jz 0.0.0 → 0.1.1

package/module/typedarray.js

@@ -0,0 +1,713 @@
+/**
+ * TypedArray module — Float64Array, Float32Array, Int32Array, etc.
+ * SIMD auto-vectorization for .map() on recognized patterns.
+ *
+ * Type=3 (TYPED): aux=elemType (3 bits), length in memory header [-8:len][-4:cap].
+ *
+ * @module typed
+ */
+
+import { typed, asF64, asI32, UNDEF_NAN, allocPtr, mkPtrIR, ptrOffsetIR, temp, tempI32 } from '../src/ir.js'
+import { emit } from '../src/emit.js'
+import { valTypeOf, lookupValType, VAL } from '../src/analyze.js'
+import { inc, PTR } from '../src/ctx.js'
+
+
+// Element types and their byte sizes
+const ELEM = {
+  Int8Array: 0, Uint8Array: 1,
+  Int16Array: 2, Uint16Array: 3,
+  Int32Array: 4, Uint32Array: 5,
+  Float32Array: 6, Float64Array: 7,
+  BigInt64Array: 7, BigUint64Array: 7,
+}
+const STRIDE = [1, 1, 2, 2, 4, 4, 4, 8]
+const SHIFT = [0, 0, 1, 1, 2, 2, 2, 3]
+const LOAD = [
+  'i32.load8_s', 'i32.load8_u', 'i32.load16_s', 'i32.load16_u',
+  'i32.load', 'i32.load', 'f32.load', 'f64.load',
+]
+const STORE = [
+  'i32.store8', 'i32.store8', 'i32.store16', 'i32.store16',
+  'i32.store', 'i32.store', 'f32.store', 'f64.store',
+]
+
+// SIMD: vector width per element type (elements per v128)
+const VEC_WIDTH = [16, 16, 8, 8, 4, 4, 4, 2] // 128 bits / element bits
+
+
+// === SIMD pattern detection ===
+
+/** Check if AST node is a constant number */
+const isConst = node => {
+  if (typeof node === 'number') return node
+  if (Array.isArray(node) && node[0] == null && typeof node[1] === 'number') return node[1]
+  return false
+}
+
+/**
+ * Analyze callback body for SIMD-vectorizable patterns.
+ * Returns { op, val } or null.
+ */
+function analyzeSimd(body, param) {
+  if (!Array.isArray(body)) return null
+  const [op, ...args] = body
+
+  // Binary: x*c, x+c, x-c, x/c (and commutative)
+  if (['+', '-', '*', '/'].includes(op) && args.length === 2) {
+    const [a, b] = args
+    const isA = a === param, isB = b === param
+    const cA = !isA && isConst(a), cB = !isB && isConst(b)
+    if (op === '*' && ((isA && cB !== false) || (isB && cA !== false)))
+      return { op: 'mul', val: isA ? cB : cA }
+    if (op === '+' && ((isA && cB !== false) || (isB && cA !== false)))
+      return { op: 'add', val: isA ? cB : cA }
+    if (op === '-' && isA && cB !== false) return { op: 'sub', val: cB }
+    if (op === '/' && isA && cB !== false) return { op: 'div', val: cB }
+  }
+
+  // Bitwise: x&c, x|c, x^c, x<<c, x>>c, x>>>c
+  if (['&', '|', '^', '<<', '>>', '>>>'].includes(op) && args.length === 2) {
+    const [a, b] = args
+    if (a === param && isConst(b) !== false) {
+      const ops = { '&': 'and', '|': 'or', '^': 'xor', '<<': 'shl', '>>': 'shr', '>>>': 'shru' }
+      return { op: ops[op], val: isConst(b) }
+    }
+  }
+
+  // Unary minus: ['u-', param]
+  if (op === 'u-' && args[0] === param) return { op: 'neg' }
+
+  // Math.abs/sqrt/ceil/floor
+  if (op === '()' && typeof args[0] === 'string' && args[0].startsWith('math.')) {
+    const method = args[0].slice(5)
+    const fnArg = args[1]
+    if (fnArg === param && ['abs', 'sqrt', 'ceil', 'floor'].includes(method))
+      return { op: method }
+  }
+
+  return null
+}
+
+
+// === SIMD + scalar WAT codegen (parameterized by type prefix) ===
+
+/** Generate SIMD v128 op. p=prefix (f64x2/f32x4/i32x4), t=const type (f64/f32/i32). */
+const simdOp = (p, t) => (op, c) => {
+  const s = `(${p}.splat (${t}.const ${c}))`
+  const ops = {
+    mul: `${p}.mul (local.get $v) ${s}`, add: `${p}.add (local.get $v) ${s}`,
+    sub: `${p}.sub (local.get $v) ${s}`, div: `${p}.div (local.get $v) ${s}`,
+    neg: `${p}.neg (local.get $v)`, abs: `${p}.abs (local.get $v)`,
+    sqrt: `${p}.sqrt (local.get $v)`, ceil: `${p}.ceil (local.get $v)`, floor: `${p}.floor (local.get $v)`,
+    // i32-only bitwise (no-op for float prefixes since analyzeSimd won't produce these for float)
