From f39fac9e4d58c98b24c31258e100872a084429f6 Mon Sep 17 00:00:00 2001
From: Con Kolivas <kernel@kolivas.org>
Date: Tue, 14 Feb 2012 17:32:40 +1100
Subject: [PATCH] Third pass reorder.

---
 poclbm120213.cl | 142 ++++++++++++++++++++++++------------------------
 1 file changed, 71 insertions(+), 71 deletions(-)

diff --git a/poclbm120213.cl b/poclbm120213.cl
index 3503675a..f2455c15 100644
--- a/poclbm120213.cl
+++ b/poclbm120213.cl
@@ -94,8 +94,8 @@ W[19]=d1;
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],b1,c1);
 W[19]+=K[4];
-W[19]+=0x80000000;
 W[23]=h1;
+W[19]+=0x80000000;
 W[23]+=W[19];
 W[20]+=fcty_e2;
 W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
@@ -117,8 +117,8 @@ W[18]+=Ma2(f1,W[19],W[20]);
 W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[17]+=Ma(W[20],W[18],W[19]);
 W[16]+=K[7];
+W[17]+=Ma(W[20],W[18],W[19]);
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
 W[16]+=Ma(W[19],W[17],W[18]);
@@ -214,8 +214,8 @@ W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
 W[19]+=Ma(W[22],W[20],W[21]);
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[5]=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=K[21];
+W[5]=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=W[5];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -295,8 +295,8 @@ W[14]=0x00a00055U;
 W[14]+=W[7];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[30];
 W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
+W[17]+=K[30];
 W[17]+=W[14];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -306,8 +306,8 @@ W[15]=fw15;
 W[15]+=W[8];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=K[31];
+W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=W[15];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -315,8 +315,8 @@ W[0]=fw01r;
 W[0]+=W[9];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[32];
 W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
+W[23]+=K[32];
 W[23]+=W[0];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -336,8 +336,8 @@ W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
 W[2]+=W[11];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[34];
 W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
+W[21]+=K[34];
 W[21]+=W[2];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -347,8 +347,8 @@ W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
 W[3]+=W[12];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=K[35];
+W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=W[3];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -356,8 +356,8 @@ W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
 W[4]+=W[13];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[36];
 W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
+W[19]+=K[36];
 W[19]+=W[4];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -367,8 +367,8 @@ W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
 W[5]+=W[14];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=K[37];
+W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=W[5];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -376,8 +376,8 @@ W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
 W[6]+=W[15];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[38];
 W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
+W[17]+=K[38];
 W[17]+=W[6];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -387,8 +387,8 @@ W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
 W[7]+=W[0];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=K[39];
+W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=W[7];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -396,8 +396,8 @@ W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
 W[8]+=W[1];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[40];
 W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
+W[23]+=K[40];
 W[23]+=W[8];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -407,8 +407,8 @@ W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
 W[9]+=W[2];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=K[41];
+W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=W[9];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -416,8 +416,8 @@ W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
 W[10]+=W[3];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[42];
 W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
+W[21]+=K[42];
 W[21]+=W[10];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -427,8 +427,8 @@ W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
 W[11]+=W[4];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=K[43];
+W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=W[11];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -436,8 +436,8 @@ W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
 W[12]+=W[5];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[44];
 W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
+W[19]+=K[44];
 W[19]+=W[12];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -447,8 +447,8 @@ W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
 W[13]+=W[6];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=K[45];
+W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=W[13];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -456,8 +456,8 @@ W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
 W[14]+=W[7];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[46];
 W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
+W[17]+=K[46];
 W[17]+=W[14];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -467,8 +467,8 @@ W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
 W[15]+=W[8];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=K[47];
+W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=W[15];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -476,8 +476,8 @@ W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
 W[0]+=W[9];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[48];
 W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
+W[23]+=K[48];
 W[23]+=W[0];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -487,8 +487,8 @@ W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
 W[1]+=W[10];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=K[49];
+W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=W[1];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -496,8 +496,8 @@ W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
 W[2]+=W[11];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[50];
 W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
+W[21]+=K[50];
 W[21]+=W[2];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -507,8 +507,8 @@ W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
 W[3]+=W[12];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=K[51];
