Make: Forward SimSIMD settings

Caution: This replaces USES_NATIVE_F16 settings with USES_FP16LIB.
unum-cloud · Nov 9, 2023 · 6f2857f · 6f2857f
1 parent 3578c50
commit 6f2857f
Show file tree

Hide file tree

Showing 9 changed files with 175 additions and 75 deletions.
diff --git a/.vscode/settings.json b/.vscode/settings.json
@@ -136,6 +136,7 @@
     "linalg",
     "longlong",
     "memmap",
+    "MSVC",
     "Napi",
     "ndarray",
     "ndim",

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -19,7 +19,7 @@ option(USEARCH_INSTALL "Install CMake targets" OFF)
 option(USEARCH_USE_OPENMP "Use OpenMP for a thread pool" OFF)
 option(USEARCH_USE_SIMSIMD "Use SimSIMD hardware-accelerated metrics" OFF)
 option(USEARCH_USE_JEMALLOC "Use JeMalloc for faster memory allocations" OFF)
-option(USEARCH_USE_NATIVE_F16 "Use native half-precision types" OFF)
+option(USEARCH_USE_FP16LIB "Use software emulation for half-precision types" OFF)
 
 option(USEARCH_BUILD_TEST_CPP "Compile a native unit test in C++" ${USEARCH_IS_MAIN_PROJECT})
 option(USEARCH_BUILD_BENCH_CPP "Compile a native benchmark in C++" ${USEARCH_IS_MAIN_PROJECT})
@@ -69,7 +69,7 @@ target_compile_definitions(
 
 # Supplementary compilation settings affecting "index_plugins.hpp"
 target_compile_definitions(
-    ${USEARCH_TARGET_NAME} INTERFACE $<$<NOT:$<BOOL:${USEARCH_USE_NATIVE_F16}>>:USEARCH_USE_NATIVE_F16=0>
+    ${USEARCH_TARGET_NAME} INTERFACE $<$<NOT:$<BOOL:${USEARCH_USE_FP16LIB}>>:USEARCH_USE_FP16LIB=0>
 )
 
 target_compile_definitions(
@@ -302,7 +302,7 @@ function (setup_target TARGET_NAME)
 
     # Supplementary compilation settings affecting "index_plugins.hpp"
     target_compile_definitions(
-        ${TARGET_NAME} PRIVATE $<$<NOT:$<BOOL:${USEARCH_USE_NATIVE_F16}>>:USEARCH_USE_NATIVE_F16=0>
+        ${TARGET_NAME} PRIVATE $<$<NOT:$<BOOL:${USEARCH_USE_FP16LIB}>>:USEARCH_USE_FP16LIB=0>
     )
 
     target_compile_definitions(${TARGET_NAME} PRIVATE $<$<NOT:$<BOOL:${USEARCH_USE_SIMSIMD}>>:USEARCH_USE_SIMSIMD=0>)

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -49,7 +49,7 @@ The CMakeLists.txt file has a number of options you can pass:
   - `USEARCH_USE_OPENMP` - use OpenMP for parallelism
   - `USEARCH_USE_SIMSIMD` - use SimSIMD for vectorization
   - `USEARCH_USE_JEMALLOC` - use Jemalloc for memory management
-  - `USEARCH_USE_NATIVE_F16` - use native half-precision floating point
+  - `USEARCH_USE_FP16LIB` - use software emulation for half-precision floating point
 
 Putting all of this together on Ubuntu and compiling the "release" version using the GCC 12 compiler:
 
@@ -58,7 +58,7 @@ cmake \
     -DCMAKE_BUILD_TYPE=Release \
     -DCMAKE_C_COMPILER=gcc-12 \
     -DCMAKE_CXX_COMPILER=g++-12 \
-    -DUSEARCH_USE_NATIVE_F16=1 \
+    -DUSEARCH_USE_FP16LIB=1 \
     -DUSEARCH_USE_OPENMP=1 \
     -DUSEARCH_USE_SIMSIMD=1 \
     -DUSEARCH_USE_JEMALLOC=1 \
@@ -80,7 +80,7 @@ cmake \
     -DCMAKE_BUILD_TYPE=Release \
     -DCMAKE_C_COMPILER="/opt/homebrew/opt/llvm/bin/clang" \
     -DCMAKE_CXX_COMPILER="/opt/homebrew/opt/llvm/bin/clang++" \
-    -DUSEARCH_USE_NATIVE_F16=1 \
+    -DUSEARCH_USE_FP16LIB=1 \
     -DUSEARCH_USE_OPENMP=1 \
     -DUSEARCH_USE_SIMSIMD=1 \
     -DUSEARCH_USE_JEMALLOC=1 \

