usearch 2.1.2 → 2.1.3

package/.github/workflows/prerelease.yml

@@ -74,8 +74,9 @@ jobs:
     name: Test Python
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
-        os: [ubuntu-20.04, macOS-11, windows-2022]
+        os: [ubuntu-22.04, macOS-11, windows-2022]
         python-version: ["3.11"]
 
     steps:
@@ -89,8 +90,16 @@ jobs:
         run: |
           python -m pip install --upgrade pip
           pip install pytest numpy
+      - name: Build locally on Ubuntu
+        run: |
+          export CC=/usr/bin/gcc-12
+          export CXX=/usr/bin/g++-12
+          export LD_LIBRARY_PATH=/usr/lib/gcc/x86_64-linux-gnu/12/:$LD_LIBRARY_PATH
+          python -m pip install .
+        if: ${{ matrix.os == 'ubuntu-22.04' }}
       - name: Build locally
         run: python -m pip install .
+        if: ${{ matrix.os != 'ubuntu-22.04' }}
       - name: Test with PyTest
         run: pytest python/scripts/
 
@@ -99,8 +108,9 @@ jobs:
     name: Test Python 3.7
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
-        os: [ubuntu-20.04]
+        os: [ubuntu-22.04]
         python-version: ["3.7"]
 
     steps:
@@ -118,7 +128,12 @@ jobs:
           pip install pytest numpy
 
       - name: Build locally
-        run: python -m pip install .
+        run: |
+          export CC=/usr/bin/gcc-12
+          export CXX=/usr/bin/g++-12
+          export LD_LIBRARY_PATH=/usr/lib/gcc/x86_64-linux-gnu/12/:$LD_LIBRARY_PATH
+          python -m pip install .
+        if: ${{ matrix.os == 'ubuntu-22.04' }}
 
       - name: Test with PyTest
         run: pytest python/scripts/

package/CMakeLists.txt

@@ -49,11 +49,26 @@ else()
     target_compile_features(${USEARCH_TARGET_NAME} INTERFACE cxx_std_11)
 endif()
 
+# Core compilation settings affecting "index.hpp"
 target_compile_definitions(
     ${USEARCH_TARGET_NAME}
     INTERFACE
     $<$<NOT:$<BOOL:${JSON_GlobalUDLs}>>:USEARCH_USE_OPENMP=0>
 )
+
+# Supplementary compilation settings affecting "index_plugins.hpp"
+target_compile_definitions(
+    ${USEARCH_TARGET_NAME}
+    INTERFACE
+    $<$<NOT:$<BOOL:${JSON_GlobalUDLs}>>:USEARCH_USE_NATIVE_F16=0>
+)
+
+target_compile_definitions(
+    ${USEARCH_TARGET_NAME}
+    INTERFACE
+    $<$<NOT:$<BOOL:${JSON_GlobalUDLs}>>:USEARCH_USE_SIMSIMD=0>
+)
+
 target_include_directories(
     ${USEARCH_TARGET_NAME}
     ${USEARCH_SYSTEM_INCLUDE} INTERFACE

package/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "usearch"
-version = "2.1.2"
+version = "2.1.3"
 authors = ["Ash Vardanian <1983160+ashvardanian@users.noreply.github.com>"]
 description = "Smaller & Faster Single-File Vector Search Engine from Unum"
 edition = "2021"

