diff --git a/.github/workflows/stockfish.yml b/.github/workflows/stockfish.yml
index 0606d685c..2608f0bf6 100644
--- a/.github/workflows/stockfish.yml
+++ b/.github/workflows/stockfish.yml
@@ -150,6 +150,10 @@ jobs:
     name: Compiler check
     if: ${{ always() }}
     uses: ./.github/workflows/avx2_compilers.yml
+  WASMCompilation:
+    name: WASM compilation
+    if: ${{ always() }}
+    uses: ./.github/workflows/wasm_compilation.yml
   # Release Jobs
   ARMCompilation:
     name: Android builds
diff --git a/.github/workflows/wasm_compilation.yml b/.github/workflows/wasm_compilation.yml
new file mode 100644
index 000000000..12f23c683
--- /dev/null
+++ b/.github/workflows/wasm_compilation.yml
@@ -0,0 +1,51 @@
+name: WASM Compilation
+
+on:
+  workflow_call:
+
+jobs:
+  wasm:
+    name: wasm (${{ matrix.arch }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [wasm32, wasm32-relaxed-simd]
+    defaults:
+      run:
+        working-directory: src
+        shell: bash
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: Install emsdk
+        working-directory: .
+        run: |
+          git clone --depth 1 https://github.com/emscripten-core/emsdk.git
+          cd emsdk
+          ./emsdk install latest
+          ./emsdk activate latest
+
+      - name: Extract the bench number from the commit history
+        run: |
+          for hash in $(git rev-list -100 HEAD); do
+            benchref=$(git show -s $hash | tac | grep -m 1 -o -x '[[:space:]]*\b[Bb]ench[ :]\+[1-9][0-9]\{5,7\}\b[[:space:]]*' | sed 's/[^0-9]//g') && break || true
+          done
+          [[ -n "$benchref" ]] && echo "benchref=$benchref" >> $GITHUB_ENV && echo "From commit: $hash" && echo "Reference bench: $benchref" || echo "No bench found"
+
+      - name: Download the used network from the fishtest framework
+        run: make net
+
+      - name: Build ${{ matrix.arch }} build
+        run: |
+          source $GITHUB_WORKSPACE/emsdk/emsdk_env.sh
+          em++ --version
+          net=$(grep -oE 'nn-[0-9a-f]+\.nnue' evaluate.h | head -1)
+          make clean
+          COMPCXX=em++ make -j build ARCH=${{ matrix.arch }} EXTRALDFLAGS="--preload-file $net"
+
+      - name: Verify bench under node
+        run: EXE=./stockfish.js RUN_PREFIX=node ../tests/signature.sh $benchref
diff --git a/src/Makefile b/src/Makefile
index ab91d1d0f..cdf0ee1a8 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -55,6 +55,8 @@ endif
 ### Executable name
 ifeq ($(target_windows),yes)
 	EXE = stockfish.exe
+else ifneq (,$(findstring wasm,$(ARCH)))
+	EXE = stockfish.js
 else
 	EXE = stockfish
 endif
@@ -122,7 +124,9 @@ VPATH = syzygy:nnue:nnue/features
 # neon = yes/no       --- -DUSE_NEON         --- Use ARM SIMD architecture
 # dotprod = yes/no    --- -DUSE_NEON_DOTPROD --- Use ARM advanced SIMD Int8 dot product instructions
 # lsx = yes/no        --- -mlsx              --- Use Loongson SIMD eXtension
-# lasx = yes/no       --- -mlasx             --- use Loongson Advanced SIMD eXtension
+# lasx = yes/no       --- -mlasx             --- Use Loongson Advanced SIMD eXtension
+# relaxedsimd = y/n   --- -mrelaxed-simd     --- Use WebAssembly relaxed SIMD extension
+# syzygy = yes/no     --- -DNO_TABLEBASES    --- Support Syzygy tablebase probing
 #
 # Note that Makefile is space sensitive, so when adding new architectures
 # or modifying existing flags, you have to make sure there are no extra spaces
@@ -149,7 +153,7 @@ ifeq ($(ARCH), $(filter $(ARCH), \
                  x86-64-bmi2 x86-64-avx2 x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
                  x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-64-altivec ppc-64-vsx ppc-32 e2k \
                  armv7 armv7-neon armv8 armv8-dotprod arm64-universal apple-silicon general-64 general-32 riscv64 \
-                 loongarch64 loongarch64-lsx loongarch64-lasx))
+                 loongarch64 loongarch64-lsx loongarch64-lasx wasm32 wasm32-relaxed-simd))
    SUPPORTED_ARCH=true
 else
    SUPPORTED_ARCH=false
@@ -186,6 +190,8 @@ dotprod = no
 arm_version = 0
 lsx = no
 lasx = no
+relaxedsimd = no
+syzygy = yes
 STRIP = strip
 