diff --git a/binding.gyp b/binding.gyp
@@ -6,36 +6,83 @@
             "include_dirs": [
                 "<!@(node -p \"require('node-addon-api').include\")",
                 "include",
-                "<!(echo $USEARCH_USE_SIMSIMD == '1' ? 'simsimd/include' : '')",
-                "<!(echo $USEARCH_USE_NATIVE_F16 == '1' ? 'fp16/include' : '')",
+                "<!(if [ -z \"$USEARCH_USE_SIMSIMD\" ]; then echo 'simsimd/include'; fi)",
+                "<!(if [ -z \"$USEARCH_USE_FP16LIB\" ]; then echo 'fp16/include'; fi)",
             ],
             "dependencies": ["<!(node -p \"require('node-addon-api').gyp\")"],
-            "defines": [
-                "<!(echo $USEARCH_USE_SIMSIMD == '1' ? 'USEARCH_USE_SIMSIMD=1' : 'USEARCH_USE_SIMSIMD=0')",
-                "<!(echo $USEARCH_USE_NATIVE_F16 == '1' ? 'USEARCH_USE_NATIVE_F16=1' : 'USEARCH_USE_NATIVE_F16=0')"
+            "conditions": [
+                [
+                    'OS=="linux"',
+                    {
+                        "cflags_cc": [
+                            "-std=c++17",
+                            "-O3",
+                            "-ffast-math",
+                            "-Wno-unknown-pragmas",
+                            "-fdiagnostics-color=always",
+                            "-fexceptions",
+                            "-g1",
+                            '<!(if [ "$USEARCH_USE_OPENMP" = "1" ]; then echo \'-fopenmp\'; fi)',
+                        ],
+                        "ldflags": [
+                            '<!(if [ "$USEARCH_USE_OPENMP" = "1" ]; then echo \'-lgomp\'; fi)'
+                        ],
+                        "defines": [
+                            "USEARCH_USE_OPENMP=<!(echo ${USEARCH_USE_OPENMP:-1})",
+                            "USEARCH_USE_SIMSIMD=<!(echo ${USEARCH_USE_SIMSIMD:-1})",
+                            "USEARCH_USE_FP16LIB=<!(echo ${USEARCH_USE_FP16LIB:-1})",
+                            "SIMSIMD_TARGET_X86_AVX512=<!(echo ${SIMSIMD_TARGET_X86_AVX512:-1})",
+                            "SIMSIMD_TARGET_ARM_SVE=<!(echo ${SIMSIMD_TARGET_ARM_SVE:-1})",
+                            "SIMSIMD_TARGET_X86_AVX2=<!(echo ${SIMSIMD_TARGET_X86_AVX2:-1})",
+                            "SIMSIMD_TARGET_ARM_NEON=<!(echo ${SIMSIMD_TARGET_ARM_NEON:-1})",
+                        ],
+                    },
+                ],
+                [
+                    'OS=="mac"',
+                    {
+                        "xcode_settings": {"MACOSX_DEPLOYMENT_TARGET": "10.15"},
+                        "cflags_cc": [
+                            "-std=c++17",
+                            "-O3",
+                            "-ffast-math",
+                            "-fcolor-diagnostics",
+                            "-Wno-unknown-pragmas",
+                            "-fexceptions",
+                            "-g1",
+                        ],
+                        "defines": [
+                            "USEARCH_USE_OPENMP=<!(echo ${USEARCH_USE_OPENMP:-0})",
+                            "USEARCH_USE_SIMSIMD=<!(echo ${USEARCH_USE_SIMSIMD:-0})",
+                            "USEARCH_USE_FP16LIB=<!(echo ${USEARCH_USE_FP16LIB:-1})",
+                            "SIMSIMD_TARGET_X86_AVX512=<!(echo ${SIMSIMD_TARGET_X86_AVX512:-0})",
+                            "SIMSIMD_TARGET_ARM_SVE=<!(echo ${SIMSIMD_TARGET_ARM_SVE:-0})",
+                            "SIMSIMD_TARGET_X86_AVX2=<!(echo ${SIMSIMD_TARGET_X86_AVX2:-1})",
+                            "SIMSIMD_TARGET_ARM_NEON=<!(echo ${SIMSIMD_TARGET_ARM_NEON:-1})",
+                        ],
+                    },
+                ],
+                [
+                    'OS=="win"',
+                    {
+                        "msvs_settings": {
+                            "VCCLCompilerTool": {
+                                "Optimization": "2",
+                                "AdditionalOptions": ["/fp:fast", "/W1"],
+                            }
+                        },
+                        "defines": [
+                            "USEARCH_USE_OPENMP=0",
+                            "USEARCH_USE_SIMSIMD=<(USEARCH_USE_SIMSIMD?USEARCH_USE_SIMSIMD:'0')",
+                            "USEARCH_USE_FP16LIB=<(USEARCH_USE_FP16LIB?USEARCH_USE_FP16LIB:'1')",
+                            "SIMSIMD_TARGET_X86_AVX512=<(SIMSIMD_TARGET_X86_AVX512?SIMSIMD_TARGET_X86_AVX512:'0')",
+                            "SIMSIMD_TARGET_ARM_SVE=<(SIMSIMD_TARGET_ARM_SVE?SIMSIMD_TARGET_ARM_SVE:'0')",
+                            "SIMSIMD_TARGET_X86_AVX2=<(SIMSIMD_TARGET_X86_AVX2?SIMSIMD_TARGET_X86_AVX2:'0')",
+                            "SIMSIMD_TARGET_ARM_NEON=<(SIMSIMD_TARGET_ARM_NEON?SIMSIMD_TARGET_ARM_NEON:'0')",
+                        ],
+                    },
+                ],
             ],
-            "cflags": [
-                "-fexceptions",
-                "-Wno-unknown-pragmas",
-                "-Wno-maybe-uninitialized",
-            ],
-            "cflags_cc": [
-                "-fexceptions",
-                "-Wno-unknown-pragmas",
-                "-Wno-maybe-uninitialized",
-                "-std=c++17",
-            ],
-            "xcode_settings": {
-                "GCC_ENABLE_CPP_EXCEPTIONS": "YES",
-                "CLANG_CXX_LIBRARY": "libc++",
-                "MACOSX_DEPLOYMENT_TARGET": "10.15",
-            },
-            "msvs_settings": {
-                "VCCLCompilerTool": {
-                    "ExceptionHandling": 1,
-                    "AdditionalOptions": ["-std:c++17"],
-                }
-            },
         }
     ]
 }
diff --git a/include/usearch/index_plugins.hpp b/include/usearch/index_plugins.hpp
@@ -13,6 +13,10 @@
 