package/README.md

@@ -39,8 +39,8 @@ Linux • MacOS • Windows • iOS • Docker • WebAssembly
 <a href="https://pypi.org/project/usearch/"> <img alt="PyPI - Downloads" src="https://img.shields.io/pypi/dm/usearch?label=pypi%20downloads"> </a>
 <a href="https://www.npmjs.com/package/usearch"> <img alt="npm" src="https://img.shields.io/npm/dy/usearch?label=npm%20dowloads"> </a>
 <a href="https://crates.io/crates/usearch"> <img alt="Crates.io" src="https://img.shields.io/crates/d/usearch?label=crate%20downloads"> </a>
-<a href="https://www.nuget.org/packages/Cloud.Unum.USearch"> <img alt="Nuget" src="https://img.shields.io/nuget/dt/Cloud.Unum.USearch?style=social&label=NuGet"> </a>
-<a href="https://hub.docker.com/r/unum/usearch"> <img alt="Docker Pulls" src="https://img.shields.io/docker/pulls/unum/usearch?style=social&label=Docker"> </a>
+<a href="https://www.nuget.org/packages/Cloud.Unum.USearch"> <img alt="Nuget" src="https://img.shields.io/nuget/dt/Cloud.Unum.USearch?label=NuGet"> </a>
+<a href="https://hub.docker.com/r/unum/usearch"> <img alt="Docker Pulls" src="https://img.shields.io/docker/pulls/unum/usearch?label=Docker"> </a>
 <img alt="GitHub code size in bytes" src="https://img.shields.io/github/languages/code-size/unum-cloud/usearch">
 </div>
 

package/conanfile.py

@@ -4,7 +4,7 @@ from conans import ConanFile
 class USearchConan(ConanFile):
 
     name = 'USearch'
-    version = '2.1.2'
+    version = '2.1.3'
     license = 'Apache 2.0'
     url = 'https://github.com/unum-cloud/usearch'
     description = 'Smaller & Faster Single-File Vector Search Engine from Unum'

package/include/usearch/index_plugins.hpp

@@ -41,6 +41,12 @@
 #endif
 
 #if USEARCH_USE_SIMSIMD
+#if __linux__
+#define SIMSIMD_TARGET_ARM_NEON 1
+#define SIMSIMD_TARGET_X86_AVX2 1
+#define SIMSIMD_TARGET_ARM_SVE 1
+#define SIMSIMD_TARGET_X86_AVX512 1
+#endif
 #include <simsimd/simsimd.h>
 #endif
 
@@ -600,9 +606,7 @@ class aligned_allocator_gt {
     using size_type = std::size_t;
     using pointer = element_at*;
     using const_pointer = element_at const*;
-    template <typename other_element_at> struct rebind {
-        using other = aligned_allocator_gt<other_element_at>;
-    };
+    template <typename other_element_at> struct rebind { using other = aligned_allocator_gt<other_element_at>; };
 
     constexpr std::size_t alignment() const { return alignment_ak; }
 
@@ -1384,25 +1388,37 @@ class metric_punned_t {
     // clang-format off
     static metric_punned_t ip_metric_f32_(std::size_t bytes_per_vector) {
         #if USEARCH_USE_SIMSIMD
-        if (hardware_supports(isa_kind_t::sve_k)) return {pun_stl_<f32_t>([=](f32_t const* a, f32_t const* b) { return simsimd_dot_f32_sve(a, b, bytes_per_vector / 4); }), bytes_per_vector, metric_kind_t::ip_k, scalar_kind_t::f32_k, isa_kind_t::sve_k};
+        #if SIMSIMD_TARGET_ARM_NEON
         if (hardware_supports(isa_kind_t::neon_k) && bytes_per_vector % 16 == 0) return {pun_stl_<f32_t>([=](f32_t const* a, f32_t const* b) { return simsimd_dot_f32x4_neon(a, b, bytes_per_vector / 4); }), bytes_per_vector, metric_kind_t::ip_k, scalar_kind_t::f32_k, isa_kind_t::neon_k};
+        #endif
+        #if SIMSIMD_TARGET_ARM_SVE
+        if (hardware_supports(isa_kind_t::sve_k)) return {pun_stl_<f32_t>([=](f32_t const* a, f32_t const* b) { return simsimd_dot_f32_sve(a, b, bytes_per_vector / 4); }), bytes_per_vector, metric_kind_t::ip_k, scalar_kind_t::f32_k, isa_kind_t::sve_k};
+        #endif
+        #if SIMSIMD_TARGET_X86_AVX2
         if (hardware_supports(isa_kind_t::avx2_k) && bytes_per_vector % 16 == 0) return {pun_stl_<f32_t>([=](f32_t const* a, f32_t const* b) { return simsimd_dot_f32x4_avx2(a, b, bytes_per_vector / 4); }), bytes_per_vector, metric_kind_t::ip_k, scalar_kind_t::f32_k, isa_kind_t::avx2_k};
         #endif
+        #endif
         return {to_stl_<metric_ip_gt<f32_t>>(bytes_per_vector), bytes_per_vector, metric_kind_t::ip_k, scalar_kind_t::f32_k, isa_kind_t::auto_k};
     }
 
