numkong 7.0.0 → 7.4.1

package/probes/arm_sve_sdot.c

@@ -0,0 +1,21 @@
+/* NumKong ISA probe: SVE I8 signed-dot (FEAT_SVEDot) */
+#if defined(_WIN32)
+#error "SVE/SME not supported on Windows ARM"
+#endif
+
+#if defined(__APPLE__) && defined(__aarch64__)
+#error "SVE not available on Apple Silicon"
+#endif
+
+#if !defined(__ARM_FEATURE_SVE)
+#error "Feature not available"
+#endif
+#include <arm_sve.h>
+int test_svesdot(void) {
+    svint32_t acc = svdup_s32(0);
+    svint8_t a = svdup_s8(1);
+    svint8_t b = svdup_s8(1);
+    acc = svdot_s32(acc, a, b);
+    return (int)svaddv_s32(svptrue_b32(), acc) >= 0 ? 0 : 1;
+}
+int main(void) { return test_svesdot(); }

package/probes/loongarch_lasx.c

@@ -0,0 +1,12 @@
+/* NumKong ISA probe: LoongArch LASX (256-bit SIMD) */
+#if !defined(__loongarch_asx)
+#error "Feature not available"
+#endif
+#include <lasxintrin.h>
+int main(void) {
+    __m256i a = __lasx_xvreplgr2vr_w(1);
+    __m256i b = __lasx_xvreplgr2vr_w(2);
+    __m256i c = __lasx_xvadd_w(a, b);
+    int r = __lasx_xvpickve2gr_w(c, 0);
+    return r == 3 ? 0 : 1;
+}

package/probes/power_vsx.c

@@ -0,0 +1,12 @@
+/* NumKong ISA probe: Power VSX (POWER9+ 128-bit SIMD) */
+#if !defined(__VSX__)
+#error "Feature not available"
+#endif
+#include <altivec.h>
+int main(void) {
+    __vector float a = vec_splats(1.0f);
+    __vector float b = vec_splats(2.0f);
+    __vector float c = vec_madd(a, b, a);
+    /* vec_extract requires POWER9+ */
+    return vec_extract(c, 0) == 3.0f ? 0 : 1;
+}