 ifneq ($(shell which clang-format-20 2> /dev/null),)
@@ -389,6 +395,27 @@ ifeq ($(ARCH),apple-silicon)
 	arm_version = 8
 endif
 
+# WASM is largely compiled using emcc's x86 compatibility layer, with occasional
+# WASM-specific intrinsics
+ifeq ($(ARCH),wasm32)
+	arch = wasm32
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	syzygy = no
+endif
+
+ifeq ($(ARCH),wasm32-relaxed-simd)
+	arch = wasm32
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	relaxedsimd = yes
+	syzygy = no
+endif
+
 ifeq ($(ARCH),ppc-32)
 	arch = ppc
 	bits = 32
@@ -480,7 +507,7 @@ ifeq ($(COMP),gcc)
 		endif
 	else ifeq ($(arch),loongarch64)
 		CXXFLAGS += -latomic
-	else
+	else ifneq ($(arch),wasm32)
 		CXXFLAGS += -m$(bits)
 		LDFLAGS += -m$(bits)
 	endif
@@ -490,8 +517,10 @@ ifeq ($(COMP),gcc)
 	endif
 
 	ifneq ($(KERNEL),Darwin)
+	ifneq ($(arch),wasm32)
 	   LDFLAGS += -Wl,--no-as-needed
 	endif
+	endif
 endif
 
 ifeq ($(target_windows),yes)
@@ -654,6 +683,7 @@ ifneq ($(comp),mingw)
 		# Haiku has pthreads in its libroot, so only link it in on other platforms
 		ifneq ($(KERNEL),Haiku)
 			ifneq ($(COMP),ndk)
+			ifneq ($(arch),wasm32)
 				LDFLAGS += -lpthread
 
 				add_lrt = yes
@@ -669,6 +699,7 @@ ifneq ($(comp),mingw)
 					LDFLAGS += -lrt
 				endif
 			endif
+			endif
 		endif
 	endif
 endif
@@ -846,6 +877,14 @@ ifeq ($(lsx),yes)
 	endif
 endif
 
+ifeq ($(arch),wasm32)
+	CXXFLAGS += -pthread -msimd128 -DUSE_POPCNT -DNNUE_EMBEDDING_OFF
+	LDFLAGS += -pthread -sINITIAL_MEMORY=64MB -sALLOW_MEMORY_GROWTH -sSTACK_SIZE=3MB
+	ifeq ($(relaxedsimd),yes)
+		CXXFLAGS += -mrelaxed-simd
+	endif
+endif
+
 ### 3.7 pext
 ifeq ($(pext),yes)
 	CXXFLAGS += -DUSE_PEXT
@@ -854,6 +893,11 @@ ifeq ($(pext),yes)
 	endif
 endif
 
