Initial WASM support.

CMakeLists.txt

@@ -40,7 +40,7 @@ set(CMAKE_C_FLAGS_DEBUG "-g -fsanitize=address")
 set(CMAKE_C_FLAGS_RELEASE "-O3")
 
 # Compiler-specific flags
-if (CMAKE_CXX_COMPILER_ID MATCHES "^(Apple)?Clang$")
+if (CMAKE_CXX_COMPILER_ID MATCHES "^(Apple)?Clang$" AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
     if (NOT APPLE)
         add_compile_options(-march=native)
     endif ()
@@ -104,15 +104,25 @@ if (SIMSIMD_BUILD_BENCHMARKS)
 endif ()
 
 if (SIMSIMD_BUILD_TESTS)
-    add_executable(simsimd_test_compile_time scripts/test.c)
-    target_link_libraries(simsimd_test_compile_time simsimd m)
-
-    add_executable(simsimd_test_run_time scripts/test.c c/lib.c)
-    target_compile_definitions(simsimd_test_run_time PRIVATE SIMSIMD_DYNAMIC_DISPATCH=1)
-    target_link_libraries(simsimd_test_run_time simsimd m)
+    if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
+        add_executable(simsimd_test_compile_time scripts/test.c)
+
+        target_compile_definitions(simsimd_test_compile_time PRIVATE _SIMSIMD_TARGET_ARM=1 SIMSIMD_TARGET_WASM=1)
+        target_compile_options(simsimd_test_compile_time PRIVATE -msimd128 -mfpu=neon)
+
+        target_link_libraries(simsimd_test_compile_time simsimd m)
+    else()
+        add_executable(simsimd_test_compile_time scripts/test.c)
+        target_compile_definitions(simsimd_test_compile_time PRIVATE SIMSIMD_TARGET_HASWELL=1)
+        target_link_libraries(simsimd_test_compile_time simsimd m)
+
+        add_executable(simsimd_test_run_time scripts/test.c c/lib.c)
+        target_compile_definitions(simsimd_test_run_time PRIVATE SIMSIMD_DYNAMIC_DISPATCH=1)
+        target_link_libraries(simsimd_test_run_time simsimd m)
+    endif()
 endif ()
 
-if (SIMSIMD_BUILD_SHARED)
+if (SIMSIMD_BUILD_SHARED AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
     set(SIMSIMD_SOURCES ${SIMSIMD_SOURCES} c/lib.c)
     add_library(simsimd_shared SHARED ${SIMSIMD_SOURCES})
     target_include_directories(simsimd_shared PUBLIC "${PROJECT_SOURCE_DIR}/include")

include/simsimd/curved.h

@@ -213,11 +213,10 @@ SIMSIMD_MAKE_COMPLEX_BILINEAR(accurate, bf16c, f64, SIMSIMD_BF16_TO_F32) // sims
 SIMSIMD_MAKE_MAHALANOBIS(accurate, bf16, f64, SIMSIMD_BF16_TO_F32)       // simsimd_mahalanobis_bf16_accurate
 
 #if _SIMSIMD_TARGET_ARM
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd"))), apply_to = function)
-
 SIMSIMD_PUBLIC void simsimd_bilinear_f32_neon(simsimd_f32_t const *a, simsimd_f32_t const *b, simsimd_f32_t const *c,
                                               simsimd_size_t n, simsimd_distance_t *result) {
     float32x4_t sum_vec = vdupq_n_f32(0);

include/simsimd/dot.h

@@ -232,7 +232,7 @@ SIMSIMD_MAKE_COMPLEX_DOT(accurate, bf16c, f64, SIMSIMD_BF16_TO_F32)  // simsimd_
 SIMSIMD_MAKE_COMPLEX_VDOT(accurate, bf16c, f64, SIMSIMD_BF16_TO_F32) // simsimd_vdot_bf16c_accurate
 