package/probes/probe.js

@@ -0,0 +1,127 @@
+#!/usr/bin/env node
+/**
+ * NumKong ISA probe script for Node.js / node-gyp builds.
+ *
+ * Try-compiles each probe .c file from probes/ to determine which ISA
+ * extensions the current compiler supports. Writes results to
+ * build/nk_probes.h as #define NK_TARGET_FOO 1/0.
+ *
+ * Usage: node scripts/probe_isa.js
+ * Called automatically via package.json "preinstall" hook.
+ */
+
+const { execSync } = require("child_process");
+const fs = require("fs");
+const path = require("path");
+const os = require("os");
+
+const cc = process.env.CC || (process.platform === "win32" ? "cl.exe" : "cc");
+const isWin = process.platform === "win32";
+
+/** Try to compile a probe file. Returns true if compilation succeeds. */
+function probeIsa(probeFile, flags) {
+    const tmpObj = path.join(
+        os.tmpdir(),
+        `nk_probe_${path.basename(probeFile, ".c")}${isWin ? ".obj" : ".o"}`,
+    );
+    try {
+        const cmd = isWin
+            ? `"${cc}" /c ${flags.join(" ")} "${probeFile}" /Fo"${tmpObj}" /nologo`
+            : `"${cc}" -c ${flags.join(" ")} "${probeFile}" -o "${tmpObj}" 2>/dev/null`;
+        execSync(cmd, { stdio: "pipe", timeout: 30000 });
+        return true;
+    } catch {
+        return false;
+    } finally {
+        try {
+            fs.unlinkSync(tmpObj);
+        } catch { }
+    }
+}
+
+// Probe table: [define, probeFile, gccFlags, msvcFlags]
+// x86 probes: GCC flags are minimal — each implies its prerequisites.
+// E.g., -mavx512vnni implies -mavx512f; -mavxvnni implies -mavx2.
+const PROBES = [
+    // x86
+    ["NK_TARGET_HASWELL", "probes/x86_haswell.c", ["-mavx2", "-mfma", "-mf16c"], ["/arch:AVX2"]],
+    ["NK_TARGET_SKYLAKE", "probes/x86_skylake.c", ["-mavx512f", "-mavx512bw", "-mavx512dq", "-mavx512vl"], ["/arch:AVX512"]],
+    ["NK_TARGET_ICELAKE", "probes/x86_icelake.c", ["-mavx512vnni", "-mavx512vl"], ["/arch:AVX512"]],
+    ["NK_TARGET_GENOA", "probes/x86_genoa.c", ["-mavx512bf16", "-mavx512vl"], ["/arch:AVX512"]],
+    ["NK_TARGET_SAPPHIRE", "probes/x86_sapphire.c", ["-mavx512fp16", "-mavx512vl"], ["/arch:AVX512"]],
+    ["NK_TARGET_SAPPHIREAMX", "probes/x86_sapphireamx.c", ["-mamx-tile", "-mamx-int8"], ["/arch:AVX512"]],
+    ["NK_TARGET_GRANITEAMX", "probes/x86_graniteamx.c", ["-mamx-tile", "-mamx-fp16"], ["/arch:AVX512"]],
+    ["NK_TARGET_DIAMOND", "probes/x86_diamond.c", ["-mavx10.2-512"], ["/arch:AVX10.2"]],
+    ["NK_TARGET_TURIN", "probes/x86_turin.c", ["-mavx512vp2intersect"], ["/arch:AVX512"]],
+    ["NK_TARGET_ALDER", "probes/x86_alder.c", ["-mavxvnni"], ["/arch:AVX2"]],
+    ["NK_TARGET_SIERRA", "probes/x86_sierra.c", ["-mavxvnniint8"], ["/arch:AVX2"]],
+    // ARM NEON base probes — msvc_flags are empty because MSVC does not define
+    // __ARM_FEATURE_* macros via /arch: flags. For MSVC header-only builds,
+    // types.h infers features from __ARM_ARCH level instead.
+    ["NK_TARGET_NEON", "probes/arm_neon.c", ["-march=armv8-a+simd"], []], // FEAT_AdvSIMD
+    ["NK_TARGET_NEONHALF", "probes/arm_neon_half.c", ["-march=armv8.2-a+simd+fp16"], ["/arch:armv8.2"]], // FEAT_FP16
+    ["NK_TARGET_NEONSDOT", "probes/arm_neon_sdot.c", ["-march=armv8.2-a+dotprod"], ["/arch:armv8.4"]], // FEAT_DotProd
+    ["NK_TARGET_NEONBFDOT", "probes/arm_neon_bfdot.c", ["-march=armv8.6-a+simd+bf16"], ["/arch:armv8.6"]], // FEAT_BF16
+    ["NK_TARGET_NEONFHM", "probes/arm_neon_fhm.c", ["-march=armv8.2-a+simd+fp16+fp16fml"], ["/arch:armv8.4"]], // FEAT_FHM
+    // ARM SVE/SME
+    ["NK_TARGET_SVE", "probes/arm_sve.c", ["-march=armv8.2-a+sve"], []],
+    ["NK_TARGET_SVEHALF", "probes/arm_sve_half.c", ["-march=armv8.2-a+sve+fp16"], []],
+    ["NK_TARGET_SVEBFDOT", "probes/arm_sve_bfdot.c", ["-march=armv8.2-a+sve+bf16"], []],
+    ["NK_TARGET_SVESDOT", "probes/arm_sve_sdot.c", ["-march=armv8.2-a+sve+dotprod"], []],
+    ["NK_TARGET_SVE2", "probes/arm_sve2.c", ["-march=armv8.2-a+sve2"], []],
+    ["NK_TARGET_SVE2P1", "probes/arm_sve2p1.c", ["-march=armv8.2-a+sve2p1"], []],
+    ["NK_TARGET_NEONFP8", "probes/arm_neonfp8.c", ["-march=armv8-a+simd+fp8dot4"], []],
+    ["NK_TARGET_SME", "probes/arm_sme.c", ["-march=armv8-a+sme"], []],
+    ["NK_TARGET_SME2", "probes/arm_sme2.c", ["-march=armv8-a+sme2"], []],
+    ["NK_TARGET_SME2P1", "probes/arm_sme2p1.c", ["-march=armv8-a+sme2p1"], []],
+    ["NK_TARGET_SMEF64", "probes/arm_sme_f64.c", ["-march=armv8-a+sme+sme-f64f64"], []],
+    ["NK_TARGET_SMEHALF", "probes/arm_sme_half.c", ["-march=armv8-a+sme+sme-f16f16"], []],
+    ["NK_TARGET_SMEBF16", "probes/arm_sme_bf16.c", ["-march=armv8-a+sme2+b16b16"], []],
+    ["NK_TARGET_SMEBI32", "probes/arm_sme_bi32.c", ["-march=armv8-a+sme2+sme-i16i32"], []],
+    ["NK_TARGET_SMELUT2", "probes/arm_sme_lut2.c", ["-march=armv8-a+sme2+lut"], []],
+    ["NK_TARGET_SMEFA64", "probes/arm_sme_fa64.c", ["-march=armv8-a+sme+sme-fa64"], []],
+    // RISC-V
+    ["NK_TARGET_RVV", "probes/riscv_rvv.c", ["-march=rv64gcv"], []],
+    ["NK_TARGET_RVVHALF", "probes/riscv_rvv_half.c", ["-march=rv64gcv_zvfh"], []],
+    ["NK_TARGET_RVVBF16", "probes/riscv_rvv_bf16.c", ["-march=rv64gcv_zvfbfwma"], []],
+    ["NK_TARGET_RVVBB", "probes/riscv_rvv_bb.c", ["-march=rv64gcv_zvbb"], []],
+    // LoongArch
+    ["NK_TARGET_LOONGSONASX", "probes/loongarch_lasx.c", ["-mlasx"], []],
+    // Power
+    ["NK_TARGET_POWERVSX", "probes/power_vsx.c", ["-mcpu=power9", "-mvsx"], []],
+    // WASM
+    ["NK_TARGET_V128RELAXED", "probes/wasm_v128relaxed.c", ["-mrelaxed-simd"], []],
+];
+
+function main() {
+    const pkgRoot = path.join(__dirname, "..");
+    fs.mkdirSync(pkgRoot, { recursive: true });
+
+    const arch = process.arch; // 'x64', 'arm64', etc.
+    const lines = [
+        "/* Auto-generated by scripts/probe_isa.js — do not edit */",
+        `/* Compiler: ${cc}, Platform: ${process.platform}, Arch: ${arch} */`,
+        "",
+    ];
+
+    let enabled = 0;
+    for (const [define, probeFile, gccFlags, msvcFlags] of PROBES) {
+        const flags = isWin ? msvcFlags : gccFlags;
+        const supported = probeIsa(path.join(pkgRoot, probeFile), flags);
+        lines.push(`#define ${define} ${supported ? 1 : 0}`);
+        if (supported) {
+            enabled++;
+            console.log(`[NumKong] Probe ${define}: supported`);
+        }
+    }
+
+    lines.push("");
+    const header = lines.join("\n");
+    const outPath = path.join(pkgRoot, "nk_probes.h");
+    fs.writeFileSync(outPath, header);
+    console.log(
+        `[NumKong] Wrote ${outPath} (${enabled} ISAs enabled out of ${PROBES.length})`,
+    );
+}
+
+main();