+### Syzygy tablebases
+ifeq ($(syzygy),no)
+	CXXFLAGS += -DNO_TABLEBASES
+endif
+
 ### 3.8.1 Try to include git info for versioning and avoid recompiles if nothing changes
 BUILD_SHA_FILE  := .build_sha.txt
 BUILD_DATE_FILE := .build_date.txt
@@ -885,7 +929,8 @@ endif
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
 ifeq ($(debug),no)
-	ifneq ($(KERNEL),Darwin)
+	ifeq ($(arch),wasm32)
+	else ifneq ($(KERNEL),Darwin)
 		LLD_BIN := $(shell command -v ld.lld 2>/dev/null)
 		ifeq ($(LLD_BIN),)
 			LLD_BIN := $(shell command -v lld 2>/dev/null)
@@ -1207,6 +1252,7 @@ config-sanity: net
 	echo "arm_version: '$(arm_version)'" && \
 	echo "lsx: '$(lsx)'" && \
 	echo "lasx: '$(lasx)'" && \
+	echo "syzygy: '$(syzygy)'" && \
 	echo "target_windows: '$(target_windows)'" && \
 	echo "" && \
 	echo "Flags:" && \
@@ -1222,7 +1268,8 @@ config-sanity: net
 	(test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
 	 test "$(arch)" = "ppc64" || test "$(arch)" = "ppc" || test "$(arch)" = "e2k" || \
 	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64" || \
-	 test "$(arch)" = "riscv64" || test "$(arch)" = "loongarch64") && \
+	 test "$(arch)" = "riscv64" || test "$(arch)" = "loongarch64" || \
+	 test "$(arch)" = "wasm32") && \
 	(test "$(bits)" = "32" || test "$(bits)" = "64") && \
 	(test "$(prefetch)" = "yes" || test "$(prefetch)" = "no") && \
 	(test "$(popcnt)" = "yes" || test "$(popcnt)" = "no") && \
@@ -1241,6 +1288,7 @@ config-sanity: net
 	(test "$(neon)" = "yes" || test "$(neon)" = "no") && \
 	(test "$(lsx)" = "yes" || test "$(lsx)" = "no") && \
 	(test "$(lasx)" = "yes" || test "$(lasx)" = "no") && \
+	(test "$(syzygy)" = "yes" || test "$(syzygy)" = "no") && \
 	(test "$(comp)" = "gcc" || test "$(comp)" = "icx" || test "$(comp)" = "mingw" || \
 	 test "$(comp)" = "clang" || test "$(comp)" = "armv7a-linux-androideabi16-clang" || \
 	 test "$(comp)" = "aarch64-linux-android21-clang")
diff --git a/src/history.h b/src/history.h
index 7ba43234d..4e0dbeb03 100644
--- a/src/history.h
+++ b/src/history.h
@@ -52,11 +52,17 @@ static_assert((CORRHIST_BASE_SIZE & (CORRHIST_BASE_SIZE - 1)) == 0,
 // the entry. The first template parameter T is the base type of the array,
 // and the second template parameter D limits the range of updates in [-D, D]
 // when we update values with the << operator
-template<typename T, int D, bool Atomic = false>
+template<typename T, int D, bool Shared = false>
 struct StatsEntry {
     static_assert(std::is_arithmetic_v<T>, "Not an arithmetic type");
 
    private:
+    static constexpr bool Atomic =
+#ifdef __wasm__
+      false;  // don't use atomics on WASM as they are always seq_cst
+#else
+      Shared;
+#endif
     std::conditional_t<Atomic, std::atomic<T>, T> entry;
 
    public:
diff --git a/src/misc.h b/src/misc.h
index cadff28cc..219e75ab8 100644
--- a/src/misc.h
+++ b/src/misc.h
@@ -384,12 +384,12 @@ class PRNG {
     }
 };
 
-inline u64 mul_hi64(u64 a, u64 b) {
-#if defined(__GNUC__) && defined(IS_64BIT)
+inline usize mul_hi64(u64 a, usize b) {
+#if defined(__GNUC__) && defined(IS_64BIT) && !defined(__wasm__)
     return (u128(a) * u128(b)) >> 64;
 #else
     u64 aL = u32(a), aH = a >> 32;
-    u64 bL = u32(b), bH = b >> 32;
+    u64 bL = u32(b), bH = u64(b) >> 32;
     u64 c1 = (aL * bL) >> 32;
     u64 c2 = aH * bL + c1;
     u64 c3 = aL * bH + u32(c2);
diff --git a/src/nnue/nnue_accumulator.cpp b/src/nnue/nnue_accumulator.cpp
index 35040b2b8..46ebe5db4 100644
--- a/src/nnue/nnue_accumulator.cpp
+++ b/src/nnue/nnue_accumulator.cpp
@@ -561,6 +561,19 @@ Bitboard get_changed_pieces(const std::array<Piece, SQUARE_NB>& oldPieces,
     uint8x8_t sameBB  = vshrn_n_u16(vreinterpretq_u16_u8(merged), 4);
 
