ecmascript_simd.js

/*
  vim: set ts=8 sts=2 et sw=2 tw=79:
  Copyright (C) 2013

  This software is provided 'as-is', without any express or implied
  warranty.  In no event will the authors be held liable for any damages
  arising from the use of this software.

  Permission is granted to anyone to use this software for any purpose,
  including commercial applications, and to alter it and redistribute it
  freely, subject to the following restrictions:

  1. The origin of this software must not be misrepresented; you must not
     claim that you wrote the original software. If you use this software
     in a product, an acknowledgment in the product documentation would be
     appreciated but is not required.
  2. Altered source versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
  3. This notice may not be removed or altered from any source distribution.
*/

// A conforming SIMD.js implementation may contain the following deviations to
// normal JS numeric behavior:
//  - Subnormal numbers may or may not be flushed to zero on input or output of
//    any SIMD operation.

// Many of the operations in SIMD.js have semantics which correspond to scalar
// operations in JS, however there are a few differences:
//  - Vector shifts don't mask the shift count.
//  - Conversions from float to int32 throw on error.
//  - Load and store operations throw when out of bounds.

(function(global) {

if (typeof global.SIMD === "undefined") {
  // SIMD module.
  global.SIMD = {};
}

if (typeof module !== "undefined") {
  // For CommonJS modules
  module.exports = global.SIMD;
}

var SIMD = global.SIMD;

// Buffers for bit casting and coercing lane values to those representable in
// the underlying lane type.
var _f32x4 = new Float32Array(4);
var _f64x2 = new Float64Array(_f32x4.buffer);
var _i32x4 = new Int32Array(_f32x4.buffer);
var _i16x8 = new Int16Array(_f32x4.buffer);
var _i8x16 = new Int8Array(_f32x4.buffer);
var _ui32x4 = new Uint32Array(_f32x4.buffer);
var _ui16x8 = new Uint16Array(_f32x4.buffer);
var _ui8x16 = new Uint8Array(_f32x4.buffer);

function convertValue(buffer, value) {
  buffer[0] = value;
  return buffer[0];
}

function convertArray(buffer, array) {
  for (var i = 0; i < array.length; i++)
    array[i] = convertValue(buffer, array[i]);
  return array;
}

// Utility functions.

function isInt32(o) {
  return (o | 0) === o;
}

function isTypedArray(o) {
  return (o instanceof Int8Array) ||
         (o instanceof Uint8Array) ||
         (o instanceof Uint8ClampedArray) ||
         (o instanceof Int16Array) ||
         (o instanceof Uint16Array) ||
         (o instanceof Int32Array) ||
         (o instanceof Uint32Array) ||
         (o instanceof Float32Array) ||
         (o instanceof Float64Array);
}

function minNum(x, y) {
  return x != x ? y :
         y != y ? x :
         Math.min(x, y);
}

function maxNum(x, y) {
  return x != x ? y :
         y != y ? x :
         Math.max(x, y);
}

function clamp(a, min, max) {
  if (a < min)
    return min;
  if (a > max)
    return max;
  return a;
}

// SIMD implementation functions

function simdCheckLaneIndex(index, lanes) {
  if (!isInt32(index))
    throw new TypeError('Lane index must be an int32');
  if (index < 0 || index >= lanes)
    throw new RangeError('Lane index must be in bounds');
}

// Global lanes array for constructing SIMD values.
var lanes = [];

function simdCreate(type) {
  return type.fn.apply(type.fn, lanes);
}

function simdToString(type, a) {
  a = type.fn.check(a);
  var str = "SIMD." + type.name + "(";
  str += type.fn.extractLane(a, 0);
  for (var i = 1; i < type.lanes; i++) {
    str += ", " + type.fn.extractLane(a, i);
  }
  return str + ")";
}

function simdToLocaleString(type, a) {
  a = type.fn.check(a);
  var str = "SIMD." + type.name + "(";
  str += type.fn.extractLane(a, 0).toLocaleString();
  for (var i = 1; i < type.lanes; i++) {
    str += ", " + type.fn.extractLane(a, i).toLocaleString();
  }
  return str + ")";
}

function simdSplat(type, s) {
  for (var i = 0; i < type.lanes; i++)
    lanes[i] = s;
  return simdCreate(type);
}

function simdReplaceLane(type, a, i, s) {
  a = type.fn.check(a);
  simdCheckLaneIndex(i, type.lanes);
  for (var j = 0; j < type.lanes; j++)
    lanes[j] = type.fn.extractLane(a, j);
  lanes[i] = s;
  return simdCreate(type);
}

function simdFrom(toType, fromType, a) {
  a = fromType.fn.check(a);
  for (var i = 0; i < fromType.lanes; i++) {
    var v = fromType.fn.extractLane(a, i);
    if (toType.minVal !== undefined &&
        (v < toType.minVal || v > toType.maxVal)) {
      throw new RangeError("Can't convert value");
    }
    lanes[i] = v;
  }
  return simdCreate(toType);
}

function simdFromBits(toType, fromType, a) {
  a = fromType.fn.check(a);
  var newValue = new toType.fn();
  newValue.s_ = new toType.view(a.s_.buffer);
  return newValue;
}

function simdSelect(type, selector, a, b) {
  selector = type.boolType.fn.check(selector);
  a = type.fn.check(a);
  b = type.fn.check(b);
  for (var i = 0; i < type.lanes; i++) {
    lanes[i] = type.boolType.fn.extractLane(selector, i) ?
               type.fn.extractLane(a, i) : type.fn.extractLane(b, i);
  }
  return simdCreate(type);
}

