usearch 2.9.0 → 2.9.2

package/simsimd/rust/lib.rs

@@ -1,27 +1,19 @@
 //! # SimSIMD - Hardware-Accelerated Similarity Metrics and Distance Functions
 //!
-//! * Zero-dependency header-only C 99 library with bindings for Python, JavaScript and Rust.
 //! * Targets ARM NEON, SVE, x86 AVX2, AVX-512 (VNNI, FP16) hardware backends.
-//! * Zero-copy compatible with NumPy, PyTorch, TensorFlow, and other tensors.
-//! * Handles f64 double-, f32 single-, and f16 half-precision, i8 integral, and binary vectors.
-//! * Up to 200x faster than scipy.spatial.distance and numpy.inner.
-//! * Used in USearch and several DBMS products.
+//! * Handles `f64` double-, `f32` single-, and `f16` half-precision, `i8` integral, and binary vectors.
+//! * Zero-dependency header-only C 99 library with bindings for Rust and other langauges.
 //!
 //! ## Implemented distance functions include:
 //!
 //! * Euclidean (L2), Inner Distance, and Cosine (Angular) spatial distances.
 //! * Hamming (~ Manhattan) and Jaccard (~ Tanimoto) binary distances.
-//! * Kullback-Leibler and Jensen–Shannon divergences for probability distributions.
+//! * Kullback-Leibler and Jensen-Shannon divergences for probability distributions.
 //!
-//! The functions in this module are exposed through a trait `SimSIMD`, which is
-//! implemented for types with SIMD support (e.g., i8 and f32). The trait provides
-//! methods for computing cosine similarity, inner product, and squared Euclidean
-//! distance between slices of these types.
-//!
-//! # Example
+//! ## Example
 //!
 //! ```rust
-//! use simsimd::SimSIMD;
+//! use simsimd::SpatialSimilarity;
 //!
 //! let a = &[1, 2, 3];
 //! let b = &[4, 5, 6];
@@ -29,74 +21,165 @@
 //! // Compute cosine similarity
 //! let cosine_sim = i8::cosine(a, b);
 //!
-//! // Compute inner product
+//! // Compute inner product distance
 //! let inner_product = i8::inner(a, b);
 //!
 //! // Compute squared Euclidean distance
 //! let sqeuclidean_dist = i8::sqeuclidean(a, b);
 //! ```
 //!
-//! # Safety
+//! ## Traits
+//!
+//! The `SpatialSimilarity` trait covers following methods:
 //!
-//! The functions declared in the `extern "C"` block are low-level bindings to
-//! SimSIMD implementation. It is crucial to ensure that the input slices have the
-//! same length (`c` parameter) to avoid undefined behavior.
+//! - `cosine(a: &[Self], b: &[Self]) -> Option<f32>`: Computes cosine similarity between two slices.
+//! - `inner(a: &[Self], b: &[Self]) -> Option<f32>`: Computes inner product distance between two slices.
+//! - `sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32>`: Computes squared Euclidean distance between two slices.
 //!
+//! The `BinarySimilarity` trait covers following methods:
 //!
-//! # Trait Implementation
+//! - `hamming(a: &[Self], b: &[Self]) -> Option<f32>`: Computes Hamming distance between two slices.
+//! - `jaccard(a: &[Self], b: &[Self]) -> Option<f32>`: Computes Jaccard index between two slices.
 //!
-//! The `SimSIMD` trait is implemented for types with SIMD support and provides
-//! three associated methods:
+//! The `ProbabilitySimilarity` trait covers following methods:
 //!
-//! - `cosine(a: &[Self], b: &[Self]) -> Option<f32>`: Computes cosine similarity between two slices.
-//! - `inner(a: &[Self], b: &[Self]) -> Option<f32>`: Computes inner product between two slices.
-//! - `sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32>`: Computes squared Euclidean distance between two slices.
+//! - `jensenshannon(a: &[Self], b: &[Self]) -> Option<f32>`: Computes Jensen-Shannon divergence between two slices.
+//! - `kullbackleibler(a: &[Self], b: &[Self]) -> Option<f32>`: Computes Kullback-Leibler divergence between two slices.
 //!
 #![allow(non_camel_case_types)]
 
 extern "C" {
-    /// Computes cosine similarity for i8 types.
     fn cosine_i8(a: *const i8, b: *const i8, c: usize) -> f32;
-
-    /// Computes cosine similarity for f32 types.
+    fn cosine_f16(a: *const u16, b: *const u16, c: usize) -> f32;
     fn cosine_f32(a: *const f32, b: *const f32, c: usize) -> f32;
+    fn cosine_f64(a: *const f64, b: *const f64, c: usize) -> f32;
 