     return ~vget_lane_u64(vreinterpret_u64_u8(sameBB), 0);
+#elif defined(USE_SSE2)
+    Bitboard sameBB = 0;
+
+    for (int i = 0; i < 64; i += 16)
+    {
+        const __m128i old_v = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&oldPieces[i]));
+        const __m128i new_v = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&newPieces[i]));
+        const __m128i same  = _mm_cmpeq_epi8(old_v, new_v);
+
+        sameBB |= static_cast<Bitboard>(_mm_movemask_epi8(same)) << i;
+    }
+
+    return ~sameBB;
 #else
     Bitboard changed = 0;
 
diff --git a/src/nnue/nnue_feature_transformer.h b/src/nnue/nnue_feature_transformer.h
index c4c92f059..12f620648 100644
--- a/src/nnue/nnue_feature_transformer.h
+++ b/src/nnue/nnue_feature_transformer.h
@@ -233,11 +233,9 @@ class FeatureTransformer {
             static_assert((HalfDimensions / 2) % OutputChunkSize == 0);
             constexpr IndexType NumOutputChunks = HalfDimensions / 2 / OutputChunkSize;
 
-    #if !defined(USE_NEON)
-            const vec_t   Zero  = vec_zero();
-            const vec_t   FtMax = vec_set_16(FtMaxVal);
-            constexpr int shift = 7;
-    #endif
+            [[maybe_unused]] const vec_t   Zero  = vec_zero();
+            [[maybe_unused]] const vec_t   FtMax = vec_set_16(FtMaxVal);
+            [[maybe_unused]] constexpr int shift = 7;
 
             const vec_t* in0 = reinterpret_cast<const vec_t*>(&(accumulation[perspectives[p]][0]));
             const vec_t* in1 =
@@ -323,6 +321,19 @@ class FeatureTransformer {
 
                     vec_t hi     = vec_mulhi_8(pa, pb);
                     vec_t result = vec_srli_8(hi, 1);
+    #elif defined(__wasm__)
+                    // _mm_mulhi_epi16 is lowered to 32-bit multiplies, so we take
+                    // a similar approach as the NEON path.
+                    vec_t mul0 = vec_packus_16(acc0a, acc0b);
+                    vec_t mul1 = vec_packus_16(acc1a, acc1b);
+
+                    vec_t low = wasm_u16x8_extmul_low_u8x16(mul0, mul1);
+                    vec_t hi  = wasm_u16x8_extmul_high_u8x16(mul0, mul1);
+
+                    // equivalent to vuzp2_u8
+                    vec_t merged = wasm_i8x16_shuffle(low, hi, 1, 3, 5, 7, 9, 11, 13, 15, 17, 19,
+                                                      21, 23, 25, 27, 29, 31);
+                    vec_t result = wasm_u8x16_shr(merged, 1);
     #else
                     vec_t sum0a = vec_slli_16(vec_max_16(vec_min_16(acc0a, FtMax), Zero), shift);
                     vec_t sum0b = vec_slli_16(vec_max_16(vec_min_16(acc0b, FtMax), Zero), shift);
diff --git a/src/nnue/nnz_helper.h b/src/nnue/nnz_helper.h
index 7bbe98437..b7542c7bc 100644
--- a/src/nnue/nnz_helper.h
+++ b/src/nnue/nnz_helper.h
@@ -132,6 +132,10 @@ struct NNZInfo {
             const uint16x8_t bits = vandq_u16(packed, vld1q_u16(Mask8));
 