+    and: `v128.and (local.get $v) (i32x4.splat (i32.const ${c}))`,
+    or: `v128.or (local.get $v) (i32x4.splat (i32.const ${c}))`,
+    xor: `v128.xor (local.get $v) (i32x4.splat (i32.const ${c}))`,
+    shl: `i32x4.shl (local.get $v) (i32.const ${c})`, shr: `i32x4.shr_s (local.get $v) (i32.const ${c})`,
+    shru: `i32x4.shr_u (local.get $v) (i32.const ${c})`,
+  }
+  return ops[op] ? `(local.set $v (${ops[op]}))` : null
+}
+
+/** Generate scalar remainder op. t=type prefix (f64/f32/i32), v=local name. */
+const scalarOp = (t, v) => (op, c) => {
+  const g = `(local.get $${v})`
+  const ops = {
+    mul: `(${t}.mul ${g} (${t}.const ${c}))`, add: `(${t}.add ${g} (${t}.const ${c}))`,
+    sub: `(${t}.sub ${g} (${t}.const ${c}))`, div: `(${t}.div ${g} (${t}.const ${c}))`,
+    neg: t === 'i32' ? `(i32.sub (i32.const 0) ${g})` : `(${t}.neg ${g})`,
+    abs: t === 'i32' ? `(select (i32.sub (i32.const 0) ${g}) ${g} (i32.lt_s ${g} (i32.const 0)))` : `(${t}.abs ${g})`,
+    sqrt: `(${t}.sqrt ${g})`, ceil: `(${t}.ceil ${g})`, floor: `(${t}.floor ${g})`,
+    and: `(i32.and ${g} (i32.const ${c}))`, or: `(i32.or ${g} (i32.const ${c}))`,
+    xor: `(i32.xor ${g} (i32.const ${c}))`, shl: `(i32.shl ${g} (i32.const ${c}))`,
+    shr: `(i32.shr_s ${g} (i32.const ${c}))`, shru: `(i32.shr_u ${g} (i32.const ${c}))`,
+  }
+  return ops[op]
+}
+
+const simdF64 = simdOp('f64x2', 'f64'), simdF32 = simdOp('f32x4', 'f32'), simdI32 = simdOp('i32x4', 'i32')
+const scalarF64 = scalarOp('f64', 'e'), scalarF32 = scalarOp('f32', 'ef'), scalarI32 = scalarOp('i32', 'ei')
+
+
+/**
+ * Generate a SIMD map function as WAT string.
+ * Takes (src: f64) → f64, returns new typed array with transform applied.
+ */
+function genSimdMap(name, elemType, pattern) {
+  const { op, val: c } = pattern
+  const stride = STRIDE[elemType]
+  const shift = SHIFT[elemType]
+  const load = LOAD[elemType], store = STORE[elemType]
+  const vw = VEC_WIDTH[elemType]
+  const vBytes = vw * stride // always 16 (128 bits)
+
+  // Choose SIMD + scalar codegen by element family
+  let simdOp, scalarOp, scalarLocal, scalarLoad, scalarStore
+  if (elemType === 7) { // Float64Array
+    simdOp = simdF64(op, c); scalarOp = scalarF64(op, c)
+    scalarLocal = '(local $e f64)'; scalarLoad = 'f64.load'; scalarStore = 'f64.store'
+  } else if (elemType === 6) { // Float32Array
+    simdOp = simdF32(op, c); scalarOp = scalarF32(op, c)
+    scalarLocal = '(local $ef f32)'; scalarLoad = 'f32.load'; scalarStore = 'f32.store'
+  } else if (elemType >= 4) { // Int32Array/Uint32Array
+    simdOp = simdI32(op, c); scalarOp = scalarI32(op, c)
+    scalarLocal = '(local $ei i32)'; scalarLoad = 'i32.load'; scalarStore = 'i32.store'
+  } else return null // i8/i16/u8/u16 — no SIMD path (would need i8x16/i16x8)
+
+  if (!simdOp || !scalarOp) return null
+
+  // Scalar remainder: load element into local, then store transform result
+  const byteOff = `(i32.add (local.get $srcOff) (i32.shl (local.get $i) (i32.const ${shift})))`
+  const dstByteOff = `(i32.add (local.get $dstOff) (i32.shl (local.get $i) (i32.const ${shift})))`
+  const scalarLoadSet = elemType === 7 ? `(local.set $e (${scalarLoad} ${byteOff}))`
+    : elemType === 6 ? `(local.set $ef (${scalarLoad} ${byteOff}))`
+    : `(local.set $ei (${scalarLoad} ${byteOff}))`
+  const scalarStoreExpr = `${scalarLoadSet}\n      (${store} ${dstByteOff} ${scalarOp})`
+
+  return `(func $${name} (param $src f64) (result f64)
+    (local $len i32) (local $srcOff i32) (local $dstOff i32) (local $dst i32)
+    (local $i i32) (local $simdLen i32) (local $byteOff i32)
+    (local $v v128)
+    ${scalarLocal}
+    (local.set $len (call $__len (local.get $src)))
+    (local.set $srcOff (call $__typed_data (local.get $src)))
+    ;; Alloc result typed array: header(8) + data. Header stores byteLen = len << ${shift}.