-    /// Computes inner product for i8 types.
     fn inner_i8(a: *const i8, b: *const i8, c: usize) -> f32;
-
-    /// Computes inner product for f32 types.
+    fn inner_f16(a: *const u16, b: *const u16, c: usize) -> f32;
     fn inner_f32(a: *const f32, b: *const f32, c: usize) -> f32;
+    fn inner_f64(a: *const f64, b: *const f64, c: usize) -> f32;
 
-    /// Computes squared Euclidean distance for i8 types.
     fn sqeuclidean_i8(a: *const i8, b: *const i8, c: usize) -> f32;
-
-    /// Computes squared Euclidean distance for f32 types.
+    fn sqeuclidean_f16(a: *const u16, b: *const u16, c: usize) -> f32;
     fn sqeuclidean_f32(a: *const f32, b: *const f32, c: usize) -> f32;
+    fn sqeuclidean_f64(a: *const f64, b: *const f64, c: usize) -> f32;
+
+    fn hamming_b8(a: *const u8, b: *const u8, c: usize) -> f32;
+    fn jaccard_b8(a: *const u8, b: *const u8, c: usize) -> f32;
+
+    fn js_f16(a: *const u16, b: *const u16, c: usize) -> f32;
+    fn js_f32(a: *const f32, b: *const f32, c: usize) -> f32;
+    fn js_f64(a: *const f64, b: *const f64, c: usize) -> f32;
+
+    fn kl_f16(a: *const u16, b: *const u16, c: usize) -> f32;
+    fn kl_f32(a: *const f32, b: *const f32, c: usize) -> f32;
+    fn kl_f64(a: *const f64, b: *const f64, c: usize) -> f32;
 }
 
-/// A trait for SIMD similarity functions.
-pub trait SimSIMD
+/// A half-precision floating point number.
+#[repr(transparent)]
+pub struct f16(u16);
+
+impl f16 {}
+
+/// `SpatialSimilarity` provides a set of trait methods for computing similarity
+/// or distance between spatial data vectors in SIMD (Single Instruction, Multiple Data) context.
+/// These methods can be used to calculate metrics like cosine similarity, inner product,
+/// and squared Euclidean distance between two slices of data.
+///
+/// Each method takes two slices of data (a and b) and returns an Option<f32>.
+/// The result is `None` if the slices are not of the same length, as these operations
+/// require one-to-one correspondence between the elements of the slices.
+/// Otherwise, it returns the computed similarity or distance as `Some(f32)`.
+pub trait SpatialSimilarity
 where
     Self: Sized,
 {
-    /// Computes cosine similarity between two slices.
+    /// Computes the cosine similarity between two slices.
+    /// The cosine similarity is a measure of similarity between two non-zero vectors
+    /// of an inner product space that measures the cosine of the angle between them.
     fn cosine(a: &[Self], b: &[Self]) -> Option<f32>;
 
-    /// Computes inner product between two slices.
+    /// Computes the inner product (also known as dot product) between two slices.
+    /// The inner product is the sum of the products of the corresponding entries
+    /// of the two sequences of numbers.
     fn inner(a: &[Self], b: &[Self]) -> Option<f32>;
 
-    /// Computes squared Euclidean distance between two slices.
+    /// Computes the squared Euclidean distance between two slices.
+    /// The squared Euclidean distance is the sum of the squared differences
+    /// between corresponding elements of the two slices.
     fn sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32>;
 }
 