function simdSwizzle(type, a, indices) {
  a = type.fn.check(a);
  for (var i = 0; i < indices.length; i++) {
    simdCheckLaneIndex(indices[i], type.lanes);
    lanes[i] = type.fn.extractLane(a, indices[i]);
  }
  return simdCreate(type);
}

function simdShuffle(type, a, b, indices) {
  a = type.fn.check(a);
  b = type.fn.check(b);
  for (var i = 0; i < indices.length; i++) {
    simdCheckLaneIndex(indices[i], 2 * type.lanes);
    lanes[i] = indices[i] < type.lanes ?
               type.fn.extractLane(a, indices[i]) :
               type.fn.extractLane(b, indices[i] - type.lanes);
  }
  return simdCreate(type);
}

function unaryNeg(a) { return -a; }
function unaryBitwiseNot(a) { return ~a; }
function unaryLogicalNot(a) { return !a; }

function simdUnaryOp(type, op, a) {
  a = type.fn.check(a);
  for (var i = 0; i < type.lanes; i++)
    lanes[i] = op(type.fn.extractLane(a, i));
  return simdCreate(type);
}

function binaryAnd(a, b) { return a & b; }
function binaryOr(a, b) { return a | b; }
function binaryXor(a, b) { return a ^ b; }
function binaryAdd(a, b) { return a + b; }
function binarySub(a, b) { return a - b; }
function binaryMul(a, b) { return a * b; }
function binaryDiv(a, b) { return a / b; }

var binaryImul;
if (typeof Math.imul !== 'undefined') {
  binaryImul = Math.imul;
} else {
  binaryImul = function(a, b) {
    var ah = (a >>> 16) & 0xffff;
    var al = a & 0xffff;
    var bh = (b >>> 16) & 0xffff;
    var bl = b & 0xffff;
    // the shift by 0 fixes the sign on the high part
    // the final |0 converts the unsigned value into a signed value
    return ((al * bl) + (((ah * bl + al * bh) << 16) >>> 0)|0);
  };
}

function simdBinaryOp(type, op, a, b) {
  a = type.fn.check(a);
  b = type.fn.check(b);
  for (var i = 0; i < type.lanes; i++)
    lanes[i] = op(type.fn.extractLane(a, i), type.fn.extractLane(b, i));
  return simdCreate(type);
}

function binaryEqual(a, b) { return a == b; }
function binaryNotEqual(a, b) { return a != b; }
function binaryLess(a, b) { return a < b; }
function binaryLessEqual(a, b) { return a <= b; }
function binaryGreater(a, b) { return a > b; }
function binaryGreaterEqual(a, b) { return a >= b; }

function simdRelationalOp(type, op, a, b) {
  a = type.fn.check(a);
  b = type.fn.check(b);
  for (var i = 0; i < type.lanes; i++)
    lanes[i] = op(type.fn.extractLane(a, i), type.fn.extractLane(b, i));
  return simdCreate(type.boolType);
}

function simdAnyTrue(type, a) {
  a = type.fn.check(a);
  for (var i = 0; i < type.lanes; i++)
    if (type.fn.extractLane(a, i)) return true;
  return false;
}

function simdAllTrue(type, a) {
  a = type.fn.check(a);
  for (var i = 0; i < type.lanes; i++)
    if (!type.fn.extractLane(a, i)) return false;
  return true;
}

function binaryShiftLeft(a, bits) { return a << bits; }
function binaryShiftRightArithmetic(a, bits) { return a >> bits; }
function binaryShiftRightLogical(a, bits) { return a >>> bits; }

function simdShiftOp(type, op, a, bits) {
  a = type.fn.check(a);
  for (var i = 0; i < type.lanes; i++)
    lanes[i] = op(type.fn.extractLane(a, i), bits);
  return simdCreate(type);
}

function simdLoad(type, tarray, index, count) {
  if (!isTypedArray(tarray))
    throw new TypeError("The 1st argument must be a typed array.");
  if (!isInt32(index))
    throw new TypeError("The 2nd argument must be an Int32.");
  var bpe = tarray.BYTES_PER_ELEMENT;
  var bytes = count * type.laneSize;
  if (index < 0 || (index * bpe + bytes) > tarray.byteLength)
    throw new RangeError("The value of index is invalid.");

  var newValue = type.fn();
  var dst = new Uint8Array(newValue.s_.buffer);
  var src = new Uint8Array(tarray.buffer, tarray.byteOffset + index * bpe, bytes);

  for (var i = 0; i < bytes; i++) {
    dst[i] = src[i];
  }

  // XXX Emscripten: Work around to fix bug https://github.com/tc39/ecmascript_simd/issues/313
  for (var i = bytes; i < 16; ++i) {
    dst[i] = 0;
  }
  // XXX Emscripten

  return newValue;
}

function simdStore(type, tarray, index, a, count) {
  if (!isTypedArray(tarray))
    throw new TypeError("The 1st argument must be a typed array.");
  if (!isInt32(index))
    throw new TypeError("The 2nd argument must be an Int32.");
  var bpe = tarray.BYTES_PER_ELEMENT;
  var bytes = count * type.laneSize;
  if (index < 0 || (index * bpe + bytes) > tarray.byteLength)
    throw new RangeError("The value of index is invalid.");

  a = type.fn.check(a);

  // The underlying buffers are copied byte by byte, to avoid float
  // canonicalization.
  var src = new Uint8Array(a.s_.buffer);
  var dst = new Uint8Array(tarray.buffer, tarray.byteOffset + index * bpe, bytes);
  for (var i = 0; i < bytes; i++) {
    dst[i] = src[i];
  }
  return a;
}

// Constructors and extractLane functions are closely related and must be
// polyfilled together.