+    (local.set $dst (call $__alloc (i32.add (i32.const 8) (i32.shl (local.get $len) (i32.const ${shift})))))
+    (i32.store (local.get $dst) (i32.shl (local.get $len) (i32.const ${shift})))
+    (i32.store (i32.add (local.get $dst) (i32.const 4)) (i32.shl (local.get $len) (i32.const ${shift})))
+    (local.set $dstOff (i32.add (local.get $dst) (i32.const 8)))
+    ;; SIMD loop: process ${vw} elements at a time
+    (local.set $simdLen (i32.and (local.get $len) (i32.const ${~(vw - 1)})))
+    (local.set $i (i32.const 0))
+    (block $sdone (loop $sloop
+      (br_if $sdone (i32.ge_u (local.get $i) (local.get $simdLen)))
+      (local.set $byteOff (i32.shl (local.get $i) (i32.const ${shift})))
+      (local.set $v (v128.load (i32.add (local.get $srcOff) (local.get $byteOff))))
+      ${simdOp}
+      (v128.store (i32.add (local.get $dstOff) (local.get $byteOff)) (local.get $v))
+      (local.set $i (i32.add (local.get $i) (i32.const ${vw})))
+      (br $sloop)))
+    ;; Scalar remainder
+    (block $rdone (loop $rloop
+      (br_if $rdone (i32.ge_u (local.get $i) (local.get $len)))
+      ${scalarStoreExpr}
+      (local.set $i (i32.add (local.get $i) (i32.const 1)))
+      (br $rloop)))
+    (call $__mkptr (i32.const ${PTR.TYPED}) (i32.const ${elemType}) (local.get $dstOff)))`
+}
+
+
+export default (ctx) => {
+  Object.assign(ctx.core.stdlibDeps, {
+    __byte_length: ['__ptr_type', '__ptr_offset', '__ptr_aux'],
+    __byte_offset: ['__ptr_type', '__ptr_offset', '__ptr_aux'],
+    __to_buffer: ['__ptr_type', '__ptr_offset', '__ptr_aux', '__mkptr'],
+  })
+
+  // .map/.filter invoke callbacks with arity 1 internally.
+  ctx.closure.floor = Math.max(ctx.closure.floor ?? 0, 1)
+
+  inc('__mkptr', '__alloc', '__len')
+
+  // === Runtime helpers: byte length, buffer coerce ===
+  // __typed_shift lives in core (needed by __len/__cap).
+
+  // __byte_length(ptr) — byte size for BUFFER/TYPED; 0 otherwise.
+  // BUFFER and owned TYPED store byteLen at [-8]. TYPED view (aux bit 3) stores byteLen
+  // at descriptor[0].
+  ctx.core.stdlib['__byte_length'] = `(func $__byte_length (param $ptr f64) (result i32)
+    (local $t i32) (local $off i32)
+    (local.set $t (call $__ptr_type (local.get $ptr)))
+    (if (result i32)
+      (i32.or
+        (i32.eq (local.get $t) (i32.const ${PTR.BUFFER}))
+        (i32.eq (local.get $t) (i32.const ${PTR.TYPED})))
+      (then
+        (local.set $off (call $__ptr_offset (local.get $ptr)))
+        (if (result i32)
+          (i32.and
+            (i32.eq (local.get $t) (i32.const ${PTR.TYPED}))
+            (i32.ne (i32.and (call $__ptr_aux (local.get $ptr)) (i32.const 8)) (i32.const 0)))
+          (then (i32.load (local.get $off)))
+          (else (i32.load (i32.sub (local.get $off) (i32.const 8))))))
+      (else (i32.const 0))))`
+
+  // __to_buffer(ptr) — return a BUFFER aliasing the same bytes (zero-copy view).
+  // BUFFER: passthrough.
+  // Owned TYPED: retag as BUFFER at same offset — the byteLen header is shared.
+  // TYPED view: retag as BUFFER at the parent data offset (descriptor[8]) — reconstructs
+  // the root ArrayBuffer so its own header supplies byteLength.
+  ctx.core.stdlib['__to_buffer'] = `(func $__to_buffer (param $ptr f64) (result f64)
+    (local $t i32) (local $off i32)
+    (local.set $t (call $__ptr_type (local.get $ptr)))
+    (if (result f64) (i32.eq (local.get $t) (i32.const ${PTR.BUFFER}))
+      (then (local.get $ptr))
+      (else
+        (local.set $off (call $__ptr_offset (local.get $ptr)))
+        (if (result f64)
+          (i32.and
+            (i32.eq (local.get $t) (i32.const ${PTR.TYPED}))
+            (i32.ne (i32.and (call $__ptr_aux (local.get $ptr)) (i32.const 8)) (i32.const 0)))
+          (then (call $__mkptr (i32.const ${PTR.BUFFER}) (i32.const 0)
+                  (i32.load (i32.add (local.get $off) (i32.const 8)))))
+          (else (call $__mkptr (i32.const ${PTR.BUFFER}) (i32.const 0) (local.get $off)))))))`
+
+  // Constructor: new Float64Array(len) | new F64Array(arr) | new F64Array(buf) | new F64Array(buf, off, len)
+  for (const [name, elemType] of Object.entries(ELEM)) {
+    const stride = STRIDE[elemType]
+    ctx.core.emit[`new.${name}`] = (lenExpr, offsetExpr, lenExpr2) => {
+      ctx.features.typedarray = true
+      const srcType = typeof lenExpr === 'string' ? lookupValType(lenExpr) : valTypeOf(lenExpr)
+      // Subview: new TypedArray(buffer, byteOffset, length) — true JS-parity view.
+      // Allocates a 16-byte descriptor [byteLen:i32][dataOff:i32][parentOff:i32][pad]
+      // and tags the TYPED ptr with aux=elemType|8. Reads/writes alias the parent,
+      // .buffer reconstructs the root BUFFER, .byteOffset = dataOff - parentOff.