     static metric_punned_t cos_metric_f16_(std::size_t bytes_per_vector) {
         #if USEARCH_USE_SIMSIMD
+        #if SIMSIMD_TARGET_X86_AVX512
         if (hardware_supports(isa_kind_t::avx512_k) && bytes_per_vector % 32 == 0) return {pun_stl_<simsimd_f16_t>([=](simsimd_f16_t const* a, simsimd_f16_t const* b) { return simsimd_cos_f16x16_avx512(a, b, bytes_per_vector / 2); }), bytes_per_vector, metric_kind_t::cos_k, scalar_kind_t::f16_k, isa_kind_t::avx512_k};
+        #endif
+        #if SIMSIMD_TARGET_ARM_NEON
         if (hardware_supports(isa_kind_t::neon_k) && bytes_per_vector % 8 == 0) return {pun_stl_<simsimd_f16_t>([=](simsimd_f16_t const* a, simsimd_f16_t const* b) { return simsimd_cos_f16x4_neon(a, b, bytes_per_vector / 2); }), bytes_per_vector, metric_kind_t::cos_k, scalar_kind_t::f16_k, isa_kind_t::neon_k};
         #endif
+        #endif
         return {to_stl_<metric_cos_gt<f16_t, f32_t>>(bytes_per_vector), bytes_per_vector, metric_kind_t::cos_k, scalar_kind_t::f16_k, isa_kind_t::auto_k};
     }
 
     static metric_punned_t cos_metric_i8_(std::size_t bytes_per_vector) {
         #if USEARCH_USE_SIMSIMD
+        #if SIMSIMD_TARGET_ARM_NEON
         if (hardware_supports(isa_kind_t::neon_k) && bytes_per_vector % 16 == 0) return {pun_stl_<int8_t>([=](int8_t const* a, int8_t const* b) { return simsimd_cos_i8x16_neon(a, b, bytes_per_vector); }), bytes_per_vector, metric_kind_t::cos_k, scalar_kind_t::i8_k, isa_kind_t::neon_k};
         #endif
+        #endif
         return {to_stl_<cos_i8_t>(bytes_per_vector), bytes_per_vector, metric_kind_t::cos_k, scalar_kind_t::i8_k, isa_kind_t::auto_k};
     }
 

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "usearch",
-    "version": "2.1.2",
+    "version": "2.1.3",
     "description": "Smaller & Faster Single-File Vector Search Engine from Unum",
     "author": "Ash Vardanian",
     "license": "Apache 2.0",

package/setup.py

@@ -7,10 +7,7 @@ from pybind11.setup_helpers import Pybind11Extension
 
 compile_args = []
 link_args = []
-macros_args = [
-    ("USEARCH_USE_NATIVE_F16", "0"),
-    ("USEARCH_USE_SIMSIMD", "0"),
-]
+macros_args = []
 
 if sys.platform == "linux":
     compile_args.append("-std=c++17")
@@ -18,6 +15,8 @@ if sys.platform == "linux":
     compile_args.append("-g")  # Simplify debugging
     compile_args.append("-Wno-unknown-pragmas")
 