// Float32x4
if (typeof SIMD.Float32x4 === "undefined" ||
    typeof SIMD.Float32x4.extractLane === "undefined") {
  SIMD.Float32x4 = function(s0, s1, s2, s3) {
    if (!(this instanceof SIMD.Float32x4)) {
      return new SIMD.Float32x4(s0, s1, s2, s3);
    }
    this.s_ = convertArray(_f32x4, new Float32Array([s0, s1, s2, s3]));
  }

  SIMD.Float32x4.extractLane = function(v, i) {
    v = SIMD.Float32x4.check(v);
    simdCheckLaneIndex(i, 4);
    return v.s_[i];
  }
}

// Miscellaneous functions that aren't easily parameterized on type.

if (typeof SIMD.Float32x4.swizzle === "undefined") {
  SIMD.Float32x4.swizzle = function(a, s0, s1, s2, s3) {
    return simdSwizzle(float32x4, a, [s0, s1, s2, s3]);
  }
}

if (typeof SIMD.Float32x4.shuffle === "undefined") {
  SIMD.Float32x4.shuffle = function(a, b, s0, s1, s2, s3) {
    return simdShuffle(float32x4, a, b, [s0, s1, s2, s3]);
  }
}

// Int32x4
if (typeof SIMD.Int32x4 === "undefined" ||
    typeof SIMD.Int32x4.extractLane === "undefined") {
  SIMD.Int32x4 = function(s0, s1, s2, s3) {
    if (!(this instanceof SIMD.Int32x4)) {
      return new SIMD.Int32x4(s0, s1, s2, s3);
    }
    this.s_ = convertArray(_i32x4, new Int32Array([s0, s1, s2, s3]));
  }

  SIMD.Int32x4.extractLane = function(v, i) {
    v = SIMD.Int32x4.check(v);
    simdCheckLaneIndex(i, 4);
    return v.s_[i];
  }
}

if (typeof SIMD.Int32x4.swizzle === "undefined") {
  SIMD.Int32x4.swizzle = function(a, s0, s1, s2, s3) {
    return simdSwizzle(int32x4, a, [s0, s1, s2, s3]);
  }
}

if (typeof SIMD.Int32x4.shuffle === "undefined") {
  SIMD.Int32x4.shuffle = function(a, b, s0, s1, s2, s3) {
    return simdShuffle(int32x4, a, b, [s0, s1, s2, s3]);
  }
}

// Int16x8
if (typeof SIMD.Int16x8 === "undefined" ||
    typeof SIMD.Int16x8.extractLane === "undefined") {
  SIMD.Int16x8 = function(s0, s1, s2, s3, s4, s5, s6, s7) {
    if (!(this instanceof SIMD.Int16x8)) {
      return new SIMD.Int16x8(s0, s1, s2, s3, s4, s5, s6, s7);
    }
    this.s_ = convertArray(_i16x8, new Int16Array([s0, s1, s2, s3, s4, s5, s6, s7]));
  }

  SIMD.Int16x8.extractLane = function(v, i) {
    v = SIMD.Int16x8.check(v);
    simdCheckLaneIndex(i, 8);
    return v.s_[i];
  }
}

if (typeof SIMD.Int16x8.swizzle === "undefined") {
  SIMD.Int16x8.swizzle = function(a, s0, s1, s2, s3, s4, s5, s6, s7) {
    return simdSwizzle(int16x8, a, [s0, s1, s2, s3, s4, s5, s6, s7]);
  }
}

if (typeof SIMD.Int16x8.shuffle === "undefined") {
  SIMD.Int16x8.shuffle = function(a, b, s0, s1, s2, s3, s4, s5, s6, s7) {
    return simdShuffle(int16x8, a, b, [s0, s1, s2, s3, s4, s5, s6, s7]);
  }
}

// Int8x16
if (typeof SIMD.Int8x16 === "undefined" ||
    typeof SIMD.Int8x16.extractLane === "undefined") {
  SIMD.Int8x16 = function(s0, s1, s2, s3, s4, s5, s6, s7,
                          s8, s9, s10, s11, s12, s13, s14, s15) {
    if (!(this instanceof SIMD.Int8x16)) {
      return new SIMD.Int8x16(s0, s1, s2, s3, s4, s5, s6, s7,
                              s8, s9, s10, s11, s12, s13, s14, s15);
    }
    this.s_ = convertArray(_i8x16, new Int8Array([s0, s1, s2, s3, s4, s5, s6, s7,
                                    s8, s9, s10, s11, s12, s13, s14, s15]));
  }

  SIMD.Int8x16.extractLane = function(v, i) {
    v = SIMD.Int8x16.check(v);
    simdCheckLaneIndex(i, 16);
    return v.s_[i];
  }
}

if (typeof SIMD.Int8x16.swizzle === "undefined") {
  SIMD.Int8x16.swizzle = function(a, s0, s1, s2, s3, s4, s5, s6, s7,
                                     s8, s9, s10, s11, s12, s13, s14, s15) {
    return simdSwizzle(int8x16, a, [s0, s1, s2, s3, s4, s5, s6, s7,
                                    s8, s9, s10, s11, s12, s13, s14, s15]);
  }
}

if (typeof SIMD.Int8x16.shuffle === "undefined") {
  SIMD.Int8x16.shuffle = function(a, b, s0, s1, s2, s3, s4, s5, s6, s7,
                                        s8, s9, s10, s11, s12, s13, s14, s15) {
    return simdShuffle(int8x16, a, b, [s0, s1, s2, s3, s4, s5, s6, s7,
                                       s8, s9, s10, s11, s12, s13, s14, s15]);
  }
}

