Uniform float improvements

.github/workflows/test.yml

@@ -79,7 +79,7 @@ jobs:
         run: |
           cargo test --target ${{ matrix.target }} --features=nightly
           cargo test --target ${{ matrix.target }} --all-features
-          cargo test --target ${{ matrix.target }} --benches --features=nightly
+          cargo test --target ${{ matrix.target }} --benches --features=nightly,small_rng
           cargo test --target ${{ matrix.target }} --manifest-path rand_distr/Cargo.toml --benches
           cargo test --target ${{ matrix.target }} --lib --tests --no-default-features
       - name: Test rand

Cargo.toml

@@ -84,4 +84,9 @@ harness = false
 [[bench]]
 name = "shuffle"
 path = "benches/shuffle.rs"
-harness = false
+harness = false
+
+[[bench]]
+name = "uniform_float"
+path = "benches/uniform_float.rs"
+harness = false

benches/uniform_float.rs

@@ -0,0 +1,103 @@
+// Copyright 2023 Developers of the Rand project.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+//! Implement benchmarks for uniform distributions over FP types
+//!
+//! Sampling methods compared:
+//!
+//! -   sample: current method: (x12 - 1.0) * (b - a) + a
+
+use core::time::Duration;
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use rand::distributions::uniform::{SampleUniform, Uniform, UniformSampler};
+use rand::prelude::*;
+use rand_chacha::ChaCha8Rng;
+use rand_pcg::{Pcg32, Pcg64};
+
+const WARM_UP_TIME: Duration = Duration::from_millis(1000);
+const MEASUREMENT_TIME: Duration = Duration::from_secs(3);
+const SAMPLE_SIZE: usize = 100_000;
+const N_RESAMPLES: usize = 10_000;
+
+macro_rules! single_random {
+    ($R:ty, $T:ty, $g:expr) => {
+        $g.bench_function(BenchmarkId::new(stringify!($T), stringify!($R)), |b| {
+            let mut rng = <$R>::from_entropy();
+            let (mut low, mut high);
+            loop {
+                low = <$T>::from_bits(rng.gen());
+                high = <$T>::from_bits(rng.gen());
+                if (low < high) && (high - low).is_normal() {
+                    break;
+                }
+            }
+
+            b.iter(|| <$T as SampleUniform>::Sampler::sample_single_inclusive(low, high, &mut rng));
+        });
+    };
+
+    ($c:expr, $T:ty) => {{
+        let mut g = $c.benchmark_group("uniform_single");
+        g.sample_size(SAMPLE_SIZE);
+        g.warm_up_time(WARM_UP_TIME);
+        g.measurement_time(MEASUREMENT_TIME);
+        g.nresamples(N_RESAMPLES);
+        single_random!(SmallRng, $T, g);
+        single_random!(ChaCha8Rng, $T, g);
+        single_random!(Pcg32, $T, g);
+        single_random!(Pcg64, $T, g);
+        g.finish();
+    }};
+}
+
+fn single_random(c: &mut Criterion) {
+    single_random!(c, f32);
+    single_random!(c, f64);
+}
+
+macro_rules! distr_random {
+    ($R:ty, $T:ty, $g:expr) => {
+        $g.bench_function(BenchmarkId::new(stringify!($T), stringify!($R)), |b| {
+            let mut rng = <$R>::from_entropy();
+            let dist = loop {
+                let low = <$T>::from_bits(rng.gen());
+                let high = <$T>::from_bits(rng.gen());
+                if let Ok(dist) = Uniform::<$T>::new_inclusive(low, high) {
+                    break dist;
+                }
+            };
+
+            b.iter(|| dist.sample(&mut rng));
+        });
+    };
+
+    ($c:expr, $T:ty) => {{
+        let mut g = $c.benchmark_group("uniform_distribution");
+        g.sample_size(SAMPLE_SIZE);
+        g.warm_up_time(WARM_UP_TIME);
+        g.measurement_time(MEASUREMENT_TIME);
+        g.nresamples(N_RESAMPLES);
+        distr_random!(SmallRng, $T, g);
+        distr_random!(ChaCha8Rng, $T, g);
+        distr_random!(Pcg32, $T, g);
+        distr_random!(Pcg64, $T, g);
+        g.finish();
+    }};
+}
+
+fn distr_random(c: &mut Criterion) {
+    distr_random!(c, f32);
+    distr_random!(c, f64);
+}
+
+criterion_group! {
+    name = benches;
+    config = Criterion::default();
+    targets = single_random, distr_random
+}
+criterion_main!(benches);

src/distributions/uniform.rs

@@ -995,6 +995,38 @@ macro_rules! uniform_float_impl {
                     }
                 }
             }
+
+            #[inline]
+            fn sample_single_inclusive<R: Rng + ?Sized, B1, B2>(low_b: B1, high_b: B2, rng: &mut R) -> Result<Self::X, Error>
+            where
+                B1: SampleBorrow<Self::X> + Sized,
+                B2: SampleBorrow<Self::X> + Sized,
+            {
+                let low = *low_b.borrow();
+                let high = *high_b.borrow();
+                #[cfg(debug_assertions)]
+                if !low.all_finite() || !high.all_finite() {
+                    return Err(Error::NonFinite);
+                }
+                if !low.all_le(high) {
+                    return Err(Error::EmptyRange);
+                }
+                let scale = high - low;
+                if !scale.all_finite() {
+                    return Err(Error::NonFinite);
+                }
+
+                // Generate a value in the range [1, 2)
+                let value1_2 =
+                    (rng.gen::<$uty>() >> $uty::splat($bits_to_discard)).into_float_with_exponent(0);
+
+                // Get a value in the range [0, 1) to avoid overflow when multiplying by scale
+                let value0_1 = value1_2 - <$ty>::splat(1.0);
+
+                // Doing multiply before addition allows some architectures
+                // to use a single instruction.
+                Ok(value0_1 * scale + low)
+            }
         }
     };
 }
@@ -1367,6 +1399,9 @@ mod tests {
                             let v = <$ty as SampleUniform>::Sampler
                                 ::sample_single(low, high, &mut rng).unwrap().extract(lane);
                             assert!(low_scalar <= v && v < high_scalar);
+                            let v = <$ty as SampleUniform>::Sampler
+                                ::sample_single_inclusive(low, high, &mut rng).unwrap().extract(lane);
+                            assert!(low_scalar <= v && v <= high_scalar);
                         }
 
                         assert_eq!(
@@ -1379,8 +1414,19 @@ mod tests {
                         assert_eq!(<$ty as SampleUniform>::Sampler
                             ::sample_single(low, high, &mut zero_rng).unwrap()
                             .extract(lane), low_scalar);
+                        assert_eq!(<$ty as SampleUniform>::Sampler
+                            ::sample_single_inclusive(low, high, &mut zero_rng).unwrap()
+                            .extract(lane), low_scalar);
+
                         assert!(max_rng.sample(my_uniform).extract(lane) < high_scalar);
                         assert!(max_rng.sample(my_incl_uniform).extract(lane) <= high_scalar);
+                        // sample_single cannot cope with max_rng:
+                        // assert!(<$ty as SampleUniform>::Sampler
+                        //     ::sample_single(low, high, &mut max_rng).unwrap()
+                        //     .extract(lane) < high_scalar);
+                        assert!(<$ty as SampleUniform>::Sampler
+                            ::sample_single_inclusive(low, high, &mut max_rng).unwrap()
+                            .extract(lane) <= high_scalar);
 
                         // Don't run this test for really tiny differences between high and low
                         // since for those rounding might result in selecting high for a very