 #include <usearch/index.hpp> // `expected_gt` and macros
 
+#if !defined(USEARCH_USE_OPENMP)
+#define USEARCH_USE_OPENMP 0
+#endif
+
 #if USEARCH_USE_OPENMP
 #include <omp.h> // `omp_get_num_threads()`
 #endif
@@ -21,18 +25,18 @@
 #include <sys/auxv.h> // `getauxval()`
 #endif
 
-#if !defined(USEARCH_USE_NATIVE_F16)
+#if !defined(USEARCH_USE_FP16LIB)
 #if defined(__AVX512F__)
-#define USEARCH_USE_NATIVE_F16 1
+#define USEARCH_USE_FP16LIB 0
 #elif defined(USEARCH_DEFINED_ARM)
 #include <arm_fp16.h> // `__fp16`
-#define USEARCH_USE_NATIVE_F16 1
+#define USEARCH_USE_FP16LIB 0
 #else
-#define USEARCH_USE_NATIVE_F16 0
+#define USEARCH_USE_FP16LIB 1
 #endif
 #endif
 
-#if !USEARCH_USE_NATIVE_F16
+#if USEARCH_USE_FP16LIB
 #include <fp16/fp16.h>
 #endif
 
@@ -41,27 +45,27 @@
 #endif
 
 #if USEARCH_USE_SIMSIMD
+
 // Propagate the `f16` settings
-#define SIMSIMD_NATIVE_F16 USEARCH_USE_NATIVE_F16
-#if defined(USEARCH_DEFINED_LINUX)
+#define SIMSIMD_NATIVE_F16 !USEARCH_USE_FP16LIB
+
+#if !defined(SIMSIMD_TARGET_X86_AVX512) && defined(USEARCH_DEFINED_LINUX)
 #define SIMSIMD_TARGET_X86_AVX512 1
+#endif
+
+#if !defined(SIMSIMD_TARGET_ARM_SVE) && defined(USEARCH_DEFINED_LINUX)
 #define SIMSIMD_TARGET_ARM_SVE 1
+#endif
+
+#if !defined(SIMSIMD_TARGET_X86_AVX2) && (defined(USEARCH_DEFINED_LINUX) || defined(USEARCH_DEFINED_APPLE))
 #define SIMSIMD_TARGET_X86_AVX2 1
+#endif
+
+#if !defined(SIMSIMD_TARGET_ARM_NEON) && (defined(USEARCH_DEFINED_LINUX) || defined(USEARCH_DEFINED_APPLE))
 #define SIMSIMD_TARGET_ARM_NEON 1
-#include <simsimd/simsimd.h>
-#elif defined(USEARCH_DEFINED_APPLE)
-#define SIMSIMD_TARGET_X86_AVX512 0
-#define SIMSIMD_TARGET_ARM_SVE 0
-#define SIMSIMD_TARGET_X86_AVX2 1
-#define SIMSIMD_TARGET_ARM_NEON 1
-#include <simsimd/simsimd.h>
-#else
-#define SIMSIMD_TARGET_X86_AVX512 0
-#define SIMSIMD_TARGET_ARM_SVE 0
-#define SIMSIMD_TARGET_X86_AVX2 0
-#define SIMSIMD_TARGET_ARM_NEON 0
-#include <simsimd/simsimd.h>
 #endif
+
+#include <simsimd/simsimd.h>
 #endif
 