+    macros_args.append(("USEARCH_USE_NATIVE_F16", "0"))
+    macros_args.append(("USEARCH_USE_SIMSIMD", "1"))
     macros_args.append(("USEARCH_USE_OPENMP", "1"))
     compile_args.append("-fopenmp")
     link_args.append("-lgomp")

package/simsimd/.github/workflows/prerelease.yml

@@ -21,8 +21,9 @@ jobs:
     name: Test Python
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
-        os: [ubuntu-20.04, macOS-11, windows-2022]
+        os: [ubuntu-22.04] # macOS-11, windows-2022
         python-version: ["3.11"]
 
     steps:
@@ -32,22 +33,54 @@ jobs:
         uses: actions/setup-python@v4
         with:
           python-version: ${{ matrix.python-version }}
+
+      - name: Set up Clang
+        run: brew install llvm
+        if: ${{ matrix.os == 'macOS-11' }}
+
+      - name: Set up MinGW
+        uses: egor-tensin/setup-mingw@v2
+        with:
+          platform: x64
+        if: ${{ matrix.os == 'windows-2022' }}
+        
       - name: Install dependencies
         run: |
           python -m pip install --no-cache-dir --upgrade pip
           pip install --no-cache-dir pytest
-      - name: Build locally
-        run: python -m pip install .
+          
+      - name: Build locally on Ubuntu
+        run: |
+          export CC=/usr/bin/gcc-12
+          export CXX=/usr/bin/g++-12
+          export LD_LIBRARY_PATH=/usr/lib/gcc/x86_64-linux-gnu/12/:$LD_LIBRARY_PATH
+          python -m pip install .
+        if: ${{ matrix.os == 'ubuntu-22.04' }}
+
+      - name: Build locally on MacOS
+        run: |
+          export LDFLAGS="-L/usr/local/opt/llvm/lib"
+          export CPPFLAGS="-I/usr/local/opt/llvm/include"
+          export CC=/usr/local/opt/llvm/bin/clang
+          export CXX=/usr/local/opt/llvm/bin/clang++
+          python -m pip install .
+        if: ${{ matrix.os == 'macOS-11' }}
+
+      - name: Build locally on Windows
+        run: |
+          python -m pip install .
+        if: ${{ matrix.os == 'windows-2022' }}
+        
       - name: Test with PyTest
         run: pytest python/test.py
-
+        
 
   test_python_37:
     name: Test Python 3.7
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-20.04]
+        os: [ubuntu-22.04]
         python-version: ["3.7"]
 
     steps:
@@ -65,9 +98,11 @@ jobs:
           pip install --no-cache-dir pytest
 
       - name: Build locally
-        run: python -m pip install .
-
+        run: |
+          export CC=/usr/bin/gcc-12
+          export CXX=/usr/bin/g++-12
+          export LD_LIBRARY_PATH=/usr/lib/gcc/x86_64-linux-gnu/12/:$LD_LIBRARY_PATH
+          python -m pip install .
+        
       - name: Test with PyTest
         run: pytest python/test.py
-
-

package/simsimd/.github/workflows/release.yml

@@ -29,7 +29,7 @@ jobs:
 
 
   rebase:
-    name: Rebase Development Branch
+    name: Rebase Dev. Branch
     needs: versioning
     runs-on: ubuntu-latest
     steps:
@@ -57,7 +57,7 @@ jobs:
     needs: versioning
     strategy:
       matrix:
-        os: [ubuntu-20.04, macOS-11, windows-2022]
+        os: [ubuntu-20.04] # macOS-11, windows-2022
 
     steps:
       - uses: actions/checkout@v3
@@ -66,13 +66,11 @@ jobs:
       - uses: actions/setup-python@v3
       
       - name: Setup Docker
-        if: matrix.os != 'windows-2022'
-        uses: crazy-max/ghaction-setup-docker@v1.0.0
-        with:
-          version: 23.0.1
-
+        if: matrix.os == 'ubuntu-20.04'
+        uses: docker-practice/actions-setup-docker@master
+        
       - name: Setup QEMU
-        if: matrix.os != 'windows-2022'
+        if: matrix.os == 'ubuntu-20.04'
         uses: docker/setup-qemu-action@v2.1.0
 