 #if _SIMSIMD_TARGET_ARM
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd"))), apply_to = function)
@@ -398,7 +398,7 @@ SIMSIMD_PUBLIC void simsimd_dot_u8_neon(simsimd_u8_t const *a_scalars, simsimd_u
 #pragma GCC pop_options
 #endif // SIMSIMD_TARGET_NEON_I8
 
-#if SIMSIMD_TARGET_NEON_F16
+#if SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd+fp16")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd+fp16"))), apply_to = function)

include/simsimd/elementwise.h

@@ -1942,7 +1942,7 @@ SIMSIMD_PUBLIC void simsimd_fma_u8_sapphire(
 #endif // _SIMSIMD_TARGET_X86
 
 #if _SIMSIMD_TARGET_ARM
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd"))), apply_to = function)

include/simsimd/probability.h

@@ -133,7 +133,7 @@ SIMSIMD_MAKE_KL(accurate, bf16, f64, SIMSIMD_BF16_TO_F32, SIMSIMD_F32_DIVISION_E
 SIMSIMD_MAKE_JS(accurate, bf16, f64, SIMSIMD_BF16_TO_F32, SIMSIMD_F32_DIVISION_EPSILON) // simsimd_js_bf16_accurate
 
 #if _SIMSIMD_TARGET_ARM
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd"))), apply_to = function)

include/simsimd/simsimd.h

@@ -196,7 +196,7 @@ typedef enum {
     simsimd_cap_sve_i8_k = 1 << 27,    ///< ARM SVE `i8` capability
     simsimd_cap_sve2_k = 1 << 28,      ///< ARM SVE2 capability
     simsimd_cap_sve2p1_k = 1 << 29,    ///< ARM SVE2p1 capability
-
+    simsimd_cap_wasm_simd_k = 1 << 30, ///< WASM capability
 } simsimd_capability_t;
 