-impl SimSIMD for i8 {
-    fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
+/// `BinarySimilarity` provides trait methods for computing similarity metrics
+/// that are commonly used with binary data vectors, such as Hamming distance
+/// and Jaccard index.
+///
+/// The methods accept two slices of binary data and return an Option<f32>
+/// indicating the computed similarity or distance, with `None` returned if the
+/// slices differ in length.
+pub trait BinarySimilarity
+where
+    Self: Sized,
+{
+    /// Computes the Hamming distance between two binary data slices.
+    /// The Hamming distance between two strings of equal length is the number of
+    /// bits at which the corresponding values are different.
+    fn hamming(a: &[Self], b: &[Self]) -> Option<f32>;
+
+    /// Computes the Jaccard index between two bitsets represented by binary data slices.
+    /// The Jaccard index, also known as the Jaccard similarity coefficient, is a statistic
+    /// used for gauging the similarity and diversity of sample sets.
+    fn jaccard(a: &[Self], b: &[Self]) -> Option<f32>;
+}
+
+/// `ProbabilitySimilarity` provides trait methods for computing similarity or divergence
+/// measures between probability distributions, such as the Jensen-Shannon divergence
+/// and the Kullback-Leibler divergence.
+///
+/// These methods are particularly useful in contexts such as information theory and
+/// machine learning, where one often needs to measure how one probability distribution
+/// differs from a second, reference probability distribution.
+pub trait ProbabilitySimilarity
+where
+    Self: Sized,
+{
+    /// Computes the Jensen-Shannon divergence between two probability distributions.
+    /// The Jensen-Shannon divergence is a method of measuring the similarity between
+    /// two probability distributions. It is based on the Kullback-Leibler divergence,
+    /// but is symmetric and always has a finite value.
+    fn jensenshannon(a: &[Self], b: &[Self]) -> Option<f32>;
+
+    /// Computes the Kullback-Leibler divergence between two probability distributions.
+    /// The Kullback-Leibler divergence is a measure of how one probability distribution
+    /// diverges from a second, expected probability distribution.
+    fn kullbackleibler(a: &[Self], b: &[Self]) -> Option<f32>;
+}
+
+impl BinarySimilarity for u8 {
+    fn hamming(a: &[Self], b: &[Self]) -> Option<f32> {
         if a.len() != b.len() {
             return None;
         }
+        let operation = unsafe { hamming_b8(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
 
-        let operation = unsafe { cosine_i8(a.as_ptr(), b.as_ptr(), a.len()) };
+    fn jaccard(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { jaccard_b8(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+}
 
+impl SpatialSimilarity for i8 {
+    fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { cosine_i8(a.as_ptr(), b.as_ptr(), a.len()) };
         Some(operation)
     }
 
@@ -104,9 +187,7 @@ impl SimSIMD for i8 {
         if a.len() != b.len() {
             return None;
         }
-
         let operation = unsafe { inner_i8(a.as_ptr(), b.as_ptr(), a.len()) };
-
         Some(operation)
     }
 
@@ -114,21 +195,21 @@ impl SimSIMD for i8 {
         if a.len() != b.len() {
             return None;
         }
-
         let operation = unsafe { sqeuclidean_i8(a.as_ptr(), b.as_ptr(), a.len()) };
-
         Some(operation)
     }
 }
 
-impl SimSIMD for f32 {
+impl SpatialSimilarity for f16 {
     fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
         if a.len() != b.len() {
             return None;
         }
 
-        let operation = unsafe { cosine_f32(a.as_ptr(), b.as_ptr(), a.len()) };
-
+        // Explicitly cast `*const f16` to `*const u16`
+        let a_ptr = a.as_ptr() as *const u16;
+        let b_ptr = b.as_ptr() as *const u16;
+        let operation = unsafe { cosine_f16(a_ptr, b_ptr, a.len()) };
         Some(operation)
     }
 
@@ -137,8 +218,66 @@ impl SimSIMD for f32 {
             return None;
         }
 
+        // Explicitly cast `*const f16` to `*const u16`
+        let a_ptr = a.as_ptr() as *const u16;
+        let b_ptr = b.as_ptr() as *const u16;
+        let operation = unsafe { inner_f16(a_ptr, b_ptr, a.len()) };
+        Some(operation)
+    }
+
+    fn sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+
+        // Explicitly cast `*const f16` to `*const u16`
+        let a_ptr = a.as_ptr() as *const u16;
+        let b_ptr = b.as_ptr() as *const u16;
+        let operation = unsafe { sqeuclidean_f16(a_ptr, b_ptr, a.len()) };
+        Some(operation)
+    }
+}
+
+impl SpatialSimilarity for f32 {
+    fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { cosine_f32(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+
+    fn inner(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
         let operation = unsafe { inner_f32(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
 