package/probes/riscv_rvv.c

@@ -0,0 +1,14 @@
+/* NumKong ISA probe: RVV 1.0 (RISC-V Vector Extension) */
+#if !defined(__riscv_v)
+#error "Feature not available"
+#endif
+#include <riscv_vector.h>
+int main(void) {
+    size_t vl = __riscv_vsetvl_e32m1(4);
+    vfloat32m1_t a = __riscv_vfmv_v_f_f32m1(1.0f, vl);
+    vfloat32m1_t b = __riscv_vfmv_v_f_f32m1(2.0f, vl);
+    vfloat32m1_t c = __riscv_vfadd_vv_f32m1(a, b, vl);
+    vfloat32m1_t sum = __riscv_vfredusum_vs_f32m1_f32m1(c, __riscv_vfmv_v_f_f32m1(0.0f, vl), vl);
+    float result = __riscv_vfmv_f_s_f32m1_f32(sum);
+    return result > 0.0f ? 0 : 1;
+}

package/probes/riscv_rvv_bb.c

@@ -0,0 +1,15 @@
+/* NumKong ISA probe: RVV Zvbb (basic bit-manipulation) */
+#if !defined(__riscv_zvbb)
+#error "Feature not available"
+#endif
+#include <riscv_vector.h>
+int main(void) {
+    size_t vl = __riscv_vsetvl_e8m1(4);
+    vuint8m1_t a = __riscv_vmv_v_x_u8m1(0xFF, vl);
+    /* vcpop.v — per-element popcount, the key Zvbb instruction */
+    vuint8m1_t popcnt = __riscv_vcpop_v_u8m1(a, vl);
+    vuint8m1_t sum = __riscv_vredsum_vs_u8m1_u8m1(popcnt, __riscv_vmv_v_x_u8m1(0, 1), vl);
+    unsigned char result = __riscv_vmv_x_s_u8m1_u8(sum);
+    /* Each lane is 0xFF → popcount 8, sum of 4 lanes = 32 */
+    return result == 32 ? 0 : 1;
+}

package/probes/riscv_rvv_bf16.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: RVV Zvfbfwma (BF16 widening FMA) */
+#if !defined(__riscv_zvfbfwma)
+#error "Feature not available"
+#endif
+#include <riscv_vector.h>
+int main(void) {
+    size_t vl = __riscv_vsetvl_e16m1(4);
+    vuint16m1_t raw = __riscv_vmv_v_x_u16m1(0x3F80, vl); /* bf16 1.0 */
+    vbfloat16m1_t a = __riscv_vreinterpret_v_u16m1_bf16m1(raw);
+    size_t vl32 = __riscv_vsetvl_e32m2(4);
+    vfloat32m2_t acc = __riscv_vfmv_v_f_f32m2(0.0f, vl32);
+    acc = __riscv_vfwmaccbf16_vv_f32m2(acc, a, a, vl);
+    vfloat32m1_t sum = __riscv_vfredusum_vs_f32m2_f32m1(acc, __riscv_vfmv_v_f_f32m1(0.0f, __riscv_vsetvl_e32m1(1)),
+                                                        vl32);
+    float result = __riscv_vfmv_f_s_f32m1_f32(sum);
+    return result > 0.0f ? 0 : 1;
+}

package/probes/riscv_rvv_half.c

@@ -0,0 +1,14 @@
+/* NumKong ISA probe: RVV Zvfh (half-precision vector) */
+#if !defined(__riscv_zvfh)
+#error "Feature not available"
+#endif
+#include <riscv_vector.h>
+int main(void) {
+    size_t vl = __riscv_vsetvl_e16m1(4);
+    vfloat16m1_t a = __riscv_vfmv_v_f_f16m1((_Float16)1.0f, vl);
+    vfloat32m2_t wide = __riscv_vfwcvt_f_f_v_f32m2(a, vl);
+    vfloat32m1_t sum = __riscv_vfredusum_vs_f32m2_f32m1(wide, __riscv_vfmv_v_f_f32m1(0.0f, __riscv_vsetvl_e32m1(1)),
+                                                        vl);
+    float result = __riscv_vfmv_f_s_f32m1_f32(sum);
+    return result > 0.0f ? 0 : 1;
+}

