From 00ae70e9dd5c30b2cdd1bea305244782adae4f5b Mon Sep 17 00:00:00 2001
From: Arjun Barrett <arjunbarrett@gmail.com>
Date: Thu, 11 Feb 2021 12:23:33 -0800
Subject: [PATCH] Optimize loops for V8 Smi

---
 adler32.flow.js  |  6 ++---
 adler32.js       |  6 ++---
 bits/40_adler.js |  6 ++---
 ctest/adler32.js |  6 ++---
 perf/bstr.js     | 18 +++++++++++++-
 perf/utf8.js     | 63 ++++++++++++++++++++++++++++++++++++++++++++++++
 6 files changed, 92 insertions(+), 13 deletions(-)

diff --git a/adler32.flow.js b/adler32.flow.js
index 7bc16c0..2cd9017 100644
--- a/adler32.flow.js
+++ b/adler32.flow.js
@@ -36,7 +36,7 @@ function adler32_bstr(bstr/*:string*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += bstr.charCodeAt(i)&0xFF;
 			b += a;
@@ -51,7 +51,7 @@ function adler32_buf(buf/*:ABuf*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = buf.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = (seed >>> 16) & 0xFFFF; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += buf[i]&0xFF;
 			b += a;
@@ -67,7 +67,7 @@ function adler32_str(str/*:string*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = str.length, M = 0, c = 0, d = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850);
+		M = Math.min(L-i, 2918);
 		while(M>0) {
 			c = str.charCodeAt(i++);
 			if(c < 0x80) { a += c; }
diff --git a/adler32.js b/adler32.js
index ad6bfb7..4a29ac3 100644
--- a/adler32.js
+++ b/adler32.js
@@ -28,7 +28,7 @@ function adler32_bstr(bstr, seed) {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += bstr.charCodeAt(i)&0xFF;
 			b += a;
@@ -43,7 +43,7 @@ function adler32_buf(buf, seed) {
 	var a = 1, b = 0, L = buf.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = (seed >>> 16) & 0xFFFF; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += buf[i]&0xFF;
 			b += a;
@@ -58,7 +58,7 @@ function adler32_str(str, seed) {
 	var a = 1, b = 0, L = str.length, M = 0, c = 0, d = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850);
+		M = Math.min(L-i, 2918);
 		while(M>0) {
 			c = str.charCodeAt(i++);
 			if(c < 0x80) { a += c; }
diff --git a/bits/40_adler.js b/bits/40_adler.js
index 6b101e1..c990207 100644
--- a/bits/40_adler.js
+++ b/bits/40_adler.js
@@ -4,7 +4,7 @@ function adler32_bstr(bstr/*:string*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += bstr.charCodeAt(i)&0xFF;
 			b += a;
@@ -19,7 +19,7 @@ function adler32_buf(buf/*:ABuf*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = buf.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = (seed >>> 16) & 0xFFFF; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += buf[i]&0xFF;
 			b += a;
@@ -35,7 +35,7 @@ function adler32_str(str/*:string*/, seed/*:?ADLER32Type*/)/*:ADLER32Type*/ {
 	var a = 1, b = 0, L = str.length, M = 0, c = 0, d = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850);
+		M = Math.min(L-i, 2918);
 		while(M>0) {
 			c = str.charCodeAt(i++);
 			if(c < 0x80) { a += c; }
diff --git a/ctest/adler32.js b/ctest/adler32.js
index ad6bfb7..4a29ac3 100644
--- a/ctest/adler32.js
+++ b/ctest/adler32.js
@@ -28,7 +28,7 @@ function adler32_bstr(bstr, seed) {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += bstr.charCodeAt(i)&0xFF;
 			b += a;
@@ -43,7 +43,7 @@ function adler32_buf(buf, seed) {
 	var a = 1, b = 0, L = buf.length, M = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = (seed >>> 16) & 0xFFFF; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += buf[i]&0xFF;
 			b += a;
@@ -58,7 +58,7 @@ function adler32_str(str, seed) {
 	var a = 1, b = 0, L = str.length, M = 0, c = 0, d = 0;
 	if(typeof seed === 'number') { a = seed & 0xFFFF; b = seed >>> 16; }
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850);
+		M = Math.min(L-i, 2918);
 		while(M>0) {
 			c = str.charCodeAt(i++);
 			if(c < 0x80) { a += c; }
diff --git a/perf/bstr.js b/perf/bstr.js
index e1319b6..acb6b25 100644
--- a/perf/bstr.js
+++ b/perf/bstr.js
@@ -15,7 +15,7 @@ function sheetjs1(bstr) {
 function sheetjs2(bstr) {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	for(var i = 0; i < L;) {
-		M = Math.min(L-i, 3850)+i;
+		M = Math.min(L-i, 2654)+i;
 		for(;i<M;i++) {
 			a += bstr.charCodeAt(i);
 			b += a;
@@ -27,6 +27,20 @@ function sheetjs2(bstr) {
 }
 