       - name: Install CIBuildWheel
@@ -92,7 +90,7 @@ jobs:
     runs-on: ubuntu-20.04
     environment:
       name: pypi
-      url: https://pypi.org/p/stringzilla
+      url: https://pypi.org/p/simsimd
     permissions:
       id-token: write
 

package/simsimd/.vscode/settings.json

@@ -85,13 +85,17 @@
         "variant": "cpp",
         "__nullptr": "cpp",
         "unordered_set": "cpp",
-        "span": "cpp"
+        "span": "cpp",
+        "arm_neon.h": "c"
     },
     "cSpell.words": [
         "dtype",
         "dtypes",
         "kann",
+        "maccs",
+        "simd",
         "simsimd",
+        "tanimoto",
         "unsw"
     ]
 }

package/simsimd/CMakeLists.txt

@@ -34,14 +34,17 @@ endif()
 
 set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -g -fsanitize=address")
 set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O3")
+set(CMAKE_C_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -g -fsanitize=address")
+set(CMAKE_C_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O3")
 
 # List of all possible compiler IDs:
 # https://cmake.org/cmake/help/latest/variable/CMAKE_LANG_COMPILER_ID.html
 if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
 elseif(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pedantic")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fmax-errors=1")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pedantic -fmax-errors=1")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=native")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -pedantic -fmax-errors=1")
 elseif(CMAKE_CXX_COMPILER_ID STREQUAL "NVIDIA" OR CMAKE_CXX_COMPILER_ID STREQUAL "NVHPC")
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr --extended-lambda")
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -gencode arch=compute_86,code=sm_86")

package/simsimd/README.md

@@ -1,8 +1,8 @@
-# SimSIMD
+# SimSIMD 📏
 
-SIMD-accelerated similarity measures, metrics, distance functions for x86 and Arm.
-Tuned for Machine Learning applications, mid-size vectors with 100-1024 dimensions.
-For Cosine (Angular) distance, the most common metric in AI, one can expect following performance.
+SIMD-accelerated similarity measures, metrics, and distance functions for x86 and Arm.
+They are tuned for Machine Learning applications and mid-size vectors with 100-1024 dimensions.
+One can expect the following performance for Cosine (Angular) distance, the most common metric in AI.
 
 | Method | Vectors | Any Length | Speed on 256b | Speed on 1024b |
 | :----- | :------ | :--------- | ------------: | -------------: |
@@ -12,7 +12,7 @@ For Cosine (Angular) distance, the most common metric in AI, one can expect foll
 | NEON   | `f16`   | ❌          |       16 GB/s |        18 GB/s |
 
 > The benchmarks were done on Arm-based "Graviton 3" CPUs powering AWS `c7g.metal` instances.
-> We only use Arm NEON implementation with vectors lengths that are multiples of 128 bits, avoiding any additional head or tail `for` loops for misaligned data.
+> We only use Arm NEON implementation with vector lengths multiples of 128 bits, avoiding additional head or tail `for` loops for misaligned data.
 > By default, we use GCC12, `-O3`, `-march=native` for benchmarks.
 > Serial versions imply auto-vectorization pragmas.
 
@@ -28,7 +28,7 @@ FetchContent_MakeAvailable(simsimd)
 include_directories(${simsimd_SOURCE_DIR}/include)
 ```
 
-Want to use in Python with [USearch](https://github.com/unum-cloud/usearch)?
+Want to use it in Python with [USearch](https://github.com/unum-cloud/usearch)?
 
 ```py
 from usearch import Index, CompiledMetric, MetricKind, MetricSignature
@@ -45,60 +45,81 @@ index = Index(256, metric=metric)
 
 ## Available Metrics
 
-In the C99 interface all function are prepended with `simsimd_` namespace prefix.
-Signature defines the number of arguments:
+In the C99 interface, all functions are prepended with the `simsimd_` namespace prefix.
+The signature defines the number of arguments:
 
 - two pointers, and length,
 - two pointers.
 