package/probes/wasm_v128relaxed.c

@@ -0,0 +1,11 @@
+/* NumKong ISA probe: WASM Relaxed SIMD (v128) */
+#if !defined(__wasm_relaxed_simd__)
+#error "WASM Relaxed SIMD not available"
+#endif
+#include <wasm_simd128.h>
+int main(void) {
+    v128_t a = wasm_f32x4_splat(1.0f);
+    v128_t b = wasm_f32x4_splat(2.0f);
+    v128_t c = wasm_f32x4_relaxed_madd(a, b, a);
+    return wasm_f32x4_extract_lane(c, 0) > 0.0f ? 0 : 1;
+}

package/probes/x86_alder.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Alder Lake (AVX-VNNI 256-bit) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVXVNNI__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int two = 2;
+    __m256i acc = _mm256_setzero_si256();
+    __m256i a = _mm256_set1_epi8((char)two);
+    __m256i b = _mm256_set1_epi8((char)(two + 1));
+    acc = _mm256_dpbusd_avx_epi32(acc, a, b);
+    return _mm256_extract_epi32(acc, 0) == 24 ? 0 : 1;
+}

package/probes/x86_diamond.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Diamond Rapids (AVX10.2) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512FP16__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int one = 1;
+    __m256i acc = _mm256_setzero_si256();
+    __m256i a = _mm256_set1_epi8((char)one);
+    __m256i b = _mm256_set1_epi8((char)one);
+    acc = _mm256_dpbe4ss_epi32(acc, a, b);
+    return _mm256_extract_epi32(acc, 0) != 0 ? 0 : 1;
+}

package/probes/x86_genoa.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Genoa (AVX-512F/BW/DQ/VL + BF16) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512BF16__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile float one = 1.0f;
+    __m512 f = _mm512_set1_ps(one);
+    __m256bh a = _mm512_cvtneps_pbh(f);
+    __m512bh wide = (__m512bh)_mm512_castsi512_ps(_mm512_inserti64x4(_mm512_setzero_si512(), (__m256i)a, 0));
+    __m512 r = _mm512_dpbf16_ps(_mm512_setzero_ps(), wide, wide);
+    return _mm512_reduce_add_ps(r) >= 0.0f ? 0 : 1;
+}

package/probes/x86_graniteamx.c

@@ -0,0 +1,19 @@
+/* NumKong ISA probe: Granite Rapids AMX (AMX-TILE + AMX-FP16) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if defined(__FreeBSD__)
+#error "AMX not supported on FreeBSD"
+#endif
+
+#if !defined(__AMX_FP16__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+#include <amxfp16intrin.h>
+int main(void) {
+    volatile int zero = 0;
+    _tile_release();
+    return zero;
+}

package/probes/x86_haswell.c

@@ -0,0 +1,11 @@
+/* NumKong ISA probe: Haswell (AVX2 + FMA + F16C) */
+#if !defined(__AVX2__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int one = 1;
+    __m256i a = _mm256_set1_epi32(one);
+    __m256i b = _mm256_add_epi32(a, a);
+    return _mm256_extract_epi32(b, 0) == 2 ? 0 : 1;
+}

package/probes/x86_icelake.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Ice Lake (AVX-512F/BW/DQ/VL + VNNI + VBMI + VPOPCNTDQ) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512VNNI__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int one = 1;
+    __m512i acc = _mm512_setzero_si512();
+    __m512i a = _mm512_set1_epi8((char)one);
+    __m512i b = _mm512_set1_epi8((char)one);
+    acc = _mm512_dpbusd_epi32(acc, a, b);
+    return (int)_mm512_reduce_add_epi32(acc) == 64 ? 0 : 1;
+}

package/probes/x86_sapphire.c

@@ -0,0 +1,16 @@
+/* NumKong ISA probe: Sapphire Rapids (AVX-512F/BW/DQ/VL + FP16) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512FP16__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile float one = 1.0f;
+    __m512h a = _mm512_set1_ph((_Float16)one);
+    __m512h b = _mm512_set1_ph((_Float16)(one + one));
+    __m512h c = _mm512_fmadd_ph(a, b, a);
+    return (int)_mm_extract_epi16(_mm256_castsi256_si128(_mm512_castsi512_si256((__m512i)c)), 0) != 0 ? 0 : 1;
+}

package/probes/x86_sapphireamx.c

@@ -0,0 +1,18 @@
+/* NumKong ISA probe: Sapphire Rapids AMX (AMX-TILE + AMX-INT8) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if defined(__FreeBSD__)
+#error "AMX not supported on FreeBSD"
+#endif
+
+#if !defined(__AMX_INT8__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int zero = 0;
+    _tile_release();
+    return zero;
+}

package/probes/x86_sierra.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Sierra Forest (AVXVNNIINT8) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVXVNNIINT8__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int two = 2;
+    __m256i acc = _mm256_setzero_si256();
+    __m256i a = _mm256_set1_epi8((char)two);
+    __m256i b = _mm256_set1_epi8((char)(two + 1));
+    acc = _mm256_dpbssd_epi32(acc, a, b);
+    return _mm256_extract_epi32(acc, 0) == 24 ? 0 : 1;
+}

package/probes/x86_skylake.c

@@ -0,0 +1,15 @@
+/* NumKong ISA probe: Skylake (AVX-512F/BW/DQ/VL) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512F__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int one = 1;
+    __m512i a = _mm512_set1_epi32(one);
+    __m512i b = _mm512_add_epi32(a, a);
+    return (int)_mm512_reduce_add_epi32(b) == 32 ? 0 : 1;
+}