             *out++ = vaddvq_u16(bits);
+        #elif defined(__wasm__)
+            __m128i packed = _mm_packus_epi32(neurons1, neurons2);
+            packed         = _mm_packs_epi16(packed, packed);
+            *out++         = ~_mm_movemask_epi8(_mm_cmpeq_epi8(packed, _mm_setzero_si128()));
         #else
             auto m1 = vec_nnz(neurons1);
             auto m2 = vec_nnz(neurons2);
diff --git a/src/nnue/simd.h b/src/nnue/simd.h
index fca5e260a..bf6c78732 100644
--- a/src/nnue/simd.h
+++ b/src/nnue/simd.h
@@ -168,7 +168,12 @@ inline __m128i _mm_cvtsi64_si128(i64 val) {
     #endif
 
     #ifdef USE_SSE41
-        #define vec_convert_8_16(a) _mm_cvtepi8_epi16(_mm_cvtsi64_si128(static_cast<i64>(a)))
+        #ifdef __wasm__
+            #define vec_convert_8_16(a) wasm_i16x8_load8x8(reinterpret_cast<const void*>(&a))
+        #else
+            #define vec_convert_8_16(a) \
+                _mm_cvtepi8_epi16(_mm_cvtsi64_si128(static_cast<int64_t>(a)))
+        #endif
     #else
 // Credit: Yoshie2000
 inline __m128i vec_convert_8_16(u64 x) {
@@ -495,10 +500,13 @@ fused(const typename VecWrapper::type& in, const T& operand, const Ts&... operan
 }
 
 [[maybe_unused]] static void m128_add_dpbusd_epi32(__m128i& acc, __m128i a, __m128i b) {
-
+    #if defined(__wasm_relaxed_simd__)
+    acc = wasm_i32x4_relaxed_dot_i8x16_i7x16_add(b, a, acc);
+    #else
     __m128i product0 = _mm_maddubs_epi16(a, b);
     product0         = _mm_madd_epi16(product0, _mm_set1_epi16(1));
     acc              = _mm_add_epi32(acc, product0);
+    #endif
 }
 
 #endif
diff --git a/src/syzygy/tbprobe.cpp b/src/syzygy/tbprobe.cpp
index b230ac59b..21d619572 100644
--- a/src/syzygy/tbprobe.cpp
+++ b/src/syzygy/tbprobe.cpp
@@ -59,6 +59,40 @@
     #include <windows.h>
 #endif
 
+#ifdef NO_TABLEBASES
+
+// Stubbed out impls
+namespace Stockfish::Tablebases {
+
+int MaxCardinality;
+
+void init(const std::string&) {}
+
+WDLScore probe_wdl(Position&, ProbeState* result) {
+    *result = FAIL;
+    return WDLDraw;
+}
+
+int probe_dtz(Position&, ProbeState* result) {
+    *result = FAIL;
+    return 0;
+}
+
+bool root_probe(Position&, Search::RootMoves&, bool, bool, const std::function<bool()>&) {
+    return false;
+}
+
+bool root_probe_wdl(Position&, Search::RootMoves&, bool) { return false; }
+
+Config rank_root_moves(
+  const OptionsMap&, Position&, Search::RootMoves&, bool, const std::function<bool()>&) {
+    return Config{};
+}
+
+}  // namespace Stockfish::Tablebases
+
+#else
+
 using namespace Stockfish::Tablebases;
 
 int Stockfish::Tablebases::MaxCardinality;
@@ -206,11 +240,11 @@ class TBFile: public std::ifstream {
 
     TBFile(const std::string& f) {
 
-#ifndef _WIN32
+    #ifndef _WIN32
         constexpr char SepChar = ':';
-#else
+    #else
         constexpr char SepChar = ';';
-#endif
+    #endif
         std::stringstream ss(Paths);
         std::string       path;
 
@@ -228,7 +262,7 @@ class TBFile: public std::ifstream {
         if (is_open())
             close();  // Need to re-open to get native file descriptor
 
-#ifndef _WIN32
+    #ifndef _WIN32
         struct stat statbuf;
         int         fd = ::open(fname.c_str(), O_RDONLY);
 