// Uint32x4
if (typeof SIMD.Uint32x4 === "undefined" ||
    typeof SIMD.Uint32x4.extractLane === "undefined") {
  SIMD.Uint32x4 = function(s0, s1, s2, s3) {
    if (!(this instanceof SIMD.Uint32x4)) {
      return new SIMD.Uint32x4(s0, s1, s2, s3);
    }
    this.s_ = convertArray(_ui32x4, new Uint32Array([s0, s1, s2, s3]));
  }

  SIMD.Uint32x4.extractLane = function(v, i) {
    v = SIMD.Uint32x4.check(v);
    simdCheckLaneIndex(i, 4);
    return v.s_[i];
  }
}

if (typeof SIMD.Uint32x4.swizzle === "undefined") {
  SIMD.Uint32x4.swizzle = function(a, s0, s1, s2, s3) {
    return simdSwizzle(uint32x4, a, [s0, s1, s2, s3]);
  }
}

if (typeof SIMD.Uint32x4.shuffle === "undefined") {
  SIMD.Uint32x4.shuffle = function(a, b, s0, s1, s2, s3) {
    return simdShuffle(uint32x4, a, b, [s0, s1, s2, s3]);
  }
}

// Uint16x8
if (typeof SIMD.Uint16x8 === "undefined" ||
    typeof SIMD.Uint16x8.extractLane === "undefined") {
  SIMD.Uint16x8 = function(s0, s1, s2, s3, s4, s5, s6, s7) {
    if (!(this instanceof SIMD.Uint16x8)) {
      return new SIMD.Uint16x8(s0, s1, s2, s3, s4, s5, s6, s7);
    }
    this.s_ = convertArray(_ui16x8, new Uint16Array([s0, s1, s2, s3, s4, s5, s6, s7]));
  }

  SIMD.Uint16x8.extractLane = function(v, i) {
    v = SIMD.Uint16x8.check(v);
    simdCheckLaneIndex(i, 8);
    return v.s_[i];
  }
}

if (typeof SIMD.Uint16x8.swizzle === "undefined") {
  SIMD.Uint16x8.swizzle = function(a, s0, s1, s2, s3, s4, s5, s6, s7) {
    return simdSwizzle(uint16x8, a, [s0, s1, s2, s3, s4, s5, s6, s7]);
  }
}

if (typeof SIMD.Uint16x8.shuffle === "undefined") {
  SIMD.Uint16x8.shuffle = function(a, b, s0, s1, s2, s3, s4, s5, s6, s7) {
    return simdShuffle(uint16x8, a, b, [s0, s1, s2, s3, s4, s5, s6, s7]);
  }
}

// Uint8x16
if (typeof SIMD.Uint8x16 === "undefined" ||
    typeof SIMD.Uint8x16.extractLane === "undefined") {
  SIMD.Uint8x16 = function(s0, s1, s2, s3, s4, s5, s6, s7,
                           s8, s9, s10, s11, s12, s13, s14, s15) {
    if (!(this instanceof SIMD.Uint8x16)) {
      return new SIMD.Uint8x16(s0, s1, s2, s3, s4, s5, s6, s7,
                               s8, s9, s10, s11, s12, s13, s14, s15);
    }
    this.s_ = convertArray(_ui8x16, new Uint8Array([s0, s1, s2, s3, s4, s5, s6, s7,
                                     s8, s9, s10, s11, s12, s13, s14, s15]));
  }

  SIMD.Uint8x16.extractLane = function(v, i) {
    v = SIMD.Uint8x16.check(v);
    simdCheckLaneIndex(i, 16);
    return v.s_[i];
  }
}

if (typeof SIMD.Uint8x16.swizzle === "undefined") {
  SIMD.Uint8x16.swizzle = function(a, s0, s1, s2, s3, s4, s5, s6, s7,
                                      s8, s9, s10, s11, s12, s13, s14, s15) {
    return simdSwizzle(uint8x16, a, [s0, s1, s2, s3, s4, s5, s6, s7,
                                     s8, s9, s10, s11, s12, s13, s14, s15]);
  }
}

if (typeof SIMD.Uint8x16.shuffle === "undefined") {
  SIMD.Uint8x16.shuffle = function(a, b, s0, s1, s2, s3, s4, s5, s6, s7,
                                         s8, s9, s10, s11, s12, s13, s14, s15) {
    return simdShuffle(uint8x16, a, b, [s0, s1, s2, s3, s4, s5, s6, s7,
                                        s8, s9, s10, s11, s12, s13, s14, s15]);
  }
}

// Bool32x4
if (typeof SIMD.Bool32x4 === "undefined" ||
    typeof SIMD.Bool32x4.extractLane === "undefined") {
  SIMD.Bool32x4 = function(s0, s1, s2, s3) {
    if (!(this instanceof SIMD.Bool32x4)) {
      return new SIMD.Bool32x4(s0, s1, s2, s3);
    }
    this.s_ = [!!s0, !!s1, !!s2, !!s3];
  }

  SIMD.Bool32x4.extractLane = function(v, i) {
    v = SIMD.Bool32x4.check(v);
    simdCheckLaneIndex(i, 4);
    return v.s_[i];
  }
}