+      if (offsetExpr != null && lenExpr2 != null) {
+        const src = temp('tvs')
+        const parentOff = tempI32('tvp')
+        const byteLen = tempI32('tvb')
+        const dst = tempI32('tvd')
+        return typed(['block', ['result', 'f64'],
+          ['local.set', `$${src}`, asF64(emit(lenExpr))],
+          ['local.set', `$${parentOff}`, ptrOffsetIR(['local.get', `$${src}`], srcType)],
+          ['local.set', `$${byteLen}`, ['i32.mul', asI32(emit(lenExpr2)), ['i32.const', stride]]],
+          ['local.set', `$${dst}`, ['call', '$__alloc', ['i32.const', 16]]],
+          ['i32.store', ['local.get', `$${dst}`], ['local.get', `$${byteLen}`]],
+          ['i32.store',
+            ['i32.add', ['local.get', `$${dst}`], ['i32.const', 4]],
+            ['i32.add', ['local.get', `$${parentOff}`], asI32(emit(offsetExpr))]],
+          ['i32.store',
+            ['i32.add', ['local.get', `$${dst}`], ['i32.const', 8]],
+            ['local.get', `$${parentOff}`]],
+          mkPtrIR(PTR.TYPED, elemType | 8, ['local.get', `$${dst}`])], 'f64')
+      }
+      // Single arg array-like source: copy elements instead of treating the pointer as a length.
+      if (srcType === VAL.ARRAY && ctx.core.emit[`${name}.from`])
+        return ctx.core.emit[`${name}.from`](lenExpr)
+      // Reinterpret on a buffer or another typed array: zero-copy view.
+      // TYPED retagged at the same offset — the byteLen header is shared with the parent.
+      // __len(view) = byteLen >> shift computes elemCount for this view's elemType.
+      if (srcType === VAL.BUFFER || srcType === VAL.TYPED) {
+        return mkPtrIR(PTR.TYPED, elemType, ['call', '$__ptr_offset', asF64(emit(lenExpr))])
+      }
+      if (srcType == null && ctx.core.emit[`${name}.from`]) {
+        // Runtime dispatch: number → allocate; array → copy elements; buffer/typed → zero-copy view.
+        const src = temp('ts')
+        const len = tempI32('tl')
+        const shift = SHIFT[elemType]
+        const numBytes = ['i32.shl', ['local.get', `$${len}`], ['i32.const', shift]]
+        const numAlloc = allocPtr({ type: PTR.TYPED, aux: elemType, len: numBytes, stride: 1, tag: 'ta' })
+        return typed(['block', ['result', 'f64'],
+          ['local.set', `$${src}`, asF64(emit(lenExpr))],
+          ['if', ['result', 'f64'],
+            ['f64.eq', ['local.get', `$${src}`], ['local.get', `$${src}`]],
+            // Regular number: treat as length, allocate fresh typed array with byteLen header
+            ['then', ['block', ['result', 'f64'],
+              ['local.set', `$${len}`, ['i32.trunc_sat_f64_s', ['local.get', `$${src}`]]],
+              numAlloc.init,
+              numAlloc.ptr]],
+            // Pointer: array → copy elements; buffer/typed → zero-copy view on same offset
+            ['else', ['if', ['result', 'f64'],
+              ['i32.eq', ['call', '$__ptr_type', ['local.get', `$${src}`]], ['i32.const', PTR.ARRAY]],
+              ['then', ctx.core.emit[`${name}.from`](src)],
+              ['else', mkPtrIR(PTR.TYPED, elemType,
+                ['call', '$__ptr_offset', ['local.get', `$${src}`]])]]]]], 'f64')
+      }
+      // Normal: allocate fresh typed array (lenExpr is numeric size). Header stores byteLen.
+      const shift = SHIFT[elemType]
+      const lenL = tempI32('tan')
+      const out = allocPtr({ type: PTR.TYPED, aux: elemType,
+        len: ['i32.shl', ['local.get', `$${lenL}`], ['i32.const', shift]], stride: 1, tag: 'ta' })
+      return typed(['block', ['result', 'f64'],
+        ['local.set', `$${lenL}`, asI32(emit(lenExpr))],
+        out.init,
+        out.ptr], 'f64')
+    }
+  }
+
+  // === ArrayBuffer (PTR.BUFFER) and DataView ===
+  // ArrayBuffer: first-class byte storage with [-8:byteLen][-4:byteCap][bytes].
+  // DataView: passthrough ptr to the same BUFFER — DataView methods operate on raw bytes via offset.
+
+  // new ArrayBuffer(n) → allocate n bytes, return as BUFFER pointer
+  const arrayBufferCtor = (sizeExpr) => {
+    const n = asI32(emit(sizeExpr))
+    const out = allocPtr({ type: PTR.BUFFER, len: n, stride: 1, tag: 'ab' })
+    return typed(['block', ['result', 'f64'], out.init, out.ptr], 'f64')
+  }
+  ctx.core.emit['new.ArrayBuffer'] = arrayBufferCtor
+
+  // new DataView(buffer) → same ptr (BUFFER). Its type tag may differ from BUFFER but methods ignore type.
+  ctx.core.emit['new.DataView'] = (bufExpr) => asF64(emit(bufExpr))
+
+  // BigInt64Array(buffer) (bare form, legacy): coerce to same data, Float64Array-compatible storage.
+  ctx.core.emit['BigInt64Array'] = (bufExpr) => {
+    ctx.features.typedarray = true
+    const va = asF64(emit(bufExpr))
+    return mkPtrIR(PTR.TYPED, 7, ['call', '$__ptr_offset', va])
+  }
+
+  // .buffer — always aliased (zero-copy). BUFFER/DataView: passthrough.
+  // Owned TYPED: retag as BUFFER at same offset — the byteLen header is shared.
+  // TYPED view: BUFFER at descriptor[8] (root parent data offset).