@@ -245,9 +279,9 @@ class TBFile: public std::ifstream {
 
         *mapping     = statbuf.st_size;
         *baseAddress = mmap(nullptr, statbuf.st_size, PROT_READ, MAP_SHARED, fd, 0);
-    #if defined(MADV_RANDOM)
+        #if defined(MADV_RANDOM)
         madvise(*baseAddress, statbuf.st_size, MADV_RANDOM);
-    #endif
+        #endif
         ::close(fd);
 
         if (*baseAddress == MAP_FAILED)
@@ -255,7 +289,7 @@ class TBFile: public std::ifstream {
             std::cerr << "Could not mmap() " << fname << std::endl;
             exit(EXIT_FAILURE);
         }
-#else
+    #else
         // Note FILE_FLAG_RANDOM_ACCESS is only a hint to Windows and as such may get ignored.
         HANDLE fd = CreateFileA(fname.c_str(), GENERIC_READ, FILE_SHARE_READ, nullptr,
                                 OPEN_EXISTING, FILE_FLAG_RANDOM_ACCESS, nullptr);
@@ -290,7 +324,7 @@ class TBFile: public std::ifstream {
                       << ", error = " << GetLastError() << std::endl;
             exit(EXIT_FAILURE);
         }
-#endif
+    #endif
         u8* data = (u8*) *baseAddress;
 
         constexpr u8 Magics[][4] = {{0xD7, 0x66, 0x0C, 0xA5}, {0x71, 0xE8, 0x23, 0x5D}};
@@ -307,12 +341,12 @@ class TBFile: public std::ifstream {
 
     static void unmap(void* baseAddress, u64 mapping) {
 
-#ifndef _WIN32
+    #ifndef _WIN32
         munmap(baseAddress, mapping);
-#else
+    #else
         UnmapViewOfFile(baseAddress);
         CloseHandle((HANDLE) mapping);
-#endif
+    #endif
     }
 };
 
@@ -728,12 +762,12 @@ int map_score(TBTable<DTZ>* entry, File f, int value, WDLScore wdl) {
     return value + 1;
 }
 
-// A temporary fix for the compiler bug with vectorization. (#4450)
-#if defined(__clang__) && defined(__clang_major__) && __clang_major__ >= 15
-    #define DISABLE_CLANG_LOOP_VEC _Pragma("clang loop vectorize(disable)")
-#else
-    #define DISABLE_CLANG_LOOP_VEC
-#endif
+    // A temporary fix for the compiler bug with vectorization. (#4450)
+    #if defined(__clang__) && defined(__clang_major__) && __clang_major__ >= 15
+        #define DISABLE_CLANG_LOOP_VEC _Pragma("clang loop vectorize(disable)")
+    #else
+        #define DISABLE_CLANG_LOOP_VEC
+    #endif
 
 // Compute a unique index out of a position and use it to probe the TB file. To
 // encode k pieces of the same type and color, first sort the pieces by square in
@@ -1802,3 +1836,5 @@ Config Tablebases::rank_root_moves(const OptionsMap&            options,
     return config;
 }
 }  // namespace Stockfish
+
+#endif  // NO_TABLEBASES
diff --git a/tests/signature.sh b/tests/signature.sh
index ef781a0f4..ad96d2ee3 100755
--- a/tests/signature.sh
+++ b/tests/signature.sh
@@ -18,7 +18,8 @@ error()
 trap 'error ${LINENO}' ERR
 
 # obtain
-eval "$RUN_PREFIX ./stockfish bench" > "$STDOUT_FILE" 2> "$STDERR_FILE" || error ${LINENO}
+EXE=${EXE:-./stockfish}
+eval "$RUN_PREFIX $EXE bench" > "$STDOUT_FILE" 2> "$STDERR_FILE" || error ${LINENO}
 signature=$(grep "Nodes searched  : " "$STDERR_FILE" | awk '{print $4}')
 
 rm -f "$STDOUT_FILE" "$STDERR_FILE"