Improved benches. (#636)

Cargo.toml

@@ -117,7 +117,7 @@ full = [
     "regex",
     "compute",
     # parses timezones used in timestamp conversions
-    "chrono-tz"
+    "chrono-tz",
 ]
 io_csv = ["io_csv_read", "io_csv_write"]
 io_csv_async = ["io_csv_read_async"]

benches/aggregate.rs

@@ -1,8 +1,8 @@
 use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::array::*;
+use arrow2::compute::aggregate::*;
 use arrow2::util::bench_util::*;
-use arrow2::{compute::aggregate::*, datatypes::DataType};
 
 fn bench_sum(arr_a: &PrimitiveArray<f32>) {
     sum(criterion::black_box(arr_a)).unwrap();
@@ -15,7 +15,7 @@ fn bench_min(arr_a: &PrimitiveArray<f32>) {
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);
-        let arr_a = create_primitive_array::<f32>(size, DataType::Float32, 0.0);
+        let arr_a = create_primitive_array::<f32>(size, 0.0);
 
         c.bench_function(&format!("sum 2^{} f32", log2_size), |b| {
             b.iter(|| bench_sum(&arr_a))
@@ -24,7 +24,7 @@ fn add_benchmark(c: &mut Criterion) {
             b.iter(|| bench_min(&arr_a))
         });
 
-        let arr_a = create_primitive_array::<f32>(size, DataType::Float32, 0.1);
+        let arr_a = create_primitive_array::<f32>(size, 0.1);
 
         c.bench_function(&format!("sum null 2^{} f32", log2_size), |b| {
             b.iter(|| bench_sum(&arr_a))

benches/arithmetic_kernels.rs

@@ -1,12 +1,9 @@
-#[macro_use]
-extern crate criterion;
-use criterion::Criterion;
+use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::array::*;
 use arrow2::util::bench_util::*;
 use arrow2::{
-    compute::arithmetics::basic::add, compute::arithmetics::basic::div_scalar, datatypes::DataType,
-    types::NativeType,
+    compute::arithmetics::basic::add, compute::arithmetics::basic::div_scalar, types::NativeType,
 };
 use num_traits::NumCast;
 use std::ops::{Add, Div};
@@ -28,8 +25,8 @@ where
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);
-        let arr_a = create_primitive_array_with_seed::<u64>(size, DataType::UInt64, 0.0, 43);
-        let arr_b = create_primitive_array_with_seed::<u64>(size, DataType::UInt64, 0.0, 42);
+        let arr_a = create_primitive_array_with_seed::<u64>(size, 0.0, 43);
+        let arr_b = create_primitive_array_with_seed::<u64>(size, 0.0, 42);
 
         c.bench_function(&format!("divide_scalar 2^{}", log2_size), |b| {
             // 4 is a very fast optimizable divisor

benches/bitmap.rs

@@ -1,12 +1,8 @@
-extern crate arrow2;
-
 use std::iter::FromIterator;
 
-use arrow2::bitmap::*;
-
 use criterion::{criterion_group, criterion_main, Criterion};
 
-//
+use arrow2::bitmap::*;
 
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {

benches/bitmap_ops.rs

@@ -1,7 +1,7 @@
-use arrow2::bitmap::Bitmap;
-
 use criterion::{criterion_group, criterion_main, Criterion};
 
+use arrow2::bitmap::Bitmap;
+
 fn bench_arrow2(lhs: &Bitmap, rhs: &Bitmap) {
     let r = lhs | rhs;
     assert!(r.null_count() > 0);

benches/bitwise.rs

@@ -4,34 +4,29 @@ use criterion::{criterion_group, criterion_main, Criterion};
 use num_traits::NumCast;
 
 use arrow2::{
-    array::PrimitiveArray,
-    compute::bitwise::*,
-    datatypes::DataType,
-    types::NativeType,
-    util::bench_util::{
-        create_boolean_array, create_primitive_array, create_primitive_array_with_seed,
-    },
+    array::PrimitiveArray, compute::bitwise::*, types::NativeType,
+    util::bench_util::create_primitive_array_with_seed,
 };
 
 fn bench_or<T>(lhs: &PrimitiveArray<T>, rhs: &PrimitiveArray<T>)
 where
     T: NativeType + BitOr<Output = T> + NumCast,
 {
-    criterion::black_box(or(lhs, rhs)).unwrap();
+    criterion::black_box(or(lhs, rhs));
 }
 