+    fn sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { sqeuclidean_f32(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+}
+
+impl SpatialSimilarity for f64 {
+    fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { cosine_f64(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+
+    fn inner(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { inner_f64(a.as_ptr(), b.as_ptr(), a.len()) };
         Some(operation)
     }
 
@@ -146,16 +285,101 @@ impl SimSIMD for f32 {
         if a.len() != b.len() {
             return None;
         }
+        let operation = unsafe { sqeuclidean_f64(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+}
+
+impl ProbabilitySimilarity for f16 {
+    fn jensenshannon(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
 
-        let operation = unsafe { sqeuclidean_f32(a.as_ptr(), b.as_ptr(), a.len()) };
+        // Explicitly cast `*const f16` to `*const u16`
+        let a_ptr = a.as_ptr() as *const u16;
+        let b_ptr = b.as_ptr() as *const u16;
+        let operation = unsafe { js_f16(a_ptr, b_ptr, a.len()) };
+        Some(operation)
+    }
+
+    fn kullbackleibler(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+
+        // Explicitly cast `*const f16` to `*const u16`
+        let a_ptr = a.as_ptr() as *const u16;
+        let b_ptr = b.as_ptr() as *const u16;
+        let operation = unsafe { kl_f16(a_ptr, b_ptr, a.len()) };
+        Some(operation)
+    }
+}
+
+impl ProbabilitySimilarity for f32 {
+    fn jensenshannon(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { js_f32(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
 
+    fn kullbackleibler(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { kl_f32(a.as_ptr(), b.as_ptr(), a.len()) };
         Some(operation)
     }
 }
 
+impl ProbabilitySimilarity for f64 {
+    fn jensenshannon(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { js_f64(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+
+    fn kullbackleibler(a: &[Self], b: &[Self]) -> Option<f32> {
+        if a.len() != b.len() {
+            return None;
+        }
+        let operation = unsafe { kl_f64(a.as_ptr(), b.as_ptr(), a.len()) };
+        Some(operation)
+    }
+}
+
+// In the older revisions of the library, the `SpatialSimilarity`
+// trait was called `SimSIMD`. The following trait provides a compatibility layer.
+pub trait SimSIMD
+where
+    Self: Sized,
+{
+    fn cosine(a: &[Self], b: &[Self]) -> Option<f32>;
+    fn inner(a: &[Self], b: &[Self]) -> Option<f32>;
+    fn sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32>;
+}
+
+impl<T: SpatialSimilarity> SimSIMD for T {
+    fn cosine(a: &[Self], b: &[Self]) -> Option<f32> {
+        SpatialSimilarity::cosine(a, b)
+    }
+    fn inner(a: &[Self], b: &[Self]) -> Option<f32> {
+        SpatialSimilarity::inner(a, b)
+    }
+    fn sqeuclidean(a: &[Self], b: &[Self]) -> Option<f32> {
+        SpatialSimilarity::sqeuclidean(a, b)
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
+    use half::f16 as HalfF16;
+
     //
     fn assert_almost_equal(left: f32, right: f32, tolerance: f32) {
         let lower = right - tolerance;
@@ -170,19 +394,19 @@ mod tests {
         let b = &[3, 97, 127];
 
         if let Some(result) = SimSIMD::cosine(a, b) {
-            assert_almost_equal(0.00012027938, result, 0.01);
             println!("The result of cosine_i8 is {:.8}", result);
+            assert_almost_equal(0.00012027938, result, 0.01);
         }
     }
 
