, T::Native)>() + size_of::<u64>())
+            + self.counts.capacity() * size_of::<i64>()
+    }
+}

datafusion/functions-aggregate/benches/count_distinct.rs

@@ -18,12 +18,12 @@
 use std::sync::Arc;
 
 use arrow::array::{
-    ArrayRef, Int8Array, Int16Array, Int64Array, UInt8Array, UInt16Array,
+    Array, ArrayRef, Int8Array, Int16Array, Int64Array, UInt8Array, UInt16Array,
 };
 use arrow::datatypes::{DataType, Field, Schema};
 use criterion::{Criterion, criterion_group, criterion_main};
 use datafusion_expr::function::AccumulatorArgs;
-use datafusion_expr::{Accumulator, AggregateUDFImpl};
+use datafusion_expr::{Accumulator, AggregateUDFImpl, EmitTo};
 use datafusion_functions_aggregate::count::Count;
 use datafusion_physical_expr::expressions::col;
 use rand::rngs::StdRng;
@@ -87,6 +87,30 @@ fn create_i16_array(n_distinct: usize) -> Int16Array {
         .collect()
 }
 
+fn prepare_args(data_type: DataType) -> (Arc<Schema>, AccumulatorArgs<'static>) {
+    let schema = Arc::new(Schema::new(vec![Field::new("f", data_type, true)]));
+    let schema_leaked: &'static Schema = Box::leak(Box::new((*schema).clone()));
+    let expr = col("f", schema_leaked).unwrap();
+    let expr_leaked: &'static _ = Box::leak(Box::new(expr));
+    let return_field: Arc<Field> = Field::new("f", DataType::Int64, true).into();
+    let return_field_leaked: &'static _ = Box::leak(Box::new(return_field.clone()));
+    let expr_field = expr_leaked.return_field(schema_leaked).unwrap();
+    let expr_field_leaked: &'static _ = Box::leak(Box::new(expr_field));
+
+    let accumulator_args = AccumulatorArgs {
+        return_field: return_field_leaked.clone(),
+        schema: schema_leaked,
+        expr_fields: std::slice::from_ref(expr_field_leaked),
+        ignore_nulls: false,
+        order_bys: &[],
+        is_reversed: false,
+        name: "count(distinct f)",
+        is_distinct: true,
+        exprs: std::slice::from_ref(expr_leaked),
+    };
+    (schema, accumulator_args)
+}
+
 fn count_distinct_benchmark(c: &mut Criterion) {
     for pct in [80, 99] {
         let n_distinct = BATCH_SIZE * pct / 100;
@@ -150,5 +174,107 @@ fn count_distinct_benchmark(c: &mut Criterion) {
     });
 }
 
-criterion_group!(benches, count_distinct_benchmark);
+/// Create group indices with uniform distribution
+fn create_uniform_groups(num_groups: usize) -> Vec<usize> {
+    let mut rng = StdRng::seed_from_u64(42);
+    (0..BATCH_SIZE)
+        .map(|_| rng.random_range(0..num_groups))
+        .collect()
+}
+
+/// Create group indices with skewed distribution (80% in 20% of groups)
+fn create_skewed_groups(num_groups: usize) -> Vec<usize> {
+    let mut rng = StdRng::seed_from_u64(42);
+    let hot_groups = (num_groups / 5).max(1);
+    (0..BATCH_SIZE)
+        .map(|_| {
+            if rng.random_range(0..100) < 80 {
+                rng.random_range(0..hot_groups)
+            } else {
+                rng.random_range(0..num_groups)
+            }
+        })
+        .collect()
+}
+
+fn count_distinct_groups_benchmark(c: &mut Criterion) {
+    let count_fn = Count::new();
+
+    let group_counts = [100, 1000, 10000];
+    let cardinalities = [("low", 20), ("mid", 80), ("high", 99)];
+    let distributions = ["uniform", "skewed"];
+
+    for num_groups in group_counts {
+        for (card_name, distinct_pct) in cardinalities {
+            for dist in distributions {
+                let name = format!("g{num_groups}_{card_name}_{dist}");
+                let n_distinct = BATCH_SIZE * distinct_pct / 100;
+                let values = Arc::new(create_i64_array(n_distinct)) as ArrayRef;
+                let group_indices = if dist == "uniform" {
+                    create_uniform_groups(num_groups)
+                } else {
+                    create_skewed_groups(num_groups)
+                };
+
+                let (_schema, args) = prepare_args(DataType::Int64);
+
+                if count_fn.groups_accumulator_supported(args.clone()) {
+                    c.bench_function(&format!("count_distinct_groups {name}"), |b| {
+                        b.iter(|| {
+                            let mut acc =
+                                count_fn.create_groups_accumulator(args.clone()).unwrap();
+                            acc.update_batch(
+                                std::slice::from_ref(&values),
+                                &group_indices,
+                                None,
+                                num_groups,
+                            )
+                            .unwrap();
+                            acc.evaluate(EmitTo::All).unwrap()
+                        })
+                    });
+                } else {
+                    let arr = values.as_any().downcast_ref::<Int64Array>().unwrap();
+                    let mut group_rows: Vec<Vec<i64>> = vec![Vec::new(); num_groups];
+                    for (idx, &group_idx) in group_indices.iter().enumerate() {
+                        if arr.is_valid(idx) {
+                            group_rows[group_idx].push(arr.value(idx));
+                        }
+                    }
+                    let group_arrays: Vec<ArrayRef> = group_rows
+                        .iter()
+                        .map(|rows| Arc::new(Int64Array::from(rows.clone())) as ArrayRef)
+                        .collect();
+
+                    c.bench_function(&format!("count_distinct_groups {name}"), |b| {
+                        b.iter(|| {
+                            let mut accumulators: Vec<_> = (0..num_groups)
+                                .map(|_| prepare_accumulator(DataType::Int64))
+                                .collect();
+
+                            for (group_idx, batch) in group_arrays.iter().enumerate() {
+                                if !batch.is_empty() {
+                                    accumulators[group_idx]
+                                        .update_batch(std::slice::from_ref(batch))
+                                        .unwrap();
+                                }
+                            }
+
+                            let _results: Vec<_> = accumulators
+                                .iter_mut()
+                                .map(|acc| acc.evaluate().unwrap())
+                                .collect();
+                        })
+                    });
+                }
+            }
+        }
+    }
+}
+
+criterion_group!(
+    benches,
+    count_distinct_benchmark,
+    count_distinct_groups_benchmark
+);
 criterion_main!(benches);