-The latter is obviously intended for cases, where the number of dimensions is hard-coded.
-Constraints define the limitations on the number of dimensions that an argument vector can have.
-
-| Name                    | Signature | ISA Extesion |  Constraints   |
-| :---------------------- | :-------: | :----------: | :------------: |
-| `dot_f32_sve`           |    ✳️✳️📏    |   Arm SVE    |                |
-| `dot_f32x4_neon`        |    ✳️✳️📏    |   Arm NEON   |  `d % 4 == 0`  |
-| `cos_f32_sve`           |    ✳️✳️📏    |   Arm SVE    |                |
-| `cos_f16_sve`           |    ✳️✳️📏    |   Arm SVE    |                |
-| `cos_f16x4_neon`        |    ✳️✳️📏    |   Arm NEON   |  `d % 4 == 0`  |
-| `cos_i8x16_neon`        |    ✳️✳️📏    |   Arm NEON   | `d % 16 == 0`  |
-| `cos_f32x4_neon`        |    ✳️✳️📏    |   Arm NEON   |  `d % 4 == 0`  |
-| `cos_f16x16_avx512`     |    ✳️✳️📏    | x86 AVX-512  | `d % 16 == 0`  |
-| `cos_f32x4_avx2`        |    ✳️✳️📏    |   x86 AVX2   |  `d % 4 == 0`  |
-| `l2sq_f32_sve`          |    ✳️✳️📏    |   Arm SVE    |                |
-| `l2sq_f16_sve`          |    ✳️✳️📏    |   Arm SVE    |                |
-| `hamming_b1x8_sve`      |    ✳️✳️📏    |   Arm SVE    |  `d % 8 == 0`  |
-| `hamming_b1x128_sve`    |    ✳️✳️📏    |   Arm SVE    | `d % 128 == 0` |
-| `hamming_b1x128_avx512` |    ✳️✳️📏    | x86 AVX-512  | `d % 128 == 0` |
-| `tanimoto_b1x8_naive`   |    ✳️✳️📏    |              |  `d % 8 == 0`  |
-| `tanimoto_maccs_naive`  |    ✳️✳️     |              |   `d == 166`   |
-| `tanimoto_maccs_neon`   |    ✳️✳️     |   Arm NEON   |   `d == 166`   |
-| `tanimoto_maccs_sve`    |    ✳️✳️     |   Arm SVE    |   `d == 166`   |
-| `tanimoto_maccs_avx512` |    ✳️✳️     | x86 AVX-512  |   `d == 166`   |
+The latter is intended for cases where the number of dimensions is hard-coded.
+Constraints define the limitations on the number of dimensions an argument vector can have.
+
+| Name                    | Signature | ISA Extension |  Constraints   |
+| :---------------------- | :-------: | :-----------: | :------------: |
+| `dot_f32_sve`           |    ✳️✳️📏    |    Arm SVE    |                |
+| `dot_f32x4_neon`        |    ✳️✳️📏    |   Arm NEON    |  `d % 4 == 0`  |
+| `cos_f32_sve`           |    ✳️✳️📏    |    Arm SVE    |                |
+| `cos_f16_sve`           |    ✳️✳️📏    |    Arm SVE    |                |
+| `cos_f16x4_neon`        |    ✳️✳️📏    |   Arm NEON    |  `d % 4 == 0`  |
+| `cos_i8x16_neon`        |    ✳️✳️📏    |   Arm NEON    | `d % 16 == 0`  |
+| `cos_f32x4_neon`        |    ✳️✳️📏    |   Arm NEON    |  `d % 4 == 0`  |
+| `cos_f16x16_avx512`     |    ✳️✳️📏    |  x86 AVX-512  | `d % 16 == 0`  |
+| `cos_f32x4_avx2`        |    ✳️✳️📏    |   x86 AVX2    |  `d % 4 == 0`  |
+| `l2sq_f32_sve`          |    ✳️✳️📏    |    Arm SVE    |                |
+| `l2sq_f16_sve`          |    ✳️✳️📏    |    Arm SVE    |                |
+| `hamming_b1x8_sve`      |    ✳️✳️📏    |    Arm SVE    |  `d % 8 == 0`  |
+| `hamming_b1x128_sve`    |    ✳️✳️📏    |    Arm SVE    | `d % 128 == 0` |
+| `hamming_b1x128_avx512` |    ✳️✳️📏    |  x86 AVX-512  | `d % 128 == 0` |
+| `tanimoto_b1x8_naive`   |    ✳️✳️📏    |               |  `d % 8 == 0`  |
+| `tanimoto_maccs_naive`  |    ✳️✳️     |               |   `d == 166`   |
+| `tanimoto_maccs_neon`   |    ✳️✳️     |   Arm NEON    |   `d == 166`   |
+| `tanimoto_maccs_sve`    |    ✳️✳️     |    Arm SVE    |   `d == 166`   |
+| `tanimoto_maccs_avx512` |    ✳️✳️     |  x86 AVX-512  |   `d == 166`   |
 