+  ctx.core.emit['.buffer'] = (obj) => {
+    if (typeof obj === 'string') {
+      const ctor = ctx.types.typedElem?.get(obj)
+      if (ctor === 'new.ArrayBuffer' || ctor === 'new.DataView') return asF64(emit(obj))
+      if (ctor?.startsWith('new.')) {
+        const isView = ctor.endsWith('.view')
+        const name = isView ? ctor.slice(4, -5) : ctor.slice(4)
+        if (ELEM[name] != null) {
+          const parentOff = isView
+            ? ['i32.load', ['i32.add', ptrOffsetIR(emit(obj), VAL.TYPED), ['i32.const', 8]]]
+            : ptrOffsetIR(emit(obj), VAL.TYPED)
+          return mkPtrIR(PTR.BUFFER, 0, parentOff)
+        }
+      }
+    }
+    inc('__to_buffer')
+    return typed(['call', '$__to_buffer', asF64(emit(obj))], 'f64')
+  }
+
+  // .byteLength — BUFFER: raw __len. Owned TYPED: elemCount * stride. View TYPED: descriptor[0].
+  ctx.core.emit['.byteLength'] = (obj) => {
+    if (typeof obj === 'string') {
+      const ctor = ctx.types.typedElem?.get(obj)
+      if (ctor === 'new.ArrayBuffer' || ctor === 'new.DataView') {
+        return typed(['f64.convert_i32_s', ['call', '$__len', asF64(emit(obj))]], 'f64')
+      }
+      if (ctor && ctor.startsWith('new.')) {
+        const isView = ctor.endsWith('.view')
+        const name = isView ? ctor.slice(4, -5) : ctor.slice(4)
+        const et = ELEM[name]
+        if (et != null) {
+          if (isView) {
+            return typed(['f64.convert_i32_s',
+              ['i32.load', ptrOffsetIR(emit(obj), VAL.TYPED)]], 'f64')
+          }
+          return typed(['f64.convert_i32_s',
+            ['i32.shl', ['call', '$__len', asF64(emit(obj))], ['i32.const', SHIFT[et]]]], 'f64')
+        }
+      }
+    }
+    inc('__byte_length')
+    return typed(['f64.convert_i32_s', ['call', '$__byte_length', asF64(emit(obj))]], 'f64')
+  }
+
+  // .byteOffset — owned: 0. View: descriptor[4] - descriptor[8].
+  ctx.core.emit['.byteOffset'] = (obj) => {
+    if (typeof obj === 'string') {
+      const ctor = ctx.types.typedElem?.get(obj)
+      if (ctor?.endsWith('.view')) {
+        const t = tempI32('bo')
+        return typed(['block', ['result', 'f64'],
+          ['local.set', `$${t}`, ptrOffsetIR(emit(obj), VAL.TYPED)],
+          ['f64.convert_i32_s',
+            ['i32.sub',
+              ['i32.load', ['i32.add', ['local.get', `$${t}`], ['i32.const', 4]]],
+              ['i32.load', ['i32.add', ['local.get', `$${t}`], ['i32.const', 8]]]]]], 'f64')
+      }
+      if (ctor?.startsWith('new.') && ELEM[ctor.slice(4)] != null) return typed(['f64.const', 0], 'f64')
+    }
+    inc('__byte_offset')
+    return typed(['f64.convert_i32_s', ['call', '$__byte_offset', asF64(emit(obj))]], 'f64')
+  }
+
+  // Runtime fallback for .byteOffset when variable view-ness is unknown.
+  ctx.core.stdlib['__byte_offset'] = `(func $__byte_offset (param $ptr f64) (result i32)
+    (local $off i32)
+    (if (result i32)
+      (i32.and
+        (i32.eq (call $__ptr_type (local.get $ptr)) (i32.const ${PTR.TYPED}))
+        (i32.ne (i32.and (call $__ptr_aux (local.get $ptr)) (i32.const 8)) (i32.const 0)))
+      (then
+        (local.set $off (call $__ptr_offset (local.get $ptr)))
+        (i32.sub
+          (i32.load (i32.add (local.get $off) (i32.const 4)))
+          (i32.load (i32.add (local.get $off) (i32.const 8)))))
+      (else (i32.const 0))))`
+
+  // ArrayBuffer.isView(x) — true iff x is a TYPED pointer. (DataView passthrough cannot be
+  // distinguished from ArrayBuffer since both are BUFFER pointers; both report false.)
+  ctx.core.emit['ArrayBuffer.isView'] = (v) => {
+    const va = asF64(emit(v))
+    return typed(['f64.convert_i32_s',
+      ['i32.eq', ['call', '$__ptr_type', va], ['i32.const', PTR.TYPED]]], 'f64')
+  }
+
+  // buf.slice(begin?, end?) on a BUFFER → fresh BUFFER with the byte range copied.
+  // Only dispatches statically when obj is a tracked ArrayBuffer/DataView variable.
