perf(expr): add initial microbenchmark for expressions

src/expr/Cargo.toml

@@ -43,3 +43,10 @@ tonic = { version = "0.2", package = "madsim-tonic" }
 
 [target.'cfg(not(madsim))'.dependencies]
 workspace-hack = { path = "../workspace-hack" }
+
+[dev-dependencies]
+criterion = "0.4"
+
+[[bench]]
+name = "expr"
+harness = false

src/expr/benches/expr.rs

@@ -0,0 +1,183 @@
+// Copyright 2022 Singularity Data
+//
+// Licensed 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.
+
+// std try_collect is slower than itertools
+// #![feature(iterator_try_collect)]
+
+// allow using `zip`.
+// `zip_eq` is a source of poor performance.
+#![allow(clippy::disallowed_methods)]
+
+use criterion::{criterion_group, criterion_main, Criterion};
+use risingwave_common::array::{DataChunk, I32Array};
+use risingwave_common::types::DataType;
+use risingwave_expr::expr::expr_binary_nonnull::*;
+use risingwave_expr::expr::*;
+use risingwave_expr::vector_op::agg::create_agg_state_unary;
+use risingwave_pb::expr::expr_node::Type;
+
+criterion_group!(benches, bench_expr, bench_raw);
+criterion_main!(benches);
+
+const CHUNK_SIZE: usize = 1024;
+
+fn bench_expr(c: &mut Criterion) {
+    let input = DataChunk::new(
+        vec![
+            I32Array::from_iter(0..CHUNK_SIZE as i32).into(),
+            I32Array::from_iter(0..CHUNK_SIZE as i32).into(),
+        ],
+        1024,
+    );
+
+    let i0 = || InputRefExpression::new(DataType::Int32, 0).boxed();
+    let i1 = || InputRefExpression::new(DataType::Int32, 1).boxed();
+    c.bench_function("expr/inputref", |bencher| {
+        let inputref = i0();
+        bencher.iter(|| inputref.eval(&input).unwrap())
+    });
+    c.bench_function("expr/constant", |bencher| {
+        let constant = LiteralExpression::new(DataType::Int32, Some(1_i32.into()));
+        bencher.iter(|| constant.eval(&input).unwrap())
+    });
+    c.bench_function("expr/add/i32", |bencher| {
+        let add = new_binary_expr(Type::Add, DataType::Int32, i0(), i1()).unwrap();
+        bencher.iter(|| add.eval(&input).unwrap())
+    });
+    c.bench_function("expr/mul/i32", |bencher| {
+        let mul = new_binary_expr(Type::Multiply, DataType::Int32, i0(), i1()).unwrap();
+        bencher.iter(|| mul.eval(&input).unwrap())
+    });
+    c.bench_function("expr/eq/i32", |bencher| {
+        let mul = new_binary_expr(Type::Equal, DataType::Int32, i0(), i1()).unwrap();
+        bencher.iter(|| mul.eval(&input).unwrap())
+    });
+    c.bench_function("expr/sum/i32", |bencher| {
+        let mut sum =
+            create_agg_state_unary(DataType::Int32, 0, AggKind::Sum, DataType::Int64, false)
+                .unwrap();
+        bencher.iter(|| sum.update_multi(&input, 0, 1024).unwrap())
+    });
+}
+
+/// Evaluate on raw Rust array.
+///
+/// This could be used as a baseline to compare and tune our expressions.
+fn bench_raw(c: &mut Criterion) {
+    // ~55ns
+    c.bench_function("raw/sum/i32", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).collect::<Vec<_>>();
+        bencher.iter(|| a.iter().sum::<i32>())
+    });
+    // ~90ns
+    c.bench_function("raw/add/i32", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).collect::<Vec<_>>();
+        let b = (0..CHUNK_SIZE as i32).collect::<Vec<_>>();
+        bencher.iter(|| {
+            a.iter()
+                .zip(b.iter())
+                .map(|(a, b)| a + b)
+                .collect::<Vec<_>>()
+        })
+    });
+    // ~600ns
+    c.bench_function("raw/add/i32/zip_eq", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).collect::<Vec<_>>();
+        let b = (0..CHUNK_SIZE as i32).collect::<Vec<_>>();
+        bencher.iter(|| {