// Bool16x8
if (typeof SIMD.Bool16x8 === "undefined" ||
    typeof SIMD.Bool16x8.extractLane === "undefined") {
  SIMD.Bool16x8 = function(s0, s1, s2, s3, s4, s5, s6, s7) {
    if (!(this instanceof SIMD.Bool16x8)) {
      return new SIMD.Bool16x8(s0, s1, s2, s3, s4, s5, s6, s7);
    }
    this.s_ = [!!s0, !!s1, !!s2, !!s3, !!s4, !!s5, !!s6, !!s7];
  }

  SIMD.Bool16x8.extractLane = function(v, i) {
    v = SIMD.Bool16x8.check(v);
    simdCheckLaneIndex(i, 8);
    return v.s_[i];
  }
}

// Bool8x16
if (typeof SIMD.Bool8x16 === "undefined" ||
    typeof SIMD.Bool8x16.extractLane === "undefined") {
  SIMD.Bool8x16 = function(s0, s1, s2, s3, s4, s5, s6, s7,
                           s8, s9, s10, s11, s12, s13, s14, s15) {
    if (!(this instanceof SIMD.Bool8x16)) {
      return new SIMD.Bool8x16(s0, s1, s2, s3, s4, s5, s6, s7,
                               s8, s9, s10, s11, s12, s13, s14, s15);
    }
    this.s_ = [!!s0, !!s1, !!s2, !!s3, !!s4, !!s5, !!s6, !!s7,
               !!s8, !!s9, !!s10, !!s11, !!s12, !!s13, !!s14, !!s15];
  }

  SIMD.Bool8x16.extractLane = function(v, i) {
    v = SIMD.Bool8x16.check(v);
    simdCheckLaneIndex(i, 16);
    return v.s_[i];
  }
}

// Type data to generate the remaining functions.

var float32x4 = {
  name: "Float32x4",
  fn: SIMD.Float32x4,
  lanes: 4,
  laneSize: 4,
  buffer: _f32x4,
  view: Float32Array,
  mulFn: binaryMul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "add", "sub", "mul", "div", "neg", "abs", "min", "max", "minNum", "maxNum",
        "reciprocalApproximation", "reciprocalSqrtApproximation", "sqrt",
        "load", "load1", "load2", "load3", "store", "store1", "store2", "store3"],
}

var int32x4 = {
  name: "Int32x4",
  fn: SIMD.Int32x4,
  lanes: 4,
  laneSize: 4,
  minVal: -0x80000000,
  maxVal: 0x7FFFFFFF,
  buffer: _i32x4,
  notFn: unaryBitwiseNot,
  view: Int32Array,
  mulFn: binaryImul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul", "neg",
        "shiftLeftByScalar", "shiftRightByScalar",
        "load", "load1", "load2", "load3", "store", "store1", "store2", "store3"],
}

var int16x8 = {
  name: "Int16x8",
  fn: SIMD.Int16x8,
  lanes: 8,
  laneSize: 2,
  minVal: -0x8000,
  maxVal: 0x7FFF,
  buffer: _i16x8,
  notFn: unaryBitwiseNot,
  view: Int16Array,
  mulFn: binaryMul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul", "neg",
        "shiftLeftByScalar", "shiftRightByScalar",
        "addSaturate", "subSaturate",
        "load", "store"],
}

var int8x16 = {
  name: "Int8x16",
  fn: SIMD.Int8x16,
  lanes: 16,
  laneSize: 1,
  minVal: -0x80,
  maxVal: 0x7F,
  buffer: _i8x16,
  notFn: unaryBitwiseNot,
  view: Int8Array,
  mulFn: binaryMul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul", "neg",
        "shiftLeftByScalar", "shiftRightByScalar",
        "addSaturate", "subSaturate",
        "load", "store"],
}

var uint32x4 = {
  name: "Uint32x4",
  fn: SIMD.Uint32x4,
  lanes: 4,
  laneSize: 4,
  minVal: 0,
  maxVal: 0xFFFFFFFF,
  unsigned: true,
  buffer: _ui32x4,
  notFn: unaryBitwiseNot,
  view: Uint32Array,
  mulFn: binaryImul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul",
        "shiftLeftByScalar", "shiftRightByScalar",
        "load", "load1", "load2", "load3", "store", "store1", "store2", "store3"],
}

var uint16x8 = {
  name: "Uint16x8",
  fn: SIMD.Uint16x8,
  lanes: 8,
  laneSize: 2,
  unsigned: true,
  minVal: 0,
  maxVal: 0xFFFF,
  buffer: _ui16x8,
  notFn: unaryBitwiseNot,
  view: Uint16Array,
  mulFn: binaryMul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul",
        "shiftLeftByScalar", "shiftRightByScalar",
        "addSaturate", "subSaturate",
        "load", "store"],
}

var uint8x16 = {
  name: "Uint8x16",
  fn: SIMD.Uint8x16,
  lanes: 16,
  laneSize: 1,
  unsigned: true,
  minVal: 0,
  maxVal: 0xFF,
  buffer: _ui8x16,
  notFn: unaryBitwiseNot,
  view: Uint8Array,
  mulFn: binaryMul,
  fns: ["check", "splat", "replaceLane", "select",
        "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
        "and", "or", "xor", "not",
        "add", "sub", "mul",
        "shiftLeftByScalar", "shiftRightByScalar",
        "addSaturate", "subSaturate",
        "load", "store"],
}

var bool32x4 = {
  name: "Bool32x4",
  fn: SIMD.Bool32x4,
  lanes: 4,
  laneSize: 4,
  notFn: unaryLogicalNot,
  fns: ["check", "splat", "replaceLane",
        "allTrue", "anyTrue", "and", "or", "xor", "not"],
}

var bool16x8 = {
  name: "Bool16x8",
  fn: SIMD.Bool16x8,
  lanes: 8,
  laneSize: 2,
  notFn: unaryLogicalNot,
  fns: ["check", "splat", "replaceLane",
        "allTrue", "anyTrue", "and", "or", "xor", "not"],
}

var bool8x16 = {
  name: "Bool8x16",
  fn: SIMD.Bool8x16,
  lanes: 16,
  laneSize: 1,
  notFn: unaryLogicalNot,
  fns: ["check", "splat", "replaceLane",
        "allTrue", "anyTrue", "and", "or", "xor", "not"],
}