package/probes/x86_turin.c

@@ -0,0 +1,17 @@
+/* NumKong ISA probe: Turin (AVX-512F + VP2INTERSECT) */
+#if defined(__APPLE__)
+#error "AVX-512 not available on macOS"
+#endif
+
+#if !defined(__AVX512VP2INTERSECT__)
+#error "Feature not available"
+#endif
+#include <immintrin.h>
+int main(void) {
+    volatile int val = 42;
+    __m512i a = _mm512_set1_epi32(val);
+    __m512i b = _mm512_set1_epi32(val);
+    __mmask16 k0, k1;
+    _mm512_2intersect_epi32(a, b, &k0, &k1);
+    return k0 != 0 ? 0 : 1;
+}

package/wasm/numkong-emscripten.js

@@ -0,0 +1,2 @@
+async function NumKongModule(moduleArg={}){var moduleRtn;var Module=moduleArg;var ENVIRONMENT_IS_WEB=!!globalThis.window;var ENVIRONMENT_IS_WORKER=!!globalThis.WorkerGlobalScope;var ENVIRONMENT_IS_NODE=globalThis.process?.versions?.node&&globalThis.process?.type!="renderer";if(ENVIRONMENT_IS_NODE){const{createRequire}=await import("node:module");var require=createRequire(import.meta.url)}var arguments_=[];var thisProgram="./this.program";var quit_=(status,toThrow)=>{throw toThrow};var _scriptName=import.meta.url;var scriptDirectory="";function locateFile(path){if(Module["locateFile"]){return Module["locateFile"](path,scriptDirectory)}return scriptDirectory+path}var readAsync,readBinary;if(ENVIRONMENT_IS_NODE){var fs=require("node:fs");if(_scriptName.startsWith("file:")){scriptDirectory=require("node:path").dirname(require("node:url").fileURLToPath(_scriptName))+"/"}readBinary=filename=>{filename=isFileURI(filename)?new URL(filename):filename;var ret=fs.readFileSync(filename);return ret};readAsync=async(filename,binary=true)=>{filename=isFileURI(filename)?new URL(filename):filename;var ret=fs.readFileSync(filename,binary?undefined:"utf8");return ret};if(process.argv.length>1){thisProgram=process.argv[1].replace(/\\/g,"/")}arguments_=process.argv.slice(2);quit_=(status,toThrow)=>{process.exitCode=status;throw toThrow}}else if(ENVIRONMENT_IS_WEB||ENVIRONMENT_IS_WORKER){try{scriptDirectory=new URL(".",_scriptName).href}catch{}{readAsync=async url=>{var response=await fetch(url,{credentials:"same-origin"});if(response.ok){return response.arrayBuffer()}throw new Error(response.status+" : "+response.url)}}}else{}var out=console.log.bind(console);var err=console.error.bind(console);var wasmBinary;var ABORT=false;var isFileURI=filename=>filename.startsWith("file://");class EmscriptenEH{}class EmscriptenSjLj extends EmscriptenEH{}var readyPromiseResolve,readyPromiseReject;var runtimeInitialized=false;function updateMemoryViews(){var b=wasmMemory.buffer;HEAP8=new Int8Array(b);HEAP16=new Int16Array(b);HEAPU8=new Uint8Array(b);HEAPU16=new Uint16Array(b);HEAP32=new Int32Array(b);HEAPU32=new Uint32Array(b);HEAPF32=new Float32Array(b);HEAPF64=new Float64Array(b);HEAP64=new BigInt64Array(b);HEAPU64=new BigUint64Array(b)}function initMemory(){if(Module["wasmMemory"]){wasmMemory=Module["wasmMemory"]}else{var INITIAL_MEMORY=Module["INITIAL_MEMORY"]||16777216;wasmMemory=new WebAssembly.Memory({initial:INITIAL_MEMORY/65536,maximum:32768})}updateMemoryViews()}function preRun(){if(Module["preRun"]){if(typeof Module["preRun"]=="function")Module["preRun"]=[Module["preRun"]];while(Module["preRun"].length){addOnPreRun(Module["preRun"].shift())}}callRuntimeCallbacks(onPreRuns)}function initRuntime(){runtimeInitialized=true;wasmExports["d"]()}function postRun(){if(Module["postRun"]){if(typeof Module["postRun"]=="function")Module["postRun"]=[Module["postRun"]];while(Module["postRun"].length){addOnPostRun(Module["postRun"].shift())}}callRuntimeCallbacks(onPostRuns)}function abort(what){Module["onAbort"]?.(what);what=`Aborted(${what})`;err(what);ABORT=true;what+=". Build with -sASSERTIONS for more info.";var e=new WebAssembly.RuntimeError(what);readyPromiseReject?.(e);throw e}var wasmBinaryFile;function findWasmBinary(){if(Module["locateFile"]){return locateFile("numkong.wasm")}return new URL("numkong.wasm",import.meta.url).href}function getBinarySync(file){if(file==wasmBinaryFile&&wasmBinary){return new Uint8Array(wasmBinary)}if(readBinary){return readBinary(file)}throw"both async and sync fetching of the wasm failed"}async function getWasmBinary(binaryFile){if(!wasmBinary){try{var response=await readAsync(binaryFile);return new Uint8Array(response)}catch{}}return getBinarySync(binaryFile)}async function instantiateArrayBuffer(binaryFile,imports){try{var binary=await getWasmBinary(binaryFile);var instance=await WebAssembly.instantiate(binary,imports);return instance}catch(reason){err(`failed to asynchronously prepare wasm: ${reason}`);abort(reason)}}async function instantiateAsync(binary,binaryFile,imports){if(!binary&&!ENVIRONMENT_IS_NODE){try{var response=fetch(binaryFile,{credentials:"same-origin"});var instantiationResult=await WebAssembly.instantiateStreaming(response,imports);return instantiationResult}catch(reason){err(`wasm streaming compile failed: ${reason}`);err("falling back to ArrayBuffer instantiation")}}return instantiateArrayBuffer(binaryFile,imports)}function getWasmImports(){var imports={a:wasmImports};return imports}async function createWasm(){function receiveInstance(instance,module){wasmExports=instance.exports;assignWasmExports(wasmExports);return wasmExports}function receiveInstantiationResult(result){return receiveInstance(result["instance"])}var info=getWasmImports();if(Module["instantiateWasm"]){return new Promise((resolve,reject)=>{Module["instantiateWasm"](info,(inst,mod)=>{resolve(receiveInstance(inst,mod))})})}wasmBinaryFile??