+            itertools::Itertools::zip_eq(a.iter(), b.iter())
+                .map(|(a, b)| a + b)
+                .collect::<Vec<_>>()
+        })
+    });
+    // ~950ns
+    c.bench_function("raw/add/Option<i32>/zip_eq", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        let b = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        bencher.iter(|| {
+            itertools::Itertools::zip_eq(a.iter(), b.iter())
+                .map(|(a, b)| match (a, b) {
+                    (Some(a), Some(b)) => Some(a + b),
+                    _ => None,
+                })
+                .collect::<Vec<_>>()
+        })
+    });
+    // ~2100ns
+    c.bench_function("raw/add/Option<i32>/zip_eq,checked", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        let b = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        struct Overflow;
+        bencher.iter(|| {
+            use itertools::Itertools;
+            itertools::Itertools::zip_eq(a.iter(), b.iter())
+                .map(|(a, b)| match (a, b) {
+                    (Some(a), Some(b)) => a.checked_add(*b).ok_or(Overflow).map(Some),
+                    _ => Ok(None),
+                })
+                .try_collect::<_, Vec<_>, Overflow>()
+        })
+    });
+    // ~2400ns
+    c.bench_function("raw/add/Option<i32>/zip_eq,checked,cast", |bencher| {
+        let a = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        let b = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+        enum Error {
+            Overflow,
+            Cast,
+        }
+        #[allow(clippy::useless_conversion)]
+        fn checked_add(a: i32, b: i32) -> Result<i32, Error> {
+            let a: i32 = a.try_into().map_err(|_| Error::Cast)?;
+            let b: i32 = b.try_into().map_err(|_| Error::Cast)?;
+            a.checked_add(b).ok_or(Error::Overflow)
+        }
+        bencher.iter(|| {
+            use itertools::Itertools;
+            itertools::Itertools::zip_eq(a.iter(), b.iter())
+                .map(|(a, b)| match (a, b) {
+                    (Some(a), Some(b)) => checked_add(*a, *b).map(Some),
+                    _ => Ok(None),
+                })
+                .try_collect::<_, Vec<_>, Error>()
+        })
+    });
+    // ~3100ns
+    c.bench_function(
+        "raw/add/Option<i32>/zip_eq,checked,cast,collect_array",
+        |bencher| {
+            let a = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+            let b = (0..CHUNK_SIZE as i32).map(Some).collect::<Vec<_>>();
+            enum Error {
+                Overflow,
+                Cast,
+            }
+            #[allow(clippy::useless_conversion)]
+            fn checked_add(a: i32, b: i32) -> Result<i32, Error> {
+                let a: i32 = a.try_into().map_err(|_| Error::Cast)?;
+                let b: i32 = b.try_into().map_err(|_| Error::Cast)?;
+                a.checked_add(b).ok_or(Error::Overflow)
+            }
+            bencher.iter(|| {
+                use itertools::Itertools;
+                itertools::Itertools::zip_eq(a.iter(), b.iter())
+                    .map(|(a, b)| match (a, b) {
+                        (Some(a), Some(b)) => checked_add(*a, *b).map(Some),
+                        _ => Ok(None),
+                    })
+                    .try_collect::<_, I32Array, Error>()
+            })
+        },
+    );
+}

src/expr/src/vector_op/agg/mod.rs

@@ -23,5 +23,5 @@ mod general_distinct_agg;
 mod general_sorted_grouper;
 mod string_agg;
 
-pub use aggregator::{AggStateFactory, BoxedAggState};
+pub use aggregator::{create_agg_state_unary, AggStateFactory, BoxedAggState};
 pub use general_sorted_grouper::{create_sorted_grouper, BoxedSortedGrouper, EqGroups};