 ## Benchmarks
 
 The benchmarks are repeated for every function with a different number of cores involved.
-Light-weight distance functions would be memory bound, implying that multi-core performance may be lower, if the bus bandwidth is not enough to saturate all the cores.
+Light-weight distance functions would be memory bound, implying that multi-core performance may be lower if the bus bandwidth cannot saturate all the cores.
 Similarly, heavy-weight distance functions running on all cores may result in CPU frequency downclocking.
-This is well illustrated by the single-core performance of `i9-13950HX`, equipped with DDR5 memory.
-
-| Method                 | Threads | Speed on 1024b |
-| :--------------------- | ------: | -------------: |
-| `tanimoto_maccs_naive` |       1 |       2.8 Gb/s |
-| `tanimoto_maccs_naive` |      32 |       1.2 Gb/s |
-| `cos_f32_naive`        |       1 |      15.3 Gb/s |
-| `cos_f32_naive`        |      32 |       4.5 Gb/s |
-| `dot_f32x4_avx2`       |       1 |      96.2 Gb/s |
-| `dot_f32x4_avx2`       |      32 |      23.6 Gb/s |
-| `cos_f32x4_avx2`       |       1 |      56.3 Gb/s |
-| `cos_f32x4_avx2`       |      32 |      13.9 Gb/s |
-
-To replicate on your hardware, please run following on Linux:
+This is well illustrated by the single-core performance of the Intel `i9-13950HX`, equipped with DDR5 memory.
+
+| Method                 | Threads | Vector Size |     Speed |
+| :--------------------- | ------: | ----------: | --------: |
+| `dot_f32x4_avx2`       |       1 |      1024 b | 96.2 GB/s |
+| `dot_f32x4_avx2`       |      32 |      1024 b | 23.6 GB/s |
+| `cos_f32_naive`        |       1 |      1024 b | 15.3 GB/s |
+| `cos_f32_naive`        |      32 |      1024 b |  4.5 GB/s |
+| `cos_f32x4_avx2`       |       1 |      1024 b | 56.3 GB/s |
+| `cos_f32x4_avx2`       |      32 |      1024 b | 13.9 GB/s |
+| `tanimoto_maccs_naive` |       1 |        21 b |  2.8 GB/s |
+| `tanimoto_maccs_naive` |      32 |        21 b |  1.2 GB/s |
+
+Switching to the Intel Sapphire Rapids server platform, we can also evaluate some of the AVX-512 extensions, including `VPOPCNTDQ` and `F16`.
+
+| Method                  | Threads | Vector Size |      Speed |
+| :---------------------- | ------: | ----------: | ---------: |
+| `dot_f32x4_avx2`        |       1 |      1024 b |  57.8 GB/s |
+| `dot_f32x4_avx2`        |     224 |      1024 b |  16.1 GB/s |
+| `cos_f32_naive`         |       1 |      1024 b |  10.7 GB/s |
+| `cos_f32_naive`         |     224 |      1024 b |   3.0 GB/s |
+| `cos_f32x4_avx2`        |       1 |      1024 b |  39.5 GB/s |
+| `cos_f32x4_avx2`        |     224 |      1024 b |  15.1 GB/s |
+| `cos_f16x16_avx512`     |       1 |      1024 b |  50.6 GB/s |
+| `cos_f16x16_avx512`     |     224 |      1024 b |  15.9 GB/s |
+| `hamming_b1x128_avx512` |       1 |      1024 b | 790.3 GB/s |
+| `hamming_b1x128_avx512` |     224 |      1024 b | 259.3 GB/s |
+| `tanimoto_maccs_naive`  |       1 |        21 b |   3.0 GB/s |
+| `tanimoto_maccs_naive`  |     224 |        21 b |   1.3 GB/s |
+| `tanimoto_maccs_avx512` |       1 |        21 b |  13.1 GB/s |
+| `tanimoto_maccs_avx512` |     224 |        21 b |   3.7 GB/s |
+
+To replicate this on your hardware, please run the following on Linux:
 
 ```sh
 git clone https://github.com/ashvardanian/SimSIMD.git && cd SimSIMD
-cmake -DCMAKE_BUILD_TYPE=Release -DSIMSIMD_BUILD_BENCHMARKS=1 -B ./build && make -C ./build && ./build/simsimd_bench
+cmake -DCMAKE_BUILD_TYPE=Release -DSIMSIMD_BUILD_BENCHMARKS=1 \
+    -DCMAKE_CXX_COMPILER="g++-12" -DCMAKE_C_COMPILER="gcc-12" \
+    -B ./build && make -C ./build && ./build/simsimd_bench
 ```
 