=findWasmBinary();var result=await instantiateAsync(wasmBinary,wasmBinaryFile,info);var exports=receiveInstantiationResult(result);return exports}class ExitStatus{name="ExitStatus";constructor(status){this.message=`Program terminated with exit(${status})`;this.status=status}}var HEAP16;var HEAP32;var HEAP64;var HEAP8;var HEAPF32;var HEAPF64;var HEAPU16;var HEAPU32;var HEAPU64;var HEAPU8;var callRuntimeCallbacks=callbacks=>{while(callbacks.length>0){callbacks.shift()(Module)}};var onPostRuns=[];var addOnPostRun=cb=>onPostRuns.push(cb);var onPreRuns=[];var addOnPreRun=cb=>onPreRuns.push(cb);var noExitRuntime=true;var wasmMemory;var getHeapMax=()=>2147483648;var alignMemory=(size,alignment)=>Math.ceil(size/alignment)*alignment;var growMemory=size=>{var oldHeapSize=wasmMemory.buffer.byteLength;var pages=(size-oldHeapSize+65535)/65536|0;try{wasmMemory.grow(pages);updateMemoryViews();return 1}catch(e){}};var _emscripten_resize_heap=requestedSize=>{var oldSize=HEAPU8.length;requestedSize>>>=0;var maxHeapSize=getHeapMax();if(requestedSize>maxHeapSize){return false}for(var cutDown=1;cutDown<=4;cutDown*=2){var overGrownHeapSize=oldSize*(1+.2/cutDown);overGrownHeapSize=Math.min(overGrownHeapSize,requestedSize+100663296);var newSize=Math.min(maxHeapSize,alignMemory(Math.max(requestedSize,overGrownHeapSize),65536));var replacement=growMemory(newSize);if(replacement){return true}}return false};{initMemory();if(Module["noExitRuntime"])noExitRuntime=Module["noExitRuntime"];if(Module["print"])out=Module["print"];if(Module["printErr"])err=Module["printErr"];if(Module["wasmBinary"])wasmBinary=Module["wasmBinary"];if(Module["arguments"])arguments_=Module["arguments"];if(Module["thisProgram"])thisProgram=Module["thisProgram"];if(Module["preInit"]){if(typeof Module["preInit"]=="function")Module["preInit"]=[Module["preInit"]];while(Module["preInit"].length>0){Module["preInit"].shift()()}}}Module["wasmMemory"]=wasmMemory;function nk_has_relaxed(){var test=new Uint8Array([0,97,115,109,1,0,0,0,1,8,1,96,3,123,123,123,1,123,3,2,1,0,10,9,1,7,0,32,0,32,1,32,2,253,175,1,11]);try{return WebAssembly.validate(test)?1:0}catch(e){return 0}}var _nk_dot_f64,_nk_dot_f32,_nk_dot_bf16,_nk_dot_f16,_nk_dot_i8,_nk_dot_u8,_nk_angular_f64,_nk_angular_f32,_nk_angular_bf16,_nk_angular_f16,_nk_angular_i8,_nk_angular_u8,_nk_euclidean_f64,_nk_euclidean_f32,_nk_euclidean_bf16,_nk_euclidean_f16,_nk_euclidean_i8,_nk_euclidean_u8,_nk_sqeuclidean_f64,_nk_sqeuclidean_f32,_nk_sqeuclidean_bf16,_nk_sqeuclidean_f16,_nk_sqeuclidean_i8,_nk_sqeuclidean_u8,_nk_hamming_u8,_nk_hamming_u1,_nk_jaccard_u16,_nk_jaccard_u1,_nk_kld_f64,_nk_kld_f32,_nk_jsd_f64,_nk_jsd_f32,_nk_dots_packed_size_f64,_nk_dots_packed_size_f32,_nk_dots_packed_size_bf16,_nk_dots_packed_size_f16,_nk_dots_packed_size_i8,_nk_dots_packed_size_u8,_nk_dots_pack_f64,_nk_dots_pack_f32,_nk_dots_pack_bf16,_nk_dots_pack_f16,_nk_dots_pack_i8,_nk_dots_pack_u8,_nk_dots_packed_f64,_nk_dots_packed_f32,_nk_dots_packed_bf16,_nk_dots_packed_f16,_nk_dots_packed_i8,_nk_dots_packed_u8,_nk_dots_symmetric_f64,_nk_dots_symmetric_f32,_nk_dots_symmetric_bf16,_nk_dots_symmetric_f16,_nk_dots_symmetric_i8,_nk_dots_symmetric_u8,_nk_angulars_packed_f64,_nk_angulars_packed_f32,_nk_angulars_packed_bf16,_nk_angulars_packed_f16,_nk_angulars_symmetric_f64,_nk_angulars_symmetric_f32,_nk_angulars_symmetric_bf16,_nk_angulars_symmetric_f16,_nk_euclideans_packed_f64,_nk_euclideans_packed_f32,_nk_euclideans_packed_bf16,_nk_euclideans_packed_f16,_nk_euclideans_symmetric_f64,_nk_euclideans_symmetric_f32,_nk_euclideans_symmetric_bf16,_nk_euclideans_symmetric_f16,_nk_capabilities,_malloc,_free,__indirect_function_table;function assignWasmExports(wasmExports){_nk_dot_f64=Module["_nk_dot_f64"]=wasmExports["e"];_nk_dot_f32=Module["_nk_dot_f32"]=wasmExports["f"];_nk_dot_bf16=Module["_nk_dot_bf16"]=wasmExports["g"];_nk_dot_f16=Module["_nk_dot_f16"]=wasmExports["h"];_nk_dot_i8=Module["_nk_dot_i8"]=wasmExports["i"];_nk_dot_u8=Module["_nk_dot_u8"]=wasmExports["j"];_nk_angular_f64=Module["_nk_angular_f64"]=wasmExports["k"];_nk_angular_f32=Module["_nk_angular_f32"]=wasmExports["l"];_nk_angular_bf16=Module["_nk_angular_bf16"]=wasmExports["m"];_nk_angular_f16=Module["_nk_angular_f16"]=wasmExports["n"];_nk_angular_i8=Module["_nk_angular_i8"]=wasmExports["o"];_nk_angular_u8=Module["_nk_angular_u8"]=wasmExports["p"];_nk_euclidean_f64=Module["_nk_euclidean_f64"]=wasmExports["q"];_nk_euclidean_f32=Module