 function sheetjs3(bstr) {
+	var a = 1, b = 0, L = bstr.length, M = 0;
+	for(var i = 0; i < L;) {
+		M = Math.min(L-i, 3850)+i;
+		for(;i<M;i++) {
+			a += bstr.charCodeAt(i);
+			b += a;
+		}
+		a = (15*(a>>>16)+(a&65535))
+		b = (15*(b>>>16)+(b&65535))
+	}
+	return ((b%65521) << 16) | (a%65521);
+}
+
+function sheetjs4(bstr) {
 	var a = 1, b = 0, L = bstr.length, M = 0;
 	for(var i = 0; i < L;) {
 		M = Math.min(L-i, 5552);
@@ -58,11 +72,13 @@ for(var i = 6; i != 14; ++i) {
 	assert.equal(res, sheetjs1(foobar));
 	assert.equal(res, sheetjs2(foobar));
 	assert.equal(res, sheetjs3(foobar));
+	assert.equal(res, sheetjs4(foobar));
 
 	var suite = new BM('binary string (' + foobar.length + ')');
 	if(i<3) suite.add('sheetjs 1', function() { for(var j = 0; j != m; ++j) sheetjs1(foobar); });
 	suite.add('sheetjs 2', function() { for(var j = 0; j != m; ++j) sheetjs2(foobar); });
 	suite.add('sheetjs 3', function() { for(var j = 0; j != m; ++j) sheetjs3(foobar); });
+	suite.add('sheetjs 4', function() { for(var j = 0; j != m; ++j) sheetjs4(foobar); });
 	suite.add('last vers', function() { for(var j = 0; j != m; ++j) old(foobar); });
 	suite.add('current v', function() { for(var j = 0; j != m; ++j) cur(foobar); });
 	suite.run();
diff --git a/perf/utf8.js b/perf/utf8.js
index 5f08191..066d9f8 100644
--- a/perf/utf8.js
+++ b/perf/utf8.js
@@ -17,6 +17,64 @@ function sheetjs1(utf8) {
 }
 
 function sheetjs2(utf8) {
+	var a = 1, b = 0, L = utf8.length, M, c, d;
+	for(var i = 0; i < L;) {
+		M = Math.min(L-i, 2654);
+		while(M>0) {
+			c = utf8.charCodeAt(i++);
+			if(c < 0x80) {
+				a += c; b += a; --M;
+			} else if(c < 0x800) {
+				a += 192|((c>>6)&31);             b += a; --M;
+				a += 128|(c&63);                  b += a; --M;
+			} else if(c >= 0xD800 && c < 0xE000) {
+				c = (c&1023)+64; d = utf8.charCodeAt(i++) & 1023;
+				a += 240|((c>>8)&7);              b += a; --M;
+				a += 128|((c>>2)&63);             b += a; --M;
+				a += 128|((d>>6)&15)|((c&3)<<4);  b += a; --M;
+				a += 128|(d&63);                  b += a; --M;
+			} else {
+				a += 224|((c>>12)&15);            b += a; --M;
+				a += 128|((c>>6)&63);             b += a; --M;
+				a += 128|(c&63);                  b += a; --M;
+			}
+		}
+		a %= 65521;
+		b %= 65521;
+	}
+	return (b << 16) | a;
+}
+
+function sheetjs3(utf8) {
+	var a = 1, b = 0, L = utf8.length, M, c, d;
+	for(var i = 0; i < L;) {
+		M = Math.min(L-i, 2918);
+		while(M>0) {
+			c = utf8.charCodeAt(i++);
+			if(c < 0x80) {
+				a += c; b += a; --M;
+			} else if(c < 0x800) {
+				a += 192|((c>>6)&31);             b += a; --M;
+				a += 128|(c&63);                  b += a; --M;
+			} else if(c >= 0xD800 && c < 0xE000) {
+				c = (c&1023)+64; d = utf8.charCodeAt(i++) & 1023;
+				a += 240|((c>>8)&7);              b += a; --M;
+				a += 128|((c>>2)&63);             b += a; --M;
+				a += 128|((d>>6)&15)|((c&3)<<4);  b += a; --M;
+				a += 128|(d&63);                  b += a; --M;
+			} else {
+				a += 224|((c>>12)&15);            b += a; --M;
+				a += 128|((c>>6)&63);             b += a; --M;
+				a += 128|(c&63);                  b += a; --M;
+			}
+		}
+		a %= 65521;
+		b %= 65521;
+	}
+	return (b << 16) | a;
+}
+
+function sheetjs4(utf8) {
 	var a = 1, b = 0, L = utf8.length, M, c, d;
 	for(var i = 0; i < L;) {
 		M = Math.min(L-i, 3850);
@@ -53,11 +111,16 @@ var res = old(foobar);
 assert.equal(res, cur(foobar));
 assert.equal(res, sheetjs1(foobar));
 assert.equal(res, sheetjs2(foobar));
+assert.equal(res, sheetjs3(foobar));
+assert.equal(res, sheetjs4(foobar));
+
 
 var BM = require('./bm');
 var suite = new BM('unicode string');
 suite.add('sheetjs 1', function() { for(var j = 0; j != 1000; ++j) sheetjs1(foobar); });
 suite.add('sheetjs 2', function() { for(var j = 0; j != 1000; ++j) sheetjs2(foobar); });
+suite.add('sheetjs 3', function() { for(var j = 0; j != 1000; ++j) sheetjs3(foobar); });
+suite.add('sheetjs 4', function() { for(var j = 0; j != 1000; ++j) sheetjs4(foobar); });
 suite.add('last vers', function() { for(var j = 0; j != 1000; ++j) old(foobar); });
 suite.add('current v', function() { for(var j = 0; j != 1000; ++j) cur(foobar); });
 suite.run();