 MacOS:

package/simsimd/bench.cxx

@@ -100,7 +100,7 @@ int main(int argc, char** argv) {
     register_<std::uint8_t, 21>("tanimoto_maccs_naive", simsimd_tanimoto_maccs_naive);
     register_<simsimd_f32_t>("cos_f32_naive", cos_f32_naive);
 
-#if defined(__ARM_FEATURE_SVE)
+#if SIMSIMD_TARGET_ARM_SVE
     register_<simsimd_f32_t>("dot_f32_sve", simsimd_dot_f32_sve);
     register_<simsimd_f32_t>("cos_f32_sve", simsimd_cos_f32_sve);
     register_<std::int16_t>("cos_f16_sve", simsimd_cos_f16_sve);
@@ -111,27 +111,26 @@ int main(int argc, char** argv) {
     register_<std::uint8_t, 21>("tanimoto_maccs_sve", simsimd_tanimoto_maccs_sve);
 #endif
 
-#if defined(__ARM_NEON)
+#if SIMSIMD_TARGET_ARM_NEON
     register_<simsimd_f32_t>("dot_f32x4_neon", simsimd_dot_f32x4_neon);
     register_<std::int16_t>("cos_f16x4_neon", simsimd_cos_f16x4_neon);
     register_<std::int8_t>("cos_i8x16_neon", simsimd_cos_i8x16_neon);
+    register_<std::int8_t>("l2sq_i8x16_neon", simsimd_l2sq_i8x16_neon);
     register_<simsimd_f32_t>("cos_f32x4_neon", simsimd_cos_f32x4_neon);
     register_<std::uint8_t, 21>("tanimoto_maccs_neon", simsimd_tanimoto_maccs_neon);
 #endif
 
-#if defined(__AVX2__)
+#if SIMSIMD_TARGET_X86_AVX2
     register_<simsimd_f32_t>("dot_f32x4_avx2", simsimd_dot_f32x4_avx2);
     register_<simsimd_f32_t>("cos_f32x4_avx2", simsimd_cos_f32x4_avx2);
 #endif
 
-#if defined(__AVX512F__)
+#if SIMSIMD_TARGET_X86_AVX512
     register_<std::int16_t>("cos_f16x16_avx512", simsimd_cos_f16x16_avx512);
     register_<std::uint8_t>("hamming_b1x128_avx512", simsimd_hamming_b1x128_avx512);
     register_<std::uint8_t, 21>("tanimoto_maccs_avx512", simsimd_tanimoto_maccs_avx512);
 #endif
 
-
-
     bm::RunSpecifiedBenchmarks();
     bm::Shutdown();
     return 0;