 fn bench_xor<T>(lhs: &PrimitiveArray<T>, rhs: &PrimitiveArray<T>)
 where
     T: NativeType + BitXor<Output = T> + NumCast,
 {
-    criterion::black_box(xor(lhs, rhs)).unwrap();
+    criterion::black_box(xor(lhs, rhs));
 }
 
 fn bench_and<T>(lhs: &PrimitiveArray<T>, rhs: &PrimitiveArray<T>)
 where
     T: NativeType + BitAnd<Output = T> + NumCast,
 {
-    criterion::black_box(and(lhs, rhs)).unwrap();
+    criterion::black_box(and(lhs, rhs));
 }
 
 fn bench_not<T>(arr: &PrimitiveArray<T>)
@@ -44,8 +39,8 @@ where
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);
-        let arr_a = create_primitive_array_with_seed::<u64>(size, DataType::UInt64, 0.0, 43);
-        let arr_b = create_primitive_array_with_seed::<u64>(size, DataType::UInt64, 0.0, 42);
+        let arr_a = create_primitive_array_with_seed::<u64>(size, 0.0, 43);
+        let arr_b = create_primitive_array_with_seed::<u64>(size, 0.0, 42);
 
         c.bench_function(&format!("or 2^{}", log2_size), |b| {
             b.iter(|| bench_or(&arr_a, &arr_b))

benches/cast_kernels.rs

@@ -15,9 +15,7 @@
 // specific language governing permissions and limitations
 // under the License.
 
-#[macro_use]
-extern crate criterion;
-use criterion::Criterion;
+use criterion::{criterion_group, criterion_main, Criterion};
 use rand::distributions::Uniform;
 use rand::Rng;
 
@@ -72,27 +70,27 @@ fn build_utf8_date_time_array(size: usize, with_nulls: bool) -> Utf8Array<i32> {
 
 // cast array from specified primitive array type to desired data type
 fn cast_array(array: &dyn Array, to_type: DataType) {
-    criterion::black_box(cast::cast(array, &to_type).unwrap());
+    criterion::black_box(cast::cast(array, &to_type, Default::default()).unwrap());
 }
 
 fn add_benchmark(c: &mut Criterion) {
     let size = 512;
-    let i32_array = create_primitive_array::<i32>(size, DataType::Int32, 0.1);
-    let i64_array = create_primitive_array::<i64>(size, DataType::Int64, 0.1);
-    let f32_array = create_primitive_array::<f32>(size, DataType::Float32, 0.1);
-    let f32_utf8_array = cast::cast(&f32_array, &DataType::Utf8).unwrap();
+    let i32_array = create_primitive_array::<i32>(size, 0.1);
+    let i64_array = create_primitive_array::<i64>(size, 0.1);
+    let f32_array = create_primitive_array::<f32>(size, 0.1);
+    let f32_utf8_array = cast::cast(&f32_array, &DataType::Utf8, Default::default()).unwrap();
 
-    let f64_array = create_primitive_array::<f64>(size, DataType::Float64, 0.1);
-    let date64_array = create_primitive_array::<i64>(size, DataType::Date64, 0.1);
-    let date32_array = create_primitive_array::<i32>(size, DataType::Date32, 0.1);
+    let f64_array = create_primitive_array::<f64>(size, 0.1);
+    let date64_array = create_primitive_array::<i64>(size, 0.1).to(DataType::Date64);
+    let date32_array = create_primitive_array::<i32>(size, 0.1).to(DataType::Date32);
     let time32s_array =
-        create_primitive_array::<i32>(size, DataType::Time32(TimeUnit::Second), 0.1);
+        create_primitive_array::<i32>(size, 0.1).to(DataType::Time32(TimeUnit::Second));
     let time64ns_array =
-        create_primitive_array::<i64>(size, DataType::Time64(TimeUnit::Nanosecond), 0.1);
-    let time_ns_array =
-        create_primitive_array::<i64>(size, DataType::Timestamp(TimeUnit::Nanosecond, None), 0.1);
-    let time_ms_array =
-        create_primitive_array::<i64>(size, DataType::Timestamp(TimeUnit::Millisecond, None), 0.1);
+        create_primitive_array::<i64>(size, 0.1).to(DataType::Time64(TimeUnit::Nanosecond));
+    let time_ns_array = create_primitive_array::<i64>(size, 0.1)
+        .to(DataType::Timestamp(TimeUnit::Nanosecond, None));
+    let time_ms_array = create_primitive_array::<i64>(size, 0.1)
+        .to(DataType::Timestamp(TimeUnit::Millisecond, None));
     let utf8_date_array = build_utf8_date_array(512, true);
     let utf8_date_time_array = build_utf8_date_time_array(512, true);
 

benches/comparison_kernels.rs

@@ -1,15 +1,15 @@
 use criterion::{criterion_group, criterion_main, Criterion};
 