// Each SIMD type has a corresponding Boolean SIMD type, which is returned by
// relational ops.
float32x4.boolType = int32x4.boolType = uint32x4.boolType = bool32x4;
int16x8.boolType = uint16x8.boolType = bool16x8;
int8x16.boolType = uint8x16.boolType = bool8x16;

// SIMD from<type> types.
float32x4.from = [int32x4, uint32x4];
int32x4.from = [float32x4, uint32x4];
int16x8.from = [uint16x8];
int8x16.from = [uint8x16];
uint32x4.from = [float32x4, int32x4];
uint16x8.from = [int16x8];
uint8x16.from = [int8x16];

// SIMD from<type>Bits types.
float32x4.fromBits = [int32x4, int16x8, int8x16, uint32x4, uint16x8, uint8x16];
int32x4.fromBits = [float32x4, int16x8, int8x16, uint32x4, uint16x8, uint8x16];
int16x8.fromBits = [float32x4, int32x4, int8x16, uint32x4, uint16x8, uint8x16];
int8x16.fromBits = [float32x4, int32x4, int16x8, uint32x4, uint16x8, uint8x16];
uint32x4.fromBits = [float32x4, int32x4, int16x8, int8x16, uint16x8, uint8x16];
uint16x8.fromBits = [float32x4, int32x4, int16x8, int8x16, uint32x4, uint8x16];
uint8x16.fromBits = [float32x4, int32x4, int16x8, int8x16, uint32x4, uint16x8];

var simdTypes = [float32x4,
                 int32x4, int16x8, int8x16,
                 uint32x4, uint16x8, uint8x16,
                 bool32x4, bool16x8, bool8x16];

// XXX Emscripten: Enable SIMD phase 2 types for Float64x2 and Bool64x2 to enable targeting SSE2 support.
simdPhase2 = true;

// SIMD Phase2 types.

if (typeof simdPhase2 !== 'undefined') {
  // Float64x2
  if (typeof SIMD.Float64x2 === "undefined" ||
      typeof SIMD.Float64x2.extractLane === "undefined") {
    SIMD.Float64x2 = function(s0, s1) {
      if (!(this instanceof SIMD.Float64x2)) {
        return new SIMD.Float64x2(s0, s1);
      }
      this.s_ = convertArray(_f64x2, new Float64Array([s0, s1]));
    }

    SIMD.Float64x2.extractLane = function(v, i) {
      v = SIMD.Float64x2.check(v);
      simdCheckLaneIndex(i, 2);
      return v.s_[i];
    }
  }

  if (typeof SIMD.Float64x2.swizzle === "undefined") {
    SIMD.Float64x2.swizzle = function(a, s0, s1) {
      return simdSwizzle(float64x2, a, [s0, s1]);
    }
  }

  if (typeof SIMD.Float64x2.shuffle === "undefined") {
    SIMD.Float64x2.shuffle = function(a, b, s0, s1) {
      return simdShuffle(float64x2, a, b, [s0, s1]);
    }
  }

  // Bool64x2
  if (typeof SIMD.Bool64x2 === "undefined" ||
      typeof SIMD.Bool64x2.extractLane === "undefined") {
    SIMD.Bool64x2 = function(s0, s1) {
      if (!(this instanceof SIMD.Bool64x2)) {
        return new SIMD.Bool64x2(s0, s1);
      }
      this.s_ = [!!s0, !!s1];
    }

    SIMD.Bool64x2.extractLane = function(v, i) {
      v = SIMD.Bool64x2.check(v);
      simdCheckLaneIndex(i, 2);
      return v.s_[i];
    }
  }

  var float64x2 = {
    name: "Float64x2",
    fn: SIMD.Float64x2,
    lanes: 2,
    laneSize: 8,
    buffer: _f64x2,
    view: Float64Array,
    mulFn: binaryMul,
    fns: ["check", "splat", "replaceLane", "select",
          "equal", "notEqual", "lessThan", "lessThanOrEqual", "greaterThan", "greaterThanOrEqual",
          "add", "sub", "mul", "div", "neg", "abs", "min", "max", "minNum", "maxNum",
          "reciprocalApproximation", "reciprocalSqrtApproximation", "sqrt",
          "load", "store"],
  }

  // XXX Emscripten: Need these functions for intrinsics, see https://github.com/tc39/ecmascript_simd/issues/316.
  float64x2.fns.push("load1");
  float64x2.fns.push("store1");
  // XXX Emscripten

  var bool64x2 = {
    name: "Bool64x2",
    fn: SIMD.Bool64x2,
    lanes: 2,
    laneSize: 8,
    notFn: unaryLogicalNot,
    fns: ["check", "splat", "replaceLane",
          "allTrue", "anyTrue", "and", "or", "xor", "not"],
  }

  float64x2.boolType = bool64x2;

  float32x4.fromBits.push(float64x2);
  int32x4.fromBits.push(float64x2);
  int16x8.fromBits.push(float64x2);
  int8x16.fromBits.push(float64x2);
  uint32x4.fromBits.push(float64x2);
  uint16x8.fromBits.push(float64x2);
  uint8x16.fromBits.push(float64x2);

  float64x2.fromBits = [float32x4, int32x4, int16x8, int8x16,
                        uint32x4, uint16x8, uint8x16];