["_nk_euclidean_f32"]=wasmExports["r"];_nk_euclidean_bf16=Module["_nk_euclidean_bf16"]=wasmExports["s"];_nk_euclidean_f16=Module["_nk_euclidean_f16"]=wasmExports["t"];_nk_euclidean_i8=Module["_nk_euclidean_i8"]=wasmExports["u"];_nk_euclidean_u8=Module["_nk_euclidean_u8"]=wasmExports["v"];_nk_sqeuclidean_f64=Module["_nk_sqeuclidean_f64"]=wasmExports["w"];_nk_sqeuclidean_f32=Module["_nk_sqeuclidean_f32"]=wasmExports["x"];_nk_sqeuclidean_bf16=Module["_nk_sqeuclidean_bf16"]=wasmExports["y"];_nk_sqeuclidean_f16=Module["_nk_sqeuclidean_f16"]=wasmExports["z"];_nk_sqeuclidean_i8=Module["_nk_sqeuclidean_i8"]=wasmExports["A"];_nk_sqeuclidean_u8=Module["_nk_sqeuclidean_u8"]=wasmExports["B"];_nk_hamming_u8=Module["_nk_hamming_u8"]=wasmExports["C"];_nk_hamming_u1=Module["_nk_hamming_u1"]=wasmExports["D"];_nk_jaccard_u16=Module["_nk_jaccard_u16"]=wasmExports["E"];_nk_jaccard_u1=Module["_nk_jaccard_u1"]=wasmExports["F"];_nk_kld_f64=Module["_nk_kld_f64"]=wasmExports["G"];_nk_kld_f32=Module["_nk_kld_f32"]=wasmExports["H"];_nk_jsd_f64=Module["_nk_jsd_f64"]=wasmExports["I"];_nk_jsd_f32=Module["_nk_jsd_f32"]=wasmExports["J"];_nk_dots_packed_size_f64=Module["_nk_dots_packed_size_f64"]=wasmExports["K"];_nk_dots_packed_size_f32=Module["_nk_dots_packed_size_f32"]=wasmExports["L"];_nk_dots_packed_size_bf16=Module["_nk_dots_packed_size_bf16"]=wasmExports["M"];_nk_dots_packed_size_f16=Module["_nk_dots_packed_size_f16"]=wasmExports["N"];_nk_dots_packed_size_i8=Module["_nk_dots_packed_size_i8"]=wasmExports["O"];_nk_dots_packed_size_u8=Module["_nk_dots_packed_size_u8"]=wasmExports["P"];_nk_dots_pack_f64=Module["_nk_dots_pack_f64"]=wasmExports["Q"];_nk_dots_pack_f32=Module["_nk_dots_pack_f32"]=wasmExports["R"];_nk_dots_pack_bf16=Module["_nk_dots_pack_bf16"]=wasmExports["S"];_nk_dots_pack_f16=Module["_nk_dots_pack_f16"]=wasmExports["T"];_nk_dots_pack_i8=Module["_nk_dots_pack_i8"]=wasmExports["U"];_nk_dots_pack_u8=Module["_nk_dots_pack_u8"]=wasmExports["V"];_nk_dots_packed_f64=Module["_nk_dots_packed_f64"]=wasmExports["W"];_nk_dots_packed_f32=Module["_nk_dots_packed_f32"]=wasmExports["X"];_nk_dots_packed_bf16=Module["_nk_dots_packed_bf16"]=wasmExports["Y"];_nk_dots_packed_f16=Module["_nk_dots_packed_f16"]=wasmExports["Z"];_nk_dots_packed_i8=Module["_nk_dots_packed_i8"]=wasmExports["_"];_nk_dots_packed_u8=Module["_nk_dots_packed_u8"]=wasmExports["$"];_nk_dots_symmetric_f64=Module["_nk_dots_symmetric_f64"]=wasmExports["aa"];_nk_dots_symmetric_f32=Module["_nk_dots_symmetric_f32"]=wasmExports["ba"];_nk_dots_symmetric_bf16=Module["_nk_dots_symmetric_bf16"]=wasmExports["ca"];_nk_dots_symmetric_f16=Module["_nk_dots_symmetric_f16"]=wasmExports["da"];_nk_dots_symmetric_i8=Module["_nk_dots_symmetric_i8"]=wasmExports["ea"];_nk_dots_symmetric_u8=Module["_nk_dots_symmetric_u8"]=wasmExports["fa"];_nk_angulars_packed_f64=Module["_nk_angulars_packed_f64"]=wasmExports["ga"];_nk_angulars_packed_f32=Module["_nk_angulars_packed_f32"]=wasmExports["ha"];_nk_angulars_packed_bf16=Module["_nk_angulars_packed_bf16"]=wasmExports["ia"];_nk_angulars_packed_f16=Module["_nk_angulars_packed_f16"]=wasmExports["ja"];_nk_angulars_symmetric_f64=Module["_nk_angulars_symmetric_f64"]=wasmExports["ka"];_nk_angulars_symmetric_f32=Module["_nk_angulars_symmetric_f32"]=wasmExports["la"];_nk_angulars_symmetric_bf16=Module["_nk_angulars_symmetric_bf16"]=wasmExports["ma"];_nk_angulars_symmetric_f16=Module["_nk_angulars_symmetric_f16"]=wasmExports["na"];_nk_euclideans_packed_f64=Module["_nk_euclideans_packed_f64"]=wasmExports["oa"];_nk_euclideans_packed_f32=Module["_nk_euclideans_packed_f32"]=wasmExports["pa"];_nk_euclideans_packed_bf16=Module["_nk_euclideans_packed_bf16"]=wasmExports["qa"];_nk_euclideans_packed_f16=Module["_nk_euclideans_packed_f16"]=wasmExports["ra"];_nk_euclideans_symmetric_f64=Module["_nk_euclideans_symmetric_f64"]=wasmExports["sa"];_nk_euclideans_symmetric_f32=Module["_nk_euclideans_symmetric_f32"]=wasmExports["ta"];_nk_euclideans_symmetric_bf16=Module["_nk_euclideans_symmetric_bf16"]=wasmExports["ua"];_nk_euclideans_symmetric_f16=Module["_nk_euclideans_symmetric_f16"]=wasmExports["va"];_nk_capabilities=Module["_nk_capabilities"]=wasmExports["wa"];_malloc=Module["_malloc"]=wasmExports["xa"];_free=Module["_free"]=wasmExports["ya"];__indirect_function_table=wasmExports["__indirect_function_table"]}var wasmImports={b:_emscripten_resize_heap,a:wasmMemory,c:nk_has_relaxed};function run(){preRun();function doRun(){Module["calledRun"]=true;if(ABORT)return;initRuntime();readyPromiseResolve?.(Module);Module["onRuntimeInitialized"]?.();postRun()}if(Module["setStatus"]){Module["setStatus"]("Running...");setTimeout(()=>{setTimeout(()=>Module["setStatus"](""),1);doRun()},1)}else{doRun()}}var wasmExports;wasmExports=await (createWasm());run();if(runtimeInitialized){moduleRtn=Module}else{moduleRtn=new Promise((resolve,reject)=>{readyPromiseResolve=resolve;readyPromiseReject=reject})}
+;return moduleRtn}export default NumKongModule;