 namespace unum {
@@ -77,7 +81,7 @@ struct uuid_t {
 class f16_bits_t;
 class i8_converted_t;
 
-#if USEARCH_USE_NATIVE_F16
+#if !USEARCH_USE_FP16LIB
 #if defined(USEARCH_DEFINED_ARM)
 using f16_native_t = __fp16;
 #else
@@ -298,7 +302,7 @@ inline expected_gt<metric_kind_t> metric_from_name(char const* name) {
 }
 
 inline float f16_to_f32(std::uint16_t u16) noexcept {
-#if USEARCH_USE_NATIVE_F16
+#if !USEARCH_USE_FP16LIB
     f16_native_t f16;
     std::memcpy(&f16, &u16, sizeof(std::uint16_t));
     return float(f16);
@@ -308,7 +312,7 @@ inline float f16_to_f32(std::uint16_t u16) noexcept {
 }
 
 inline std::uint16_t f32_to_f16(float f32) noexcept {
-#if USEARCH_USE_NATIVE_F16
+#if !USEARCH_USE_FP16LIB
     f16_native_t f16 = f16_native_t(f32);
     std::uint16_t u16;
     std::memcpy(&u16, &f16, sizeof(std::uint16_t));

diff --git a/python/lib.cpp b/python/lib.cpp
@@ -850,7 +850,7 @@ PYBIND11_MODULE(compiled, m) {
 
     m.attr("USES_OPENMP") = py::int_(USEARCH_USE_OPENMP);
     m.attr("USES_SIMSIMD") = py::int_(USEARCH_USE_SIMSIMD);
-    m.attr("USES_NATIVE_F16") = py::int_(USEARCH_USE_NATIVE_F16);
+    m.attr("USES_FP16LIB") = py::int_(USEARCH_USE_FP16LIB);
 
     py::enum_<metric_punned_signature_t>(m, "MetricSignature")
         .value("ArrayArray", metric_punned_signature_t::array_array_k)

diff --git a/python/scripts/cluster.ipynb b/python/scripts/cluster.ipynb
@@ -34,10 +34,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from usearch.index import Index, USES_SIMSIMD, USES_NATIVE_F16\n",
+    "from usearch.index import Index, USES_SIMSIMD, USES_FP16LIB\n",
     "\n",
     "index = Index(ndim=vectors.shape[1], metric=\"cos\", dtype=\"i8\")\n",
-    "index.hardware_acceleration, USES_SIMSIMD, USES_NATIVE_F16"
+    "index.hardware_acceleration, USES_SIMSIMD, USES_FP16LIB"
    ]
   },
   {

diff --git a/python/usearch/index.py b/python/usearch/index.py
@@ -44,7 +44,7 @@
     DEFAULT_EXPANSION_SEARCH,
     USES_OPENMP,
     USES_SIMSIMD,
-    USES_NATIVE_F16,
+    USES_FP16LIB,
 )
 
 MetricKindBitwise = (
@@ -1204,7 +1204,7 @@ def specs(self) -> Dict[str, Union[str, int, bool]]:
             "path": self.path,
             "compiled_with_openmp": USES_OPENMP,
             "compiled_with_simsimd": USES_SIMSIMD,
-            "compiled_with_native_f16": USES_NATIVE_F16,
+            "compiled_with_native_f16": USES_FP16LIB,
         }
 
     def __repr__(self) -> str:
@@ -1242,7 +1242,7 @@ def _repr_pretty_(self, printer, cycle) -> str:
                 "- binary",
                 f"-- uses OpenMP: {USES_OPENMP}",
                 f"-- uses SimSIMD: {USES_SIMSIMD}",
-                f"-- supports half-precision: {USES_NATIVE_F16}",
+                f"-- supports half-precision: {USES_FP16LIB}",
                 f"-- uses hardware acceleration: {self.hardware_acceleration}",
                 "- state",
                 f"-- size: {self.size:,} vectors",