     #[test]
     fn test_cosine_f32() {
         let a = &[1.0, 2.0, 3.0];
-        let b = &[1.0, 2.0, 3.0];
+        let b = &[4.0, 5.0, 6.0];
 
         if let Some(result) = SimSIMD::cosine(a, b) {
-            assert_almost_equal(0.004930496, result, 0.01);
             println!("The result of cosine_f32 is {:.8}", result);
+            assert_almost_equal(0.025, result, 0.01);
         }
     }
 
@@ -192,8 +416,8 @@ mod tests {
         let b = &[4, 5, 6];
 
         if let Some(result) = SimSIMD::inner(a, b) {
-            assert_almost_equal(0.029403687, result, 0.01);
             println!("The result of inner_i8 is {:.8}", result);
+            assert_almost_equal(0.029403687, result, 0.01);
         }
     }
 
@@ -203,8 +427,8 @@ mod tests {
         let b = &[4.0, 5.0, 6.0];
 
         if let Some(result) = SimSIMD::inner(a, b) {
-            assert_almost_equal(-31.0, result, 0.01);
             println!("The result of inner_f32 is {:.8}", result);
+            assert_almost_equal(-31.0, result, 0.01);
         }
     }
 
@@ -214,8 +438,8 @@ mod tests {
         let b = &[4, 5, 6];
 
         if let Some(result) = SimSIMD::sqeuclidean(a, b) {
-            assert_almost_equal(27.0, result, 0.01);
             println!("The result of sqeuclidean_i8 is {:.8}", result);
+            assert_almost_equal(27.0, result, 0.01);
         }
     }
 
@@ -225,8 +449,102 @@ mod tests {
         let b = &[4.0, 5.0, 6.0];
 
         if let Some(result) = SimSIMD::sqeuclidean(a, b) {
-            assert_almost_equal(27.0, result, 0.01);
             println!("The result of sqeuclidean_f32 is {:.8}", result);
+            assert_almost_equal(27.0, result, 0.01);
+        }
+    }
+
+    // Adding new tests for bit-level distances
+    #[test]
+    fn test_hamming_u8() {
+        let a = &[0b01010101, 0b11110000, 0b10101010]; // Binary representations for clarity
+        let b = &[0b01010101, 0b11110000, 0b10101010];
+
+        if let Some(result) = BinarySimilarity::hamming(a, b) {
+            println!("The result of hamming_u8 is {:.8}", result);
+            assert_almost_equal(0.0, result, 0.01); // Perfect match
+        }
+    }
+
+    #[test]
+    fn test_jaccard_u8() {
+        // For binary data, treat each byte as a set of bits
+        let a = &[0b11110000, 0b00001111, 0b10101010];
+        let b = &[0b11110000, 0b00001111, 0b01010101];
+
+        if let Some(result) = BinarySimilarity::jaccard(a, b) {
+            println!("The result of jaccard_u8 is {:.8}", result);
+            assert_almost_equal(0.5, result, 0.01); // Example value
+        }
+    }
+
+    // Adding new tests for probability similarities
+    #[test]
+    fn test_js_f32() {
+        let a = &[0.1, 0.9, 0.0];
+        let b = &[0.2, 0.8, 0.0];
+
+        if let Some(result) = ProbabilitySimilarity::jensenshannon(a, b) {
+            println!("The result of js_f32 is {:.8}", result);
+            assert_almost_equal(0.01, result, 0.01); // Example value
+        }
+    }
+
+    #[test]
+    fn test_kl_f32() {
+        let a = &[0.1, 0.9, 0.0];
+        let b = &[0.2, 0.8, 0.0];
+
+        if let Some(result) = ProbabilitySimilarity::kullbackleibler(a, b) {
+            println!("The result of kl_f32 is {:.8}", result);
+            assert_almost_equal(0.036, result, 0.01); // Example value
+        }
+    }
+
+    #[test]
+    fn test_cosine_f16_same() {
+        // Assuming these u16 values represent f16 bit patterns, and they are identical
+        let a_u16: &[u16] = &[15360, 16384, 17408]; // Corresponding to some f16 values
+        let b_u16: &[u16] = &[15360, 16384, 17408]; // Same as above for simplicity
+
+        // Reinterpret cast from &[u16] to &[f16]
+        // SAFETY: This is safe as long as the representations are guaranteed to be identical,
+        // which they are for transparent structs wrapping the same type.
+        let a_f16: &[f16] =
+            unsafe { std::slice::from_raw_parts(a_u16.as_ptr() as *const f16, a_u16.len()) };
+        let b_f16: &[f16] =
+            unsafe { std::slice::from_raw_parts(b_u16.as_ptr() as *const f16, b_u16.len()) };
+
+        if let Some(result) = SimSIMD::cosine(a_f16, b_f16) {
+            println!("The result of cosine_f16 is {:.8}", result);
+            assert_almost_equal(0.0, result, 0.01); // Example value, adjust according to actual expected value
+        }
+    }
+
+    #[test]
+    fn test_cosine_f16_interop() {
+        let a_half: Vec<HalfF16> = vec![1.0, 2.0, 3.0]
+            .iter()
+            .map(|&x| HalfF16::from_f32(x))
+            .collect();
+        let b_half: Vec<HalfF16> = vec![4.0, 5.0, 6.0]
+            .iter()
+            .map(|&x| HalfF16::from_f32(x))
+            .collect();
+
+        // SAFETY: This is safe as long as the memory representations are guaranteed to be identical,
+        // which they are due to both being #[repr(transparent)] wrappers around u16.
+        let a_simsimd: &[f16] =
+            unsafe { std::slice::from_raw_parts(a_half.as_ptr() as *const f16, a_half.len()) };
+        let b_simsimd: &[f16] =
+            unsafe { std::slice::from_raw_parts(b_half.as_ptr() as *const f16, b_half.len()) };
+
+        // Use the reinterpret-casted slices with your SpatialSimilarity implementation
+        if let Some(result) = SpatialSimilarity::cosine(a_simsimd, b_simsimd) {
+            // Expected value might need adjustment depending on actual cosine functionality
+            // Assuming identical vectors yield cosine similarity of 1.0
+            println!("The result of cosine_f16 (interop) is {:.8}", result);
+            assert_almost_equal(0.025, result, 0.01);
         }
     }
 }

