Final utf8 decoder opt (2-3x speed boost)

README.md

@@ -36,7 +36,7 @@ node v0.10.26, 1mb sample:
    inflate-dankogai x 32.16 ops/sec ±0.13% (56 runs sampled)
    inflate-imaya x 30.35 ops/sec ±0.92% (53 runs sampled)
  ! inflate-pako x 69.89 ops/sec ±1.46% (71 runs sampled)
-   inflate-pako-string x 12.46 ops/sec ±5.65% (36 runs sampled)
+   inflate-pako-string x 19.22 ops/sec ±1.86% (49 runs sampled)
    inflate-pako-untyped x 17.19 ops/sec ±0.85% (32 runs sampled)
  * inflate-zlib x 70.03 ops/sec ±1.64% (81 runs sampled)
 
@@ -52,7 +52,7 @@ node v0.11.12, 1mb sample:
    inflate-dankogai x 42.19 ops/sec ±0.09% (56 runs sampled)
    inflate-imaya x 79.68 ops/sec ±1.07% (68 runs sampled)
  ! inflate-pako x 97.52 ops/sec ±0.83% (80 runs sampled)
-   inflate-pako-string x 19.07 ops/sec ±12.67% (32 runs sampled)
+   inflate-pako-string x 45.19 ops/sec ±1.69% (57 runs sampled)
    inflate-pako-untyped x 24.35 ops/sec ±2.59% (40 runs sampled)
  * inflate-zlib x 60.32 ops/sec ±1.36% (69 runs sampled)
 ```

benchmark/profile.js

@@ -5,8 +5,8 @@ var pako = require('../index.js');
 
 var data = new Uint8Array(fs.readFileSync(__dirname +'/samples/lorem_1mb.txt'));
 
-var deflated = pako.deflate(data, { level: 6 });
+var deflated = pako.deflate(data, { level: 6/*, to: 'string'*/ });
 
 for (var i=0; i<200; i++) {
-  pako.inflate(deflated);
+  pako.inflate(deflated, { to: 'string' });
 }

lib/utils/strings.js

@@ -12,7 +12,7 @@ try { String.fromCharCode.apply(null, [0]); } catch(__) { STR_APPLY_OK = false;
 
 // Table with utf8 lengths (calculated by first byte of sequence)
 // Note, that 5 & 6-byte values and some 4-byte values can not be represented in JS,
-// because max possible codepoiny is 0x10ffff
+// because max possible codepoint is 0x10ffff
 var _utf8len = new utils.Buf8(256);
 for (var i=0; i<256; i++) {
   _utf8len[i] = (i >= 252 ? 6 : i >= 248 ? 5 : i >= 240 ? 4 : i >= 224 ? 3 : i >= 192 ? 2 : 1);
@@ -21,12 +21,10 @@ _utf8len[254]=_utf8len[254]=1; // Invalid sequence start
 
 
 // convert string to array (typed, when possible)
-// src: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Base64_encoding_and_decoding
 exports.string2buf = function (str) {
   var buf, c, c2, m_pos, i, str_len = str.length, buf_len = 0;
 
-  /* mapping... */
-
+  // count binary size
   for (m_pos = 0; m_pos < str_len; m_pos++) {
     c = str.charCodeAt(m_pos);
     if ((c & 0xfc00) === 0xd800 && (m_pos+1 < str_len)) {
@@ -39,10 +37,10 @@ exports.string2buf = function (str) {
     buf_len += c < 0x80 ? 1 : c < 0x800 ? 2 : c < 0x10000 ? 3 : 4;
   }
 
+  // allocate buffer
   buf = new utils.Buf8(buf_len);
 
-  /* transcription... */
-
+  // convert
   for (i=0, m_pos = 0; i < buf_len; m_pos++) {
     c = str.charCodeAt(m_pos);
     if ((c & 0xfc00) === 0xd800 && (m_pos+1 < str_len)) {
@@ -103,59 +101,41 @@ exports.binstring2buf = function(str) {
 
 
 // convert array to string
-// src: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Base64_encoding_and_decoding
 exports.buf2string = function (buf, max) {
-  /*jshint nonstandard:true*/
-  // That's not as fast as via String.fromCharCode.apply
-  /*return decodeURIComponent(escape(exports.buf2binstring(
-    (buf.length === max) ?
-      buf
-    :
-      buf.subarray ? buf.subarray(0, max) : buf.slice(0, max)
-  )));*/
-
-  var str, i, out, part, char_len;
+  var str, i, out, c, c_len;
   var len = max || buf.length;
 
   // Reserve max possible length (2 words per char)
-  var utf16buf = new utils.Buf16(len*2);
+  // NB: by unknown reasons, Array is significantly faster for
+  //     String.fromCharCode.apply than Uint16Array.
+  var utf16buf = new Array(len*2);
 
-  for (out=0, i=0; i<len; i++) {
-    part = buf[i];
-    char_len = _utf8len[part];
+  for (out=0, i=0; i<len;) {
+    c = buf[i++];
+    // quick process ascii
+    if (c < 0x80) { utf16buf[out++] = c; continue; }
 
-    // edge case - broken sequence
-    if (i + char_len > len) {
-      utf16buf[out++] = 0xfffd;
-      break; // end of string reached, stop
+    c_len = _utf8len[c];
+    // skip 5 & 6 byte codes
+    if (c_len > 4) { utf16buf[out++] = 0xfffd; i += c_len-1; continue; }
+
+    // apply mask on first byte
+    c &= c_len === 2 ? 0x1f : c_len === 3 ? 0x0f : 0x07;
+    // join the rest
+    while (c_len > 1 && i < len) {
+      c = (c << 6) | (buf[i++] & 0x3f);
+      c_len--;
     }
-    switch (char_len) {
-      case 1:
-        utf16buf[out++] = part;
-        break;
-      case 2:
-        utf16buf[out++] = ((part & 0x1f) << 6) | (buf[++i] & 0x3f);
-        break;
-      case 3:
-        utf16buf[out++] = ((part & 0x0f) << 12) | ((buf[++i] & 0x3f) << 6) | (buf[++i] & 0x3f);
-        break;
-      case 4:
-        // surrogate pair
-        part = ((part & 0x07) << 18) | ((buf[++i] & 0x3f) << 12) | ((buf[++i] & 0x3f) << 6) + (buf[++i] & 0x3f);
-        part -= 0x10000;
-        utf16buf[out++] = 0xd800 | ((part >> 10) & 0x3ff);
-        utf16buf[out++] = 0xdc00 | (part & 0x3ff);
-        break;
-      // 5 & 6 bytes uticodes not supported in UTF16 (JS),
-      // so fill with dummy symbol & update scan position.
-      case 5:
-        i += 4;
-        utf16buf[out++] = 0xfffd;
-        break;
-      case 6:
-        i += 5;
-        utf16buf[out++] = 0xfffd;
-        break;
+
+    // terminated by end of string?
+    if (c_len > 1) { utf16buf[out++] = 0xfffd; continue; }
+
+    if (c < 0x10000) {
+      utf16buf[out++] = c;
+    } else {
+      c -= 0x10000;
+      utf16buf[out++] = 0xd800 | ((c >> 10) & 0x3ff);
+      utf16buf[out++] = 0xdc00 | (c & 0x3ff);
     }
   }