+  ctx.core.emit['.buf:slice'] = (obj, beginExpr, endExpr) => {
+    const src = temp('bss')
+    const beg = tempI32('bsb')
+    const end = tempI32('bse')
+    const bytes = tempI32('bsn')
+    const out = allocPtr({ type: PTR.BUFFER, len: ['local.get', `$${bytes}`], stride: 1, tag: 'bsd' })
+    const beginWat = beginExpr == null ? ['i32.const', 0] : asI32(emit(beginExpr))
+    const endWat = endExpr == null
+      ? ['call', '$__len', ['local.get', `$${src}`]]
+      : asI32(emit(endExpr))
+    return typed(['block', ['result', 'f64'],
+      ['local.set', `$${src}`, asF64(emit(obj))],
+      ['local.set', `$${beg}`, beginWat],
+      ['local.set', `$${end}`, endWat],
+      ['local.set', `$${bytes}`, ['i32.sub', ['local.get', `$${end}`], ['local.get', `$${beg}`]]],
+      ['if',
+        ['i32.lt_s', ['local.get', `$${bytes}`], ['i32.const', 0]],
+        ['then', ['local.set', `$${bytes}`, ['i32.const', 0]]]],
+      out.init,
+      ['memory.copy',
+        ['local.get', `$${out.local}`],
+        ['i32.add', ['call', '$__ptr_offset', ['local.get', `$${src}`]], ['local.get', `$${beg}`]],
+        ['local.get', `$${bytes}`]],
+      out.ptr], 'f64')
+  }
+
+  // DataView set methods: extract i32 offset from f64 ptr, store value
+  const DV_SET = {
+    setInt8: 'i32.store8', setUint8: 'i32.store8',
+    setInt16: 'i32.store16', setUint16: 'i32.store16',
+    setInt32: 'i32.store', setUint32: 'i32.store',
+    setFloat32: 'f32.store', setFloat64: 'f64.store',
+    setBigInt64: 'i64.store', setBigUint64: 'i64.store',
+  }
+  for (const [method, storeOp] of Object.entries(DV_SET)) {
+    ctx.core.emit[`.${method}`] = (dv, off, val, _le) => {
+      const dvOff = ptrOffsetIR(emit(dv), VAL.BUFFER)
+      const addr = ['i32.add', dvOff, asI32(emit(off))]
+      let v = emit(val)
+      if (method.includes('BigInt') || method.includes('BigUint'))
+        v = typed(['i64.reinterpret_f64', asF64(v)], 'i64')
+      else if (method.includes('Float64')) v = asF64(v)
+      else if (method.includes('Float32')) v = typed(['f32.demote_f64', asF64(v)], 'f32')
+      else v = asI32(v)
+      return [storeOp, addr, v]
+    }
+  }
+
+  // DataView get methods: extract i32 offset, load value, return as f64
+  const DV_GET = {
+    getInt8: ['i32.load8_s', 'i32'], getUint8: ['i32.load8_u', 'i32'],
+    getInt16: ['i32.load16_s', 'i32'], getUint16: ['i32.load16_u', 'i32'],
+    getInt32: ['i32.load', 'i32'], getUint32: ['i32.load', 'i32'],
+    getFloat32: ['f32.load', 'f32'], getFloat64: ['f64.load', 'f64'],
+    getBigInt64: ['i64.load', 'i64'], getBigUint64: ['i64.load', 'i64'],
+  }
+  for (const [method, [loadOp, resultType]] of Object.entries(DV_GET)) {
+    ctx.core.emit[`.${method}`] = (dv, off, _le) => {
+      const addr = ['i32.add', ptrOffsetIR(emit(dv), VAL.BUFFER), asI32(emit(off))]
+      const raw = typed([loadOp, addr], resultType)
+      if (resultType === 'f64') return raw
+      if (resultType === 'f32') return typed(['f64.promote_f32', raw], 'f64')
+      if (resultType === 'i64') return typed(['f64.reinterpret_i64', raw], 'f64')
+      return typed(['f64.convert_i32_s', raw], 'f64')
+    }
+  }
+
+  // TypedArray.from(arr) — convert regular array to typed array
+  for (const [name, elemType] of Object.entries(ELEM)) {
+    const stride = STRIDE[elemType], store = STORE[elemType]
+    ctx.core.emit[`${name}.from`] = (src) => {
+      ctx.features.typedarray = true
+      const srcL = temp('tfs')
+      const len = tempI32('tfl'), i = tempI32('tfi'), off = tempI32('tfo')
+      const out = allocPtr({ type: PTR.TYPED, aux: elemType,
+        len: ['i32.mul', ['local.get', `$${len}`], ['i32.const', stride]], stride: 1, tag: 'tf' })
+      const t = out.local
+      const id = ctx.func.uniq++
+      const storeExpr = elemType === 7 ? ['f64.store',
+          ['i32.add', ['local.get', `$${t}`], ['i32.mul', ['local.get', `$${i}`], ['i32.const', stride]]],
+          ['f64.load', ['i32.add', ['local.get', `$${off}`], ['i32.shl', ['local.get', `$${i}`], ['i32.const', 3]]]]]
+        : elemType === 6 ? ['f32.store',
+          ['i32.add', ['local.get', `$${t}`], ['i32.mul', ['local.get', `$${i}`], ['i32.const', stride]]],
+          ['f32.demote_f64', ['f64.load', ['i32.add', ['local.get', `$${off}`], ['i32.shl', ['local.get', `$${i}`], ['i32.const', 3]]]]]]
+        : [store,
+          ['i32.add', ['local.get', `$${t}`], ['i32.mul', ['local.get', `$${i}`], ['i32.const', stride]]],
+          [(elemType & 1) ? 'i32.trunc_f64_u' : 'i32.trunc_f64_s',
+            ['f64.load', ['i32.add', ['local.get', `$${off}`], ['i32.shl', ['local.get', `$${i}`], ['i32.const', 3]]]]]]
+      return typed(['block', ['result', 'f64'],
+        ['local.set', `$${srcL}`, asF64(emit(src))],
+        ['local.set', `$${off}`, ['call', '$__ptr_offset', ['local.get', `$${srcL}`]]],
+        ['local.set', `$${len}`, ['call', '$__len', ['local.get', `$${srcL}`]]],
+        out.init,
+        ['local.set', `$${i}`, ['i32.const', 0]],
+        ['block', `$brk${id}`, ['loop', `$loop${id}`,
+          ['br_if', `$brk${id}`, ['i32.ge_s', ['local.get', `$${i}`], ['local.get', `$${len}`]]],
+          storeExpr,
+          ['local.set', `$${i}`, ['i32.add', ['local.get', `$${i}`], ['i32.const', 1]]],
+          ['br', `$loop${id}`]]],
+        out.ptr], 'f64')
+    }
+  }
+
+  // .length handled by ptr.js's __len (reads from memory header [-8:len])
+
+  /** Resolve element type + view-ness for a known TypedArray variable.
