Improved performance of utf8 comparison (1.7x-4x)

benches/comparison_kernels.rs

@@ -42,6 +42,16 @@ fn add_benchmark(c: &mut Criterion) {
         c.bench_function(&format!("bool scalar 2^{}", log2_size), |b| {
             b.iter(|| bench_op_scalar(&arr_a, &BooleanScalar::from(Some(true)), Operator::Eq))
         });
+
+        let arr_a = create_string_array::<i32>(size, 0.1, 42);
+        let arr_b = create_string_array::<i32>(size, 0.1, 43);
+        c.bench_function(&format!("utf8 2^{}", log2_size), |b| {
+            b.iter(|| bench_op(&arr_a, &arr_b, Operator::Eq))
+        });
+
+        c.bench_function(&format!("utf8 2^{}", log2_size), |b| {
+            b.iter(|| bench_op_scalar(&arr_a, &Utf8Scalar::<i32>::from(Some("abc")), Operator::Eq))
+        });
     })
 }
 

benches/filter_kernels.rs

@@ -113,7 +113,7 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| bench_built_filter(&sparse_filter, &data_array))
     });
 
-    let data_array = create_string_array::<i32>(size, 0.5);
+    let data_array = create_string_array::<i32>(size, 0.5, 42);
     c.bench_function("filter context string", |b| {
         b.iter(|| bench_built_filter(&filter, &data_array))
     });

benches/sort_kernel.rs

@@ -80,7 +80,7 @@ fn add_benchmark(c: &mut Criterion) {
             b.iter(|| bench_lexsort(&arr_a, &arr_b))
         });
 
-        let arr_a = create_string_array::<i32>(size, 0.1);
+        let arr_a = create_string_array::<i32>(size, 0.1, 42);
         c.bench_function(&format!("sort utf8 null 2^{}", log2_size), |b| {
             b.iter(|| bench_sort(&arr_a))
         });

benches/take_kernels.rs

@@ -91,36 +91,36 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| bench_take(&values, &indices))
     });
 
-    let values = create_string_array::<i32>(512, 0.0);
+    let values = create_string_array::<i32>(512, 0.0, 42);
     let indices = create_random_index(512, 0.0);
     c.bench_function("take str 512", |b| b.iter(|| bench_take(&values, &indices)));
 
-    let values = create_string_array::<i32>(1024, 0.0);
+    let values = create_string_array::<i32>(1024, 0.0, 42);
     let indices = create_random_index(1024, 0.0);
     c.bench_function("take str 1024", |b| {
         b.iter(|| bench_take(&values, &indices))
     });
 
-    let values = create_string_array::<i32>(512, 0.0);
+    let values = create_string_array::<i32>(512, 0.0, 42);
     let indices = create_random_index(512, 0.5);
     c.bench_function("take str null indices 512", |b| {
         b.iter(|| bench_take(&values, &indices))
     });
 
-    let values = create_string_array::<i32>(1024, 0.0);
+    let values = create_string_array::<i32>(1024, 0.0, 42);
     let indices = create_random_index(1024, 0.5);
     c.bench_function("take str null indices 1024", |b| {
         b.iter(|| bench_take(&values, &indices))
     });
 
-    let values = create_string_array::<i32>(1024, 0.5);
+    let values = create_string_array::<i32>(1024, 0.5, 42);
 
     let indices = create_random_index(1024, 0.0);
     c.bench_function("take str null values 1024", |b| {
         b.iter(|| bench_take(&values, &indices))
     });
 
-    let values = create_string_array::<i32>(1024, 0.5);
+    let values = create_string_array::<i32>(1024, 0.5, 42);
     let indices = create_random_index(1024, 0.5);
     c.bench_function("take str null values null indices 1024", |b| {
         b.iter(|| bench_take(&values, &indices))

benches/write_ipc.rs

@@ -35,7 +35,7 @@ fn add_benchmark(c: &mut Criterion) {
     });
 
     (0..=10).step_by(2).for_each(|i| {
-        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1);
+        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1, 42);
         let a = format!("write utf8 2^{}", 10 + i);
         c.bench_function(&a, |b| b.iter(|| write(array).unwrap()));
     });

benches/write_parquet.rs

@@ -55,13 +55,13 @@ fn add_benchmark(c: &mut Criterion) {
     });
 
     (0..=10).step_by(2).for_each(|i| {
-        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1);
+        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1, 42);
         let a = format!("write utf8 2^{}", 10 + i);
         c.bench_function(&a, |b| b.iter(|| write(array, Encoding::Plain).unwrap()));
     });
 
     (0..=10).step_by(2).for_each(|i| {
-        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1);
+        let array = &create_string_array::<i32>(1024 * 2usize.pow(i), 0.1, 42);
         let a = format!("write utf8 delta 2^{}", 10 + i);
         c.bench_function(&a, |b| {
             b.iter(|| write(array, Encoding::DeltaLengthByteArray).unwrap())

src/compute/comparison/utf8.rs

@@ -37,12 +37,9 @@ where
     let validity = combine_validities(lhs.validity(), rhs.validity());
 
     let values = lhs
-        .iter()
-        .zip(rhs.iter())
-        .map(|(lhs, rhs)| match (lhs, rhs) {
-            (Some(lhs), Some(rhs)) => op(lhs, rhs),
-            _ => false,
-        });
+        .values_iter()
+        .zip(rhs.values_iter())
+        .map(|(lhs, rhs)| op(lhs, rhs));
     let values = Bitmap::from_trusted_len_iter(values);
 
     Ok(BooleanArray::from_data(values, validity))
@@ -57,10 +54,7 @@ where
 {
     let validity = lhs.validity().clone();
 
-    let values = lhs.iter().map(|lhs| match lhs {
-        None => false,
-        Some(lhs) => op(lhs, rhs),
-    });
+    let values = lhs.values_iter().map(|lhs| op(lhs, rhs));
     let values = Bitmap::from_trusted_len_iter(values);
 
     BooleanArray::from_data(values, validity)

src/util/bench_util.rs

@@ -95,15 +95,15 @@ where
 }
 
 /// Creates an random (but fixed-seeded) array of a given size and null density
-pub fn create_string_array<O: Offset>(size: usize, null_density: f32) -> Utf8Array<O> {
-    let rng = &mut seedable_rng();
+pub fn create_string_array<O: Offset>(size: usize, null_density: f32, seed: u64) -> Utf8Array<O> {
+    let mut rng = StdRng::seed_from_u64(seed);
 
     (0..size)
         .map(|_| {
             if rng.gen::<f32>() < null_density {
                 None
             } else {
-                let value = rng
+                let value = (&mut rng)
                     .sample_iter(&Alphanumeric)
                     .take(4)
                     .map(char::from)