+use arrow2::compute::comparison::{eq, eq_scalar};
 use arrow2::scalar::*;
 use arrow2::util::bench_util::*;
-use arrow2::{compute::comparison::eq, datatypes::DataType};
 
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);
 
-        let arr_a = create_primitive_array_with_seed::<f32>(size, DataType::Float32, 0.0, 42);
-        let arr_b = create_primitive_array_with_seed::<f32>(size, DataType::Float32, 0.0, 43);
+        let arr_a = create_primitive_array_with_seed::<f32>(size, 0.0, 42);
+        let arr_b = create_primitive_array_with_seed::<f32>(size, 0.0, 43);
 
         c.bench_function(&format!("f32 2^{}", log2_size), |b| {
             b.iter(|| eq(&arr_a, &arr_b))

benches/concatenate.rs

@@ -1,27 +1,24 @@
-extern crate arrow2;
+use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::{
     compute::concatenate::concatenate,
-    datatypes::DataType,
     util::bench_util::{create_boolean_array, create_primitive_array},
 };
 
-use criterion::{criterion_group, criterion_main, Criterion};
-
 fn add_benchmark(c: &mut Criterion) {
     (20..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);
 
-        let array1 = create_primitive_array::<i32>(8, DataType::Int32, 0.5);
-        let array2 = create_primitive_array::<i32>(size + 1, DataType::Int32, 0.5);
+        let array1 = create_primitive_array::<i32>(8, 0.5);
+        let array2 = create_primitive_array::<i32>(size + 1, 0.5);
 
         c.bench_function(&format!("int32 concat aligned 2^{}", log2_size), |b| {
             b.iter(|| {
                 let _ = concatenate(&[&array1, &array2]);
             })
         });
 
-        let array1 = create_primitive_array::<i32>(9, DataType::Int32, 0.5);
+        let array1 = create_primitive_array::<i32>(9, 0.5);
 
         c.bench_function(&format!("int32 concat unaligned 2^{}", log2_size), |b| {
             b.iter(|| {

benches/count_zeros.rs

@@ -1,7 +1,7 @@
-use arrow2::bitmap::utils::count_zeros;
-
 use criterion::{criterion_group, criterion_main, Criterion};
 
+use arrow2::bitmap::utils::count_zeros;
+
 fn add_benchmark(c: &mut Criterion) {
     (10..=20).step_by(2).for_each(|log2_size| {
         let size = 2usize.pow(log2_size);

benches/filter_kernels.rs

@@ -14,17 +14,15 @@
 // KIND, either express or implied.  See the License for the
 // specific language governing permissions and limitations
 // under the License.
-extern crate arrow2;
-
 use std::sync::Arc;
 
+use criterion::{criterion_group, criterion_main, Criterion};
+
 use arrow2::array::*;
 use arrow2::compute::filter::{build_filter, filter, filter_record_batch, Filter};
 use arrow2::datatypes::{DataType, Field, Schema};
 use arrow2::record_batch::RecordBatch;
-
 use arrow2::util::bench_util::{create_boolean_array, create_primitive_array, create_string_array};
-use criterion::{criterion_group, criterion_main, Criterion};
 
 fn bench_filter(data_array: &dyn Array, filter_array: &BooleanArray) {
     criterion::black_box(filter(data_array, filter_array).unwrap());
@@ -43,12 +41,12 @@ fn add_benchmark(c: &mut Criterion) {
         let filter_array =
             BooleanArray::from_data(DataType::Boolean, filter_array.values().clone(), None);
 
-        let arr_a = create_primitive_array::<f32>(size, DataType::Float32, 0.0);
+        let arr_a = create_primitive_array::<f32>(size, 0.0);
         c.bench_function(&format!("filter 2^{} f32", log2_size), |b| {
             b.iter(|| bench_filter(&arr_a, &filter_array))
         });
 
-        let arr_a = create_primitive_array::<f32>(size, DataType::Float32, 0.1);
+        let arr_a = create_primitive_array::<f32>(size, 0.1);
 
         c.bench_function(&format!("filter null 2^{} f32", log2_size), |b| {
             b.iter(|| bench_filter(&arr_a, &filter_array))
@@ -64,7 +62,7 @@ fn add_benchmark(c: &mut Criterion) {
     let dense_filter = build_filter(&dense_filter_array).unwrap();
     let sparse_filter = build_filter(&sparse_filter_array).unwrap();
 