 /**
@@ -504,6 +504,8 @@ SIMSIMD_PUBLIC simsimd_capability_t _simsimd_capabilities_arm(void) {
         (simsimd_cap_sve2_k * (supports_sve2)) |                                                      //
         (simsimd_cap_sve2p1_k * (supports_sve2p1)) |                                                  //
         (simsimd_cap_serial_k));
+#elif defined(SIMSIMD_TARGET_WASM)
+    return simsimd_cap_wasm_simd_k;
 #else // if !_SIMSIMD_DEFINED_LINUX
     return simsimd_cap_serial_k;
 #endif
@@ -1465,6 +1467,7 @@ SIMSIMD_PUBLIC int simsimd_uses_sve_f16(void) { return _SIMSIMD_TARGET_ARM && SI
 SIMSIMD_PUBLIC int simsimd_uses_sve_bf16(void) { return _SIMSIMD_TARGET_ARM && SIMSIMD_TARGET_SVE_BF16; }
 SIMSIMD_PUBLIC int simsimd_uses_sve_i8(void) { return _SIMSIMD_TARGET_ARM && SIMSIMD_TARGET_SVE_I8; }
 SIMSIMD_PUBLIC int simsimd_uses_sve2(void) { return _SIMSIMD_TARGET_ARM && SIMSIMD_TARGET_SVE2; }
+SIMSIMD_PUBLIC int simsimd_uses_wasm_simd(void) { return _SIMSIMD_TARGET_ARM && SIMSIMD_TARGET_WASM; }
 SIMSIMD_PUBLIC int simsimd_uses_haswell(void) { return _SIMSIMD_TARGET_X86 && SIMSIMD_TARGET_HASWELL; }
 SIMSIMD_PUBLIC int simsimd_uses_skylake(void) { return _SIMSIMD_TARGET_X86 && SIMSIMD_TARGET_SKYLAKE; }
 SIMSIMD_PUBLIC int simsimd_uses_ice(void) { return _SIMSIMD_TARGET_X86 && SIMSIMD_TARGET_ICE; }
@@ -1527,7 +1530,7 @@ SIMSIMD_PUBLIC void simsimd_dot_f16(simsimd_f16_t const *a, simsimd_f16_t const
                                     simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE_F16
     simsimd_dot_f16_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON_F16
+#elif SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
     simsimd_dot_f16_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SAPPHIRE
     simsimd_dot_f16_sapphire(a, b, n, d);
@@ -1553,7 +1556,7 @@ SIMSIMD_PUBLIC void simsimd_dot_f32(simsimd_f32_t const *a, simsimd_f32_t const
                                     simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_dot_f32_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_dot_f32_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_dot_f32_skylake(a, b, n, d);
@@ -1577,7 +1580,7 @@ SIMSIMD_PUBLIC void simsimd_dot_f16c(simsimd_f16c_t const *a, simsimd_f16c_t con
                                      simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE_F16
     simsimd_dot_f16c_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON_F16
+#elif SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
     simsimd_dot_f16c_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SAPPHIRE
     simsimd_dot_f16c_sapphire(a, b, n, d);
@@ -1601,7 +1604,7 @@ SIMSIMD_PUBLIC void simsimd_dot_f32c(simsimd_f32c_t const *a, simsimd_f32c_t con
                                      simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_dot_f32c_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_dot_f32c_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_dot_f32c_skylake(a, b, n, d);
@@ -1625,7 +1628,7 @@ SIMSIMD_PUBLIC void simsimd_vdot_f16c(simsimd_f16c_t const *a, simsimd_f16c_t co
                                       simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_vdot_f16c_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_dot_f16c_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SAPPHIRE
     simsimd_dot_f16c_sapphire(a, b, n, d);
@@ -1649,7 +1652,7 @@ SIMSIMD_PUBLIC void simsimd_vdot_f32c(simsimd_f32c_t const *a, simsimd_f32c_t co
                                       simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_vdot_f32c_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_dot_f32c_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_dot_f32c_skylake(a, b, n, d);
@@ -1711,7 +1714,7 @@ SIMSIMD_PUBLIC void simsimd_cos_f16(simsimd_f16_t const *a, simsimd_f16_t const
                                     simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE_F16
     simsimd_cos_f16_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON_F16
+#elif SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
     simsimd_cos_f16_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SAPPHIRE
     simsimd_cos_f16_sapphire(a, b, n, d);
@@ -1739,7 +1742,7 @@ SIMSIMD_PUBLIC void simsimd_cos_f32(simsimd_f32_t const *a, simsimd_f32_t const
                                     simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_cos_f32_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_cos_f32_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_cos_f32_skylake(a, b, n, d);
@@ -1753,7 +1756,7 @@ SIMSIMD_PUBLIC void simsimd_cos_f64(simsimd_f64_t const *a, simsimd_f64_t const
                                     simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_cos_f64_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_cos_f64_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_cos_f64_skylake(a, b, n, d);
@@ -1789,7 +1792,7 @@ SIMSIMD_PUBLIC void simsimd_l2sq_f16(simsimd_f16_t const *a, simsimd_f16_t const
                                      simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE_F16
     simsimd_l2sq_f16_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON_F16
+#elif SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
     simsimd_l2sq_f16_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SAPPHIRE
     simsimd_l2sq_f16_sapphire(a, b, n, d);
@@ -1817,7 +1820,7 @@ SIMSIMD_PUBLIC void simsimd_l2sq_f32(simsimd_f32_t const *a, simsimd_f32_t const
                                      simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_l2sq_f32_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_l2sq_f32_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_l2sq_f32_skylake(a, b, n, d);
@@ -1831,7 +1834,7 @@ SIMSIMD_PUBLIC void simsimd_l2sq_f64(simsimd_f64_t const *a, simsimd_f64_t const
                                      simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_l2sq_f64_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_l2sq_f64_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_l2sq_f64_skylake(a, b, n, d);
@@ -1935,7 +1938,7 @@ SIMSIMD_PUBLIC void simsimd_hamming_b8(simsimd_b8_t const *a, simsimd_b8_t const
                                        simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_hamming_b8_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_hamming_b8_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_ICE
     simsimd_hamming_b8_ice(a, b, n, d);
@@ -1949,7 +1952,7 @@ SIMSIMD_PUBLIC void simsimd_jaccard_b8(simsimd_b8_t const *a, simsimd_b8_t const
                                        simsimd_distance_t *d) {
 #if SIMSIMD_TARGET_SVE
     simsimd_jaccard_b8_sve(a, b, n, d);
-#elif SIMSIMD_TARGET_NEON
+#elif SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_jaccard_b8_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_ICE
     simsimd_jaccard_b8_ice(a, b, n, d);
@@ -1990,7 +1993,7 @@ SIMSIMD_PUBLIC void simsimd_kl_bf16(simsimd_bf16_t const *a, simsimd_bf16_t cons
 }
 SIMSIMD_PUBLIC void simsimd_kl_f32(simsimd_f32_t const *a, simsimd_f32_t const *b, simsimd_size_t n,
                                    simsimd_distance_t *d) {
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_kl_f32_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_kl_f32_skylake(a, b, n, d);
@@ -2018,7 +2021,7 @@ SIMSIMD_PUBLIC void simsimd_js_bf16(simsimd_bf16_t const *a, simsimd_bf16_t cons
 }
 SIMSIMD_PUBLIC void simsimd_js_f32(simsimd_f32_t const *a, simsimd_f32_t const *b, simsimd_size_t n,
                                    simsimd_distance_t *d) {
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
     simsimd_js_f32_neon(a, b, n, d);
 #elif SIMSIMD_TARGET_SKYLAKE
     simsimd_js_f32_skylake(a, b, n, d);