package/stringzilla/CONTRIBUTING.md

@@ -136,6 +136,12 @@ cppcheck --project=build_artifacts/compile_commands.json --enable=all
 clang-tidy-11 -p build_artifacts
 ```
 
+I'd recommend putting the following breakpoints:
+
+- `__asan::ReportGenericError` - to detect illegal memory accesses.
+- `__GI_exit` - to stop at exit points - the end of running any executable.
+- `__builtin_unreachable` - to catch all the places where the code is expected to be unreachable.
+
 ### Benchmarking
 
 For benchmarks, you can use the following commands:
@@ -374,6 +380,17 @@ echo "export PATH=/usr/share/swift/usr/bin:$PATH" >> ~/.bashrc
 source ~/.bashrc
 ```
 
+You can check the available images on [`swift.org/download` page](https://www.swift.org/download/#releases).
+For x86 CPUs, the following commands would work:
+
+```bash
+wget https://download.swift.org/swift-5.9.2-release/ubuntu2204/swift-5.9.2-RELEASE/swift-5.9.2-RELEASE-ubuntu22.04.tar.gz
+tar xzf swift-5.9.2-RELEASE-ubuntu22.04.tar.gz
+sudo mv swift-5.9.2-RELEASE-ubuntu22.04 /usr/share/swift
+echo "export PATH=/usr/share/swift/usr/bin:$PATH" >> ~/.bashrc
+source ~/.bashrc
+```
+
 Alternatively, on Linux, the official Swift Docker image can be used for builds and tests:
 
 ```bash

package/stringzilla/Cargo.lock

@@ -19,7 +19,7 @@ checksum = "13e3bf6590cbc649f4d1a3eefc9d5d6eb746f5200ffb04e5e142700b8faa56e7"
 
 [[package]]
 name = "stringzilla"
-version = "3.0.0"
+version = "3.3.1"
 dependencies = [
  "cc",
 ]

package/stringzilla/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "stringzilla"
-version = "3.2.0"
+version = "3.5.0"
 authors = ["Ash Vardanian <1983160+ashvardanian@users.noreply.github.com>"]
 description = "Faster SIMD-accelerated string search, sorting, fingerprints, and edit distances"
 edition = "2021"