package/wasm/numkong.d.ts

@@ -0,0 +1,14 @@
+/**
+ * @brief Self-contained browser ESM entry point for NumKong WASM.
+ * @file javascript/numkong-browser.ts
+ *
+ * Auto-initializes the Emscripten module on import via top-level await.
+ * The Emscripten glue (`numkong-emscripten.js`) and binary (`numkong.wasm`)
+ * must be co-located with this file (same directory or CDN prefix).
+ *
+ * Usage:
+ *   import { dot, euclidean } from './numkong.js';
+ *   console.log(dot(new Float32Array([1,2,3]), new Float32Array([4,5,6])));
+ */
+export { TensorBase, VectorBase, VectorView, Vector, MatrixBase, Matrix, PackedMatrix, DType, TypedArray, KernelFamily, dtypeToString, outputDtype, Float16Array, BFloat16Array, E4M3Array, E5M2Array, BinaryArray, isFloat16Array, isBFloat16Array, isE4M3Array, isE5M2Array, isBinaryArray, } from './types.js';
+export { dot, inner, euclidean, sqeuclidean, angular, hamming, jaccard, kullbackleibler, jensenshannon, getCapabilities, hasCapability, dotsPack, dotsPackedSize, dotsPacked, angularsPacked, euclideansPacked, dotsSymmetric, angularsSymmetric, euclideansSymmetric, } from './numkong-wasm.js';