include/simsimd/spatial.h

@@ -251,7 +251,7 @@ SIMSIMD_MAKE_L2SQ(accurate, bf16, f64, SIMSIMD_BF16_TO_F32) // simsimd_l2sq_bf16
 SIMSIMD_MAKE_L2(accurate, bf16, f64, SIMSIMD_BF16_TO_F32)   // simsimd_l2_bf16_accurate
 
 #if _SIMSIMD_TARGET_ARM
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd"))), apply_to = function)
@@ -395,7 +395,7 @@ SIMSIMD_PUBLIC void simsimd_cos_f64_neon(simsimd_f64_t const *a, simsimd_f64_t c
 #pragma GCC pop_options
 #endif // SIMSIMD_TARGET_NEON
 
-#if SIMSIMD_TARGET_NEON_F16
+#if SIMSIMD_TARGET_NEON_F16 || SIMSIMD_TARGET_WASM
 #pragma GCC push_options
 #pragma GCC target("arch=armv8.2-a+simd+fp16")
 #pragma clang attribute push(__attribute__((target("arch=armv8.2-a+simd+fp16"))), apply_to = function)

include/simsimd/types.h

@@ -227,7 +227,7 @@
 #include <intrin.h>
 #else
 
-#if SIMSIMD_TARGET_NEON
+#if SIMSIMD_TARGET_NEON || SIMSIMD_TARGET_WASM
 #include <arm_neon.h>
 #endif
 

scripts/test.c

@@ -69,6 +69,7 @@ void test_utilities(void) {
     int uses_sapphire = simsimd_uses_sapphire();
     int uses_turin = simsimd_uses_turin();
     int uses_sierra = simsimd_uses_sierra();
+    int uses_wasm = simsimd_uses_wasm_simd();
 
     assert(uses_neon == ((capabilities & simsimd_cap_neon_k) != 0));
     assert(uses_sve == ((capabilities & simsimd_cap_sve_k) != 0));
@@ -79,6 +80,7 @@ void test_utilities(void) {
     assert(uses_sapphire == ((capabilities & simsimd_cap_sapphire_k) != 0));
     assert(uses_turin == ((capabilities & simsimd_cap_turin_k) != 0));
     assert(uses_sierra == ((capabilities & simsimd_cap_sierra_k) != 0));
+    assert(uses_wasm == ((capabilities & simsimd_cap_wasm_simd_k) != 0));
 }
 
 /**