/*
  // XXX Emscripten: Removed to fix https://github.com/tc39/ecmascript_simd/issues/314
  int32x4.fromBits = [float32x4, int16x8, int8x16, uint32x4, uint16x8, uint8x16];
  int16x8.fromBits = [float32x4, int32x4, int8x16, uint32x4, uint16x8, uint8x16];
  int8x16.fromBits = [float32x4, int32x4, int16x8, uint32x4, uint16x8, uint8x16];
  uint32x4.fromBits = [float32x4, int32x4, int16x8, int8x16, uint16x8, uint8x16];
  uint16x8.fromBits = [float32x4, int32x4, int16x8, int8x16, uint32x4, uint8x16];
  uint8x16.fromBits = [float32x4, int32x4, int16x8, int8x16, uint32x4, uint16x8];
*/

  simdTypes.push(float64x2);
  simdTypes.push(bool64x2);
}

// SIMD prototype functions.
var prototypeFns = {
  valueOf:
    function(type) {
      return function() {
        throw new TypeError(type.name + " cannot be converted to a number");
      }
    },

  toString:
    function(type) {
      return function() {
        return simdToString(type, this);
      }
    },

  toLocaleString:
    function(type) {
      return function() {
        return simdToLocaleString(type, this);
      }
    },
};

// SIMD constructor functions.

var simdFns = {
  check:
    function(type) {
      return function(a) {
        if (!(a instanceof type.fn)) {
          throw new TypeError("Argument is not a " + type.name + ".");
        }
        return a;
      }
    },

  splat:
    function(type) {
      return function(s) { return simdSplat(type, s); }
    },

  replaceLane:
    function(type) {
      return function(a, i, s) { return simdReplaceLane(type, a, i, s); }
    },

  allTrue:
    function(type) {
      return function(a) { return simdAllTrue(type, a); }
    },

  anyTrue:
    function(type) {
      return function(a) { return simdAnyTrue(type, a); }
    },

  and:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binaryAnd, a, b);
      }
    },

  or:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binaryOr, a, b);
      }
    },

  xor:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binaryXor, a, b);
      }
    },

  not:
    function(type) {
      return function(a) {
        return simdUnaryOp(type, type.notFn, a);
      }
    },

  equal:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryEqual, a, b);
      }
    },

  notEqual:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryNotEqual, a, b);
      }
    },

  lessThan:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryLess, a, b);
      }
    },

  lessThanOrEqual:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryLessEqual, a, b);
      }
    },

  greaterThan:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryGreater, a, b);
      }
    },

  greaterThanOrEqual:
    function(type) {
      return function(a, b) {
        return simdRelationalOp(type, binaryGreaterEqual, a, b);
      }
    },

  add:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binaryAdd, a, b);
      }
    },

  sub:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binarySub, a, b);
      }
    },

  mul:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, type.mulFn, a, b);
      }
    },

  div:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, binaryDiv, a, b);
      }
    },

  neg:
    function(type) {
      return function(a) {
        return simdUnaryOp(type, unaryNeg, a);
      }
    },

  abs:
    function(type) {
      return function(a) {
        return simdUnaryOp(type, Math.abs, a);
      }
    },

  min:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, Math.min, a, b);
      }
    },

  max:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, Math.max, a, b);
      }
    },

  minNum:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, minNum, a, b);
      }
    },

  maxNum:
    function(type) {
      return function(a, b) {
        return simdBinaryOp(type, maxNum, a, b);
      }
    },

  load:
    function(type) {
      return function(tarray, index) {
        return simdLoad(type, tarray, index, type.lanes);
      }
    },

  load1:
    function(type) {
      return function(tarray, index) {
        return simdLoad(type, tarray, index, 1);
      }
    },

  load2:
    function(type) {
      return function(tarray, index) {
        return simdLoad(type, tarray, index, 2);
      }
    },

  load3:
    function(type) {
      return function(tarray, index) {
        return simdLoad(type, tarray, index, 3);
      }
    },

  store:
    function(type) {
      return function(tarray, index, a) {
        return simdStore(type, tarray, index, a, type.lanes);
      }
    },

  store1:
    function(type) {
      return function(tarray, index, a) {
        return simdStore(type, tarray, index, a, 1);
      }
    },

  store2:
    function(type) {
      return function(tarray, index, a) {
        return simdStore(type, tarray, index, a, 2);
      }
    },

  store3:
    function(type) {
      return function(tarray, index, a) {
        return simdStore(type, tarray, index, a, 3);
      }
    },

  select:
    function(type) {
      return function(selector, a, b) {
        return simdSelect(type, selector, a, b);
      }
    },


  reciprocalApproximation:
    function(type) {
      return function(a) {
        a = type.fn.check(a);
        return type.fn.div(type.fn.splat(1.0), a);
      }
    },

  reciprocalSqrtApproximation:
    function(type) {
      return function(a) {
        a = type.fn.check(a);
        return type.fn.reciprocalApproximation(type.fn.sqrt(a));
      }
    },

  sqrt:
    function(type) {
      return function(a) {
        return simdUnaryOp(type, Math.sqrt, a);
      }
    },