+   *  Returns { et, isView } or null. */
+  const resolveElem = (arr) => {
+    const ctor = typeof arr === 'string' && ctx.types.typedElem?.get(arr)
+    if (!ctor) return null
+    const isView = ctor.endsWith('.view')
+    const name = isView ? ctor.slice(4, -5) : ctor.slice(4)
+    const et = ELEM[name]
+    return et == null ? null : { et, isView }
+  }
+
+  /** Emit the real data byte-address for a typed array IR node.
+   *  Owned: low 32 bits of the NaN-box (or the unboxed local directly).
+   *  View: load descriptor[4]. Uses ptrOffsetIR so unboxed-TYPED locals pass through
+   *  without a rebox-then-unbox round trip, and globals fold to inline bit-extract. */
+  const typedDataAddr = (objIR, isView) => isView
+    ? ['i32.load', ['i32.add', ptrOffsetIR(objIR, VAL.TYPED), ['i32.const', 4]]]
+    : ptrOffsetIR(objIR, VAL.TYPED)
+
+  // Runtime-dispatch typed index: checks ptr_type + aux to load with correct stride.
+  // For TYPED views (aux bit 3), $off indirects through descriptor[4] to real data.
+  // Factory — collapses to ARRAY-only f64 indexing when no TYPED pointer can reach here.
+  // Identical factory in array.js; whichever module loads last wins the registration.
+  ctx.core.stdlib['__typed_idx'] = () => {
+    if (!ctx.features.typedarray && !ctx.features.external) {
+      return `(func $__typed_idx (param $ptr f64) (param $i i32) (result f64)
+    (local $len i32)
+    (local.set $len (call $__len (local.get $ptr)))
+    (if (result f64)
+      (i32.or
+        (i32.lt_s (local.get $i) (i32.const 0))
+        (i32.ge_u (local.get $i) (local.get $len)))
+      (then (f64.const nan:${UNDEF_NAN}))
+      (else (f64.load (i32.add (call $__ptr_offset (local.get $ptr)) (i32.shl (local.get $i) (i32.const 3)))))))`
+    }
+    return `(func $__typed_idx (param $ptr f64) (param $i i32) (result f64)
+    (local $off i32) (local $et i32) (local $len i32) (local $aux i32)
+    (local.set $aux (call $__ptr_aux (local.get $ptr)))
+    (local.set $off (call $__ptr_offset (local.get $ptr)))
+    (if
+      (i32.and
+        (i32.eq (call $__ptr_type (local.get $ptr)) (i32.const ${PTR.TYPED}))
+        (i32.ne (i32.and (local.get $aux) (i32.const 8)) (i32.const 0)))
+      (then (local.set $off (i32.load (i32.add (local.get $off) (i32.const 4))))))
+    (local.set $len (call $__len (local.get $ptr)))
+    (if (result f64)
+      (i32.or
+        (i32.lt_s (local.get $i) (i32.const 0))
+        (i32.ge_u (local.get $i) (local.get $len)))
+      (then (f64.const nan:${UNDEF_NAN}))
+      (else
+        (if (result f64) (i32.eq (call $__ptr_type (local.get $ptr)) (i32.const ${PTR.TYPED}))
+          (then
+            (local.set $et (i32.and (local.get $aux) (i32.const 7)))
+            (if (result f64) (i32.ge_u (local.get $et) (i32.const 6))
+              (then (if (result f64) (i32.eq (local.get $et) (i32.const 7))
+                (then (f64.load (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 3)))))
+                (else (f64.promote_f32 (f32.load (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 2))))))))
+              (else (if (result f64) (i32.ge_u (local.get $et) (i32.const 4))
+                (then (if (result f64) (i32.and (local.get $et) (i32.const 1))
+                  (then (f64.convert_i32_u (i32.load (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 2))))))
+                  (else (f64.convert_i32_s (i32.load (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 2))))))))
+                (else (if (result f64) (i32.ge_u (local.get $et) (i32.const 2))
+                  (then (if (result f64) (i32.and (local.get $et) (i32.const 1))
+                    (then (f64.convert_i32_u (i32.load16_u (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 1))))))
+                    (else (f64.convert_i32_s (i32.load16_s (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 1))))))))
+                  (else (if (result f64) (i32.and (local.get $et) (i32.const 1))
+                    (then (f64.convert_i32_u (i32.load8_u (i32.add (local.get $off) (local.get $i)))))
+                    (else (f64.convert_i32_s (i32.load8_s (i32.add (local.get $off) (local.get $i)))))))))))))
+          (else (f64.load (i32.add (local.get $off) (i32.shl (local.get $i) (i32.const 3)))))))))`
+  }
+
+  // Type-aware TypedArray read: arr[i]
+  ctx.core.emit['.typed:[]'] = (arr, idx) => {
+    const r = resolveElem(arr)
+    if (r == null) return null // unknown type, fallback to generic
+    const { et, isView } = r
+    const objIR = emit(arr), vi = asI32(emit(idx))
+    const off = ['i32.add', typedDataAddr(objIR, isView), ['i32.shl', vi, ['i32.const', SHIFT[et]]]]
+    if (et === 7) return typed(['f64.load', off], 'f64') // Float64Array
+    if (et === 6) return typed(['f64.promote_f32', ['f32.load', off]], 'f64') // Float32Array
+    // Integer types: load and convert to f64 (unsigned types use unsigned conversion)
+    return typed([(et & 1) ? 'f64.convert_i32_u' : 'f64.convert_i32_s', [LOAD[et], off]], 'f64')
+  }
+
+  // Type-aware TypedArray write: arr[i] = val
+  ctx.core.emit['.typed:[]='] = (arr, idx, val) => {
+    const r = resolveElem(arr)
+    if (r == null) return null
+    const { et, isView } = r
+    const objIR = emit(arr), vi = asI32(emit(idx)), vv = asF64(emit(val))
+    const off = ['i32.add', typedDataAddr(objIR, isView), ['i32.shl', vi, ['i32.const', SHIFT[et]]]]
+    if (et === 7) return ['f64.store', off, vv] // Float64Array
+    if (et === 6) return ['f32.store', off, ['f32.demote_f64', vv]] // Float32Array
+    // Integer types: truncate f64 to i32, then store. Peel f64.convert_i32_*(x) → x:
+    // store of bitwise-result already-i32 needs no round-trip.