-    let data_array = create_primitive_array::<u8>(size, DataType::UInt8, 0.0);
+    let data_array = create_primitive_array::<u8>(size, 0.0);
 
     c.bench_function("filter u8", |b| {
         b.iter(|| bench_filter(&data_array, &filter_array))
@@ -86,7 +84,7 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| bench_built_filter(&sparse_filter, &data_array))
     });
 
-    let data_array = create_primitive_array::<u8>(size, DataType::UInt8, 0.5);
+    let data_array = create_primitive_array::<u8>(size, 0.5);
     c.bench_function("filter context u8 w NULLs", |b| {
         b.iter(|| bench_built_filter(&filter, &data_array))
     });
@@ -97,7 +95,7 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| bench_built_filter(&sparse_filter, &data_array))
     });
 
-    let data_array = create_primitive_array::<f32>(size, DataType::Float32, 0.5);
+    let data_array = create_primitive_array::<f32>(size, 0.5);
     c.bench_function("filter f32", |b| {
         b.iter(|| bench_filter(&data_array, &filter_array))
     });
@@ -125,7 +123,7 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| bench_built_filter(&sparse_filter, &data_array))
     });
 
-    let data_array = create_primitive_array::<f32>(size, DataType::Float32, 0.0);
+    let data_array = create_primitive_array::<f32>(size, 0.0);
 
     let field = Field::new("c1", data_array.data_type().clone(), true);
     let schema = Schema::new(vec![field]);

benches/from_trusted_len_iter.rs

@@ -1,9 +1,7 @@
-extern crate arrow2;
+use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::{array::PrimitiveArray, bitmap::*, buffer::*};
 
-use criterion::{criterion_group, criterion_main, Criterion};
-
 fn add_benchmark(c: &mut Criterion) {
     let values = 0..1026;
 

benches/growable.rs

@@ -1,17 +1,14 @@
-extern crate arrow2;
+use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::{
     array::growable::{Growable, GrowablePrimitive},
-    datatypes::DataType,
     util::bench_util::create_primitive_array,
 };
 
-use criterion::{criterion_group, criterion_main, Criterion};
-
 fn add_benchmark(c: &mut Criterion) {
     let values = (0..1026).rev();
 
-    let i32_array = create_primitive_array::<i32>(1026 * 10, DataType::Int32, 0.0);
+    let i32_array = create_primitive_array::<i32>(1026 * 10, 0.0);
     c.bench_function("growable::primitive::non_null::non_null", |b| {
         b.iter(|| {
             let mut a = GrowablePrimitive::new(vec![&i32_array], false, 1026 * 10);
@@ -22,7 +19,7 @@ fn add_benchmark(c: &mut Criterion) {
         })
     });
 
-    let i32_array = create_primitive_array::<i32>(1026 * 10, DataType::Int32, 0.0);
+    let i32_array = create_primitive_array::<i32>(1026 * 10, 0.0);
     c.bench_function("growable::primitive::non_null::null", |b| {
         b.iter(|| {
             let mut a = GrowablePrimitive::new(vec![&i32_array], true, 1026 * 10);
@@ -36,7 +33,7 @@ fn add_benchmark(c: &mut Criterion) {
         })
     });
 
-    let i32_array = create_primitive_array::<i32>(1026 * 10, DataType::Int32, 0.1);
+    let i32_array = create_primitive_array::<i32>(1026 * 10, 0.1);
 
     let values = values.collect::<Vec<_>>();
     c.bench_function("growable::primitive::null::non_null", |b| {

benches/hash_kernel.rs

@@ -1,20 +1,17 @@
-extern crate arrow2;
+use criterion::{criterion_group, criterion_main, Criterion};
 
 use arrow2::compute::hash::hash;
-use arrow2::datatypes::DataType;
 use arrow2::util::bench_util::*;
 
-use criterion::{criterion_group, criterion_main, Criterion};
-
 fn add_benchmark(c: &mut Criterion) {
     let log2_size = 10;
     let size = 2usize.pow(log2_size);
 
-    let arr_a = create_primitive_array::<i32>(size, DataType::Int32, 0.0);
+    let arr_a = create_primitive_array::<i32>(size, 0.0);
 
     c.bench_function(&format!("i32 2^{}", log2_size), |b| b.iter(|| hash(&arr_a)));
 
-    let arr_a = create_primitive_array::<i64>(size, DataType::Int64, 0.0);
+    let arr_a = create_primitive_array::<i64>(size, 0.0);
 
     c.bench_function(&format!("i64 2^{}", log2_size), |b| b.iter(|| hash(&arr_a)));