  shiftLeftByScalar:
    function(type) {
      return function(a, bits) {
        if (bits>>>0 >= type.laneSize * 8)
          return type.fn.splat(0);
        return simdShiftOp(type, binaryShiftLeft, a, bits);
      }
    },

  shiftRightByScalar:
    function(type) {
      if (type.unsigned) {
        return function(a, bits) {
          if (bits>>>0 >= type.laneSize * 8)
            return type.fn.splat(0);
          return simdShiftOp(type, binaryShiftRightLogical, a, bits);
        }
      } else {
        return function(a, bits) {
          if (bits>>>0 >= type.laneSize * 8)
            bits = type.laneSize * 8 - 1;
          return simdShiftOp(type, binaryShiftRightArithmetic, a, bits);
        }
      }
    },

  addSaturate:
    function(type) {
      function addSaturate(a, b) {
        return clamp(a + b, type.minVal, type.maxVal);
      }
      return function(a, b) { return simdBinaryOp(type, addSaturate, a, b); }
    },

  subSaturate:
    function(type) {
      function subSaturate(a, b) {
        return clamp(a - b, type.minVal, type.maxVal);
      }
      return function(a, b) { return simdBinaryOp(type, subSaturate, a, b); }
    },
}

// Install functions.

simdTypes.forEach(function(type) {
  // Install each prototype function on each SIMD prototype.
  var simdFn = type.fn;
  var proto = simdFn.prototype;
  for (var name in prototypeFns) {
    if (!proto.hasOwnProperty(name))
      proto[name] = prototypeFns[name](type);
  }
  // Install regular functions.
  type.fns.forEach(function(name) {
    if (typeof simdFn[name] === "undefined")
      simdFn[name] = simdFns[name](type);
  });
  // Install 'fromTIMD' functions.
  if (type.from) {
    type.from.forEach(function(fromType) {
      var name = "from" + fromType.name;
      var toType = type;  // pull type into closure.
      if (typeof type.fn[name] === "undefined") {
        type.fn[name] =
            function(a) { return simdFrom(toType, fromType, a); }
      }
    });
  }
  // Install 'fromTIMDBits' functions.
  if (type.fromBits) {
    type.fromBits.forEach(function(fromType) {
      var name = "from" + fromType.name + "Bits";
      var toType = type;  // pull type into closure.
      if (typeof type.fn[name] === "undefined") {
        type.fn[name] =
            function(a) { return simdFromBits(toType, fromType, a); }
      }
    });
  }
});

// If we're in a browser, the global namespace is named 'window'. If we're
// in node, it's named 'global'. If we're in a web worker, it's named
// 'self'. If we're in a shell, 'this' might work.
})(typeof window !== "undefined"
   ? window
   : (typeof process === 'object' &&
      typeof require === 'function' &&
      typeof global === 'object')
     ? global
     : typeof self === 'object'
       ? self
       : this);


// XXX Emscripten-specific below XXX

// Work around Firefox Nightly bug that Float64x2 comparison return a Int32x4 instead of a Bool64x2.
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.equal(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevEqual = SIMD.Float64x2.equal;
    SIMD.Float64x2.equal = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevEqual(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    }
    console.error('Warning: Patching up SIMD.Float64x2.equal to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.notEqual(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevNotEqual = SIMD.Float64x2.notEqual;
    SIMD.Float64x2.notEqual = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevNotEqual(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    } 
    console.error('Warning: Patching up SIMD.Float64x2.notEqual to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.greaterThan(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevGreaterThan = SIMD.Float64x2.greaterThan;
    SIMD.Float64x2.greaterThan = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevGreaterThan(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    } 
    console.error('Warning: Patching up SIMD.Float64x2.greaterThan to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.greaterThanOrEqual(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevGreaterThanOrEqual = SIMD.Float64x2.greaterThanOrEqual;
    SIMD.Float64x2.greaterThanOrEqual = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevGreaterThanOrEqual(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    } 
    console.error('Warning: Patching up SIMD.Float64x2.greaterThanOrEqual to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.lessThan(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevLessThan = SIMD.Float64x2.lessThan;
    SIMD.Float64x2.lessThan = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevLessThan(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    } 
    console.error('Warning: Patching up SIMD.Float64x2.lessThan to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}
try {
  if (SIMD.Int32x4.check(SIMD.Float64x2.lessThanOrEqual(SIMD.Float64x2.splat(5.0), SIMD.Float64x2.splat(5.0)))) {
    SIMD.Float64x2.prevLessThanOrEqual = SIMD.Float64x2.lessThanOrEqual;
    SIMD.Float64x2.lessThanOrEqual = function(a, b) {
      var int32x4 = SIMD.Float64x2.prevLessThanOrEqual(a, b);
      return SIMD.Bool64x2(SIMD.Int32x4.extractLane(int32x4, 1) != 0, SIMD.Int32x4.extractLane(int32x4, 3) != 0);
    } 
    console.error('Warning: Patching up SIMD.Float64x2.lessThanOrEqual to return a Bool64x2 instead of Int32x4!');
  }
} catch(e) {}


if (!SIMD.Int32x4.fromBool64x2Bits) {
  SIMD.Int32x4.fromBool64x2Bits = function(bool64x2) {
    var lane0 = SIMD.Bool64x2.extractLane(bool64x2, 0)?-1:0;
    var lane1 = SIMD.Bool64x2.extractLane(bool64x2, 1)?-1:0;
    return SIMD.Int32x4(lane0, lane0, lane1, lane1);
  }
}