+    const isConv = Array.isArray(vv) && (vv[0] === 'f64.convert_i32_s' || vv[0] === 'f64.convert_i32_u')
+    const i32val = isConv ? vv[1] : [(et & 1) ? 'i32.trunc_f64_u' : 'i32.trunc_f64_s', vv]
+    return [STORE[et], off, i32val]
+  }
+
+  // .map() on TypedArrays — SIMD auto-vectorization when pattern detected
+  ctx.core.emit['.typed:map'] = (arr, fn) => {
+    // Resolve element type + view-ness from variable tracking
+    const ctor = typeof arr === 'string' && ctx.types.typedElem?.get(arr)
+    const isView = ctor?.endsWith('.view')
+    const elemName = isView ? ctor.slice(4, -5) : ctor?.slice(4)
+    const elemType = elemName && ELEM[elemName]
+
+    // Try SIMD: inline arrow with recognizable pattern
+    if (elemType != null && Array.isArray(fn) && fn[0] === '=>') {
+      const [, rawParam, body] = fn
+      const param = Array.isArray(rawParam) && rawParam[0] === '()' ? rawParam[1] : rawParam
+      const pattern = analyzeSimd(body, param)
+
+      if (pattern) {
+        const id = ctx.func.uniq++
+        const funcName = `__simd_map_${id}`
+        const wat = genSimdMap(funcName, elemType, pattern)
+        if (wat) {
+          ctx.core.stdlib[funcName] = wat
+          inc(funcName, '__typed_data', '__len')
+          return typed(['call', `$${funcName}`, asF64(emit(arr))], 'f64')
+        }
+      }
+    }
+
+    // Scalar fallback: proper typed-array map (preserves element type)
+    if (elemType != null) {
+      const va = emit(arr), vf = emit(fn)
+      const len = tempI32('tml'), ptr = tempI32('tmp'), i = tempI32('tmi')
+      const stride = STRIDE[elemType], shift = SHIFT[elemType]
+      const dst = allocPtr({ type: PTR.TYPED, aux: elemType,
+        len: ['i32.shl', ['local.get', `$${len}`], ['i32.const', shift]], stride: 1, tag: 'tmo' })
+      const out = dst.local
+
+      const loadElem = () => {
+        const off = ['i32.add', ['local.get', `$${ptr}`], ['i32.shl', ['local.get', `$${i}`], ['i32.const', shift]]]
+        if (elemType === 7) return typed(['f64.load', off], 'f64')
+        if (elemType === 6) return typed(['f64.promote_f32', ['f32.load', off]], 'f64')
+        return typed([(elemType & 1) ? 'f64.convert_i32_u' : 'f64.convert_i32_s', [LOAD[elemType], off]], 'f64')
+      }
+      const storeElem = (val) => {
+        const off = ['i32.add', ['local.get', `$${out}`], ['i32.shl', ['local.get', `$${i}`], ['i32.const', shift]]]
+        if (elemType === 7) return ['f64.store', off, val]
+        if (elemType === 6) return ['f32.store', off, ['f32.demote_f64', val]]
+        return [STORE[elemType], off, [(elemType & 1) ? 'i32.trunc_f64_u' : 'i32.trunc_f64_s', val]]
+      }
+
+      const id = ctx.func.uniq++
+      return typed(['block', ['result', 'f64'],
+        ['local.set', `$${ptr}`, typedDataAddr(va, isView)],
+        ['local.set', `$${len}`, ['call', '$__len', asF64(va)]],
+        dst.init,
+        ['local.set', `$${i}`, ['i32.const', 0]],
+        ['block', `$brk${id}`, ['loop', `$loop${id}`,
+          ['br_if', `$brk${id}`, ['i32.ge_s', ['local.get', `$${i}`], ['local.get', `$${len}`]]],
+          storeElem(asF64(ctx.closure.call(vf, [loadElem()]))),
+          ['local.set', `$${i}`, ['i32.add', ['local.get', `$${i}`], ['i32.const', 1]]],
+          ['br', `$loop${id}`]]],
+        dst.ptr], 'f64')
+    }
+
+    // Unknown typed array type: fall back to generic array .map
+    if (ctx.core.emit['.map']) return ctx.core.emit['.map'](arr, fn)
+    return null
+  }
+}