+W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=W[3];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -516,8 +516,8 @@ W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
 W[4]+=W[13];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[52];
 W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
+W[19]+=K[52];
 W[19]+=W[4];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -527,8 +527,8 @@ W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
 W[5]+=W[14];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=K[53];
+W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=W[5];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -536,8 +536,8 @@ W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
 W[6]+=W[15];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[54];
 W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
+W[17]+=K[54];
 W[17]+=W[6];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -547,8 +547,8 @@ W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
 W[7]+=W[0];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=K[55];
+W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=W[7];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -556,8 +556,8 @@ W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
 W[8]+=W[1];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[56];
 W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
+W[23]+=K[56];
 W[23]+=W[8];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -567,8 +567,8 @@ W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
 W[9]+=W[2];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=K[57];
+W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=W[9];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -576,8 +576,8 @@ W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
 W[10]+=W[3];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[58];
 W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
+W[21]+=K[58];
 W[21]+=W[10];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -587,8 +587,8 @@ W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
 W[11]+=W[4];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=K[59];
+W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=W[11];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -596,8 +596,8 @@ W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
 W[12]+=W[5];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[60];
 W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
+W[19]+=K[60];
 W[19]+=W[12];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -607,8 +607,8 @@ W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
 W[13]+=W[6];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=K[61];
+W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=W[13];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -616,8 +616,8 @@ W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
 W[14]+=W[7];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[62];
 W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
+W[17]+=K[62];
 W[17]+=W[14];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -627,8 +627,8 @@ W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
 W[15]+=W[8];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=K[63];
+W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=W[15];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -654,8 +654,8 @@ W[22]+=(0x9b05688cU^(W[19]&0xca0b3af3U));
 W[22]+=K[1];
 W[2]=W[18];
 W[2]+=state2;
-W[18]=0x3c6ef372U;
 W[22]+=W[1];
+W[18]=0x3c6ef372U;
 W[18]+=W[22];
 W[23]+=0x08909ae5U;
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -665,8 +665,8 @@ W[21]=0x9b05688cU;
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],0x510e527fU);
 W[21]+=K[2];
-W[17]=0xbb67ae85U;
 W[21]+=W[2];
+W[17]=0xbb67ae85U;
 W[17]+=W[21];
 W[22]+=Ma2(0xbb67ae85U,W[23],0x6a09e667U);
 W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
@@ -676,8 +676,8 @@ W[20]=0x510e527fU;
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
 W[20]+=K[3];
-W[16]=0x6a09e667U;
 W[20]+=W[3];
+W[16]=0x6a09e667U;
 W[16]+=W[20];
 W[21]+=Ma2(0x6a09e667U,W[22],W[23]);
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -761,8 +761,8 @@ W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[16];
 W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
+W[23]+=K[16];
 W[23]+=W[0];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -816,8 +816,8 @@ W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
 W[6]+=0x00000100U;
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[22];
 W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
+W[17]+=K[22];
 W[17]+=W[6];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -827,8 +827,8 @@ W[7]+=0x11002000U;
 W[7]+=W[0];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=K[23];
+W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=W[7];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -836,8 +836,8 @@ W[8]=0x80000000;
 W[8]+=W[1];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[24];
 W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
+W[23]+=K[24];
 W[23]+=W[8];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -891,8 +891,8 @@ W[14]=0x00400022U;
 W[14]+=W[7];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[30];
 W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
+W[17]+=K[30];
 W[17]+=W[14];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -912,8 +912,8 @@ W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
 W[0]+=W[9];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[32];
 W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
+W[23]+=K[32];
 W[23]+=W[0];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -923,8 +923,8 @@ W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
 W[1]+=W[10];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=K[33];
+W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=W[1];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -932,8 +932,8 @@ W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
 W[2]+=W[11];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[34];
 W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
+W[21]+=K[34];
 W[21]+=W[2];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -943,8 +943,8 @@ W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
 W[3]+=W[12];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=K[35];
+W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=W[3];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -952,8 +952,8 @@ W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
 W[4]+=W[13];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[36];
 W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
+W[19]+=K[36];
 W[19]+=W[4];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -963,8 +963,8 @@ W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
 W[5]+=W[14];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=K[37];
+W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=W[5];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -972,8 +972,8 @@ W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
 W[6]+=W[15];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[38];
 W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
+W[17]+=K[38];
 W[17]+=W[6];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -983,8 +983,8 @@ W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
 W[7]+=W[0];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=K[39];
+W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=W[7];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -992,8 +992,8 @@ W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
 W[8]+=W[1];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[40];
 W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
+W[23]+=K[40];
 W[23]+=W[8];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -1003,8 +1003,8 @@ W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
 W[9]+=W[2];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=K[41];
+W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=W[9];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -1012,8 +1012,8 @@ W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
 W[10]+=W[3];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[42];
 W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
+W[21]+=K[42];
 W[21]+=W[10];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -1023,8 +1023,8 @@ W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
 W[11]+=W[4];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=K[43];
+W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=W[11];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -1032,8 +1032,8 @@ W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
 W[12]+=W[5];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[44];
 W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
+W[19]+=K[44];
 W[19]+=W[12];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -1043,8 +1043,8 @@ W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
 W[13]+=W[6];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=K[45];
+W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
 W[18]+=W[13];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -1052,8 +1052,8 @@ W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
 W[14]+=W[7];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[46];
 W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
+W[17]+=K[46];
 W[17]+=W[14];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -1063,8 +1063,8 @@ W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
 W[15]+=W[8];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=K[47];
+W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
 W[16]+=W[15];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -1072,8 +1072,8 @@ W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
 W[0]+=W[9];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[48];
 W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
+W[23]+=K[48];
 W[23]+=W[0];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -1083,8 +1083,8 @@ W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
 W[1]+=W[10];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=K[49];
+W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
 W[22]+=W[1];
 W[18]+=W[22];
 W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
@@ -1092,8 +1092,8 @@ W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
 W[2]+=W[11];
 W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
 W[21]+=ch(W[18],W[19],W[20]);
-W[21]+=K[50];
 W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
+W[21]+=K[50];
 W[21]+=W[2];
 W[22]+=Ma(W[17],W[23],W[16]);
 W[17]+=W[21];
@@ -1103,8 +1103,8 @@ W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
 W[3]+=W[12];
 W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
 W[20]+=ch(W[17],W[18],W[19]);
-W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=K[51];
+W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
 W[20]+=W[3];
 W[16]+=W[20];
 W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
@@ -1112,8 +1112,8 @@ W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
 W[4]+=W[13];
 W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[19]+=ch(W[16],W[17],W[18]);
-W[19]+=K[52];
 W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
+W[19]+=K[52];
 W[19]+=W[4];
 W[20]+=Ma(W[23],W[21],W[22]);
 W[23]+=W[19];
@@ -1123,8 +1123,8 @@ W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
 W[5]+=W[14];
 W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
 W[18]+=ch(W[23],W[16],W[17]);
-W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=K[53];
+W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
 W[18]+=W[5];
 W[22]+=W[18];
 W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
@@ -1132,8 +1132,8 @@ W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
 W[6]+=W[15];
 W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
 W[17]+=ch(W[22],W[23],W[16]);
-W[17]+=K[54];
 W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
+W[17]+=K[54];
 W[17]+=W[6];
 W[18]+=Ma(W[21],W[19],W[20]);
 W[21]+=W[17];
@@ -1143,8 +1143,8 @@ W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
 W[7]+=W[0];
 W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
 W[16]+=ch(W[21],W[22],W[23]);
-W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=K[55];
+W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
 W[16]+=W[7];
 W[20]+=W[16];
 W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
@@ -1152,8 +1152,8 @@ W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
 W[8]+=W[1];
 W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
 W[23]+=ch(W[20],W[21],W[22]);
-W[23]+=K[56];
 W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
+W[23]+=K[56];
 W[23]+=W[8];
 W[16]+=Ma(W[19],W[17],W[18]);
 W[19]+=W[23];
@@ -1163,8 +1163,8 @@ W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
 W[9]+=W[2];
 W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
 W[22]+=ch(W[19],W[20],W[21]);
-W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=K[57];
+W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
 W[22]+=W[9];
 W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
 W[10]+=W[3];
@@ -1184,8 +1184,8 @@ W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
 W[20]+=W[11];
 W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
 W[12]+=W[5];
-W[23]+=W[19];
 W[16]+=W[20];
+W[23]+=W[19];
 W[23]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
 W[23]+=ch(W[16],W[17],W[18]);
 W[23]+=K[60];