poclbm120213.cl 38 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302
  1. // -ck modified kernel taken from Phoenix taken from poclbm, with aspects of
  2. // phatk and others.
  3. // Modified version copyright 2011-2012 Con Kolivas
  4. // This file is taken and modified from the public-domain poclbm project, and
  5. // we have therefore decided to keep it public-domain in Phoenix.
  6. #ifdef VECTORS4
  7. typedef uint4 u;
  8. #elif defined VECTORS2
  9. typedef uint2 u;
  10. #else
  11. typedef uint u;
  12. #endif
  13. __constant uint K[64] = {
  14. 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
  15. 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
  16. 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
  17. 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
  18. 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
  19. 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
  20. 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
  21. 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
  22. };
  23. // This part is not from the stock poclbm kernel. It's part of an optimization
  24. // added in the Phoenix Miner.
  25. // Some AMD devices have a BFI_INT opcode, which behaves exactly like the
  26. // SHA-256 ch function, but provides it in exactly one instruction. If
  27. // detected, use it for ch. Otherwise, construct ch out of simpler logical
  28. // primitives.
  29. #ifdef BITALIGN
  30. #pragma OPENCL EXTENSION cl_amd_media_ops : enable
  31. #define rotr(x, y) amd_bitalign((u)x, (u)x, (u)y)
  32. #ifdef BFI_INT
  33. // Well, slight problem... It turns out BFI_INT isn't actually exposed to
  34. // OpenCL (or CAL IL for that matter) in any way. However, there is
  35. // a similar instruction, BYTE_ALIGN_INT, which is exposed to OpenCL via
  36. // amd_bytealign, takes the same inputs, and provides the same output.
  37. // We can use that as a placeholder for BFI_INT and have the application
  38. // patch it after compilation.
  39. // This is the BFI_INT function
  40. #define ch(x, y, z) amd_bytealign(x, y, z)
  41. // Ma can also be implemented in terms of BFI_INT...
  42. #define Ma(x, y, z) amd_bytealign( (z^x), (y), (x) )
  43. #else // BFI_INT
  44. // Later SDKs optimise this to BFI INT without patching and GCN
  45. // actually fails if manually patched with BFI_INT
  46. #define ch(x, y, z) bitselect((u)z, (u)y, (u)x)
  47. #define Ma(x, y, z) bitselect((u)x, (u)y, (u)z ^ (u)x)
  48. #endif
  49. #else // BITALIGN
  50. #define ch(x, y, z) (z ^ (x & (y ^ z)))
  51. #define Ma(x, y, z) ((x & z) | (y & (x | z)))
  52. #define rotr(x, y) rotate((u)x, (u)(32 - y))
  53. #endif
  54. // AMD's KernelAnalyzer throws errors compiling the kernel if we use
  55. // amd_bytealign on constants with vectors enabled, so we use this to avoid
  56. // problems. (this is used 4 times, and likely optimized out by the compiler.)
  57. #define Ma2(x, y, z) ((y & z) | (x & (y | z)))
  58. __kernel void search(const uint state0, const uint state1, const uint state2, const uint state3,
  59. const uint state4, const uint state5, const uint state6, const uint state7,
  60. const uint b1, const uint c1, const uint d1,
  61. const uint f1, const uint g1, const uint h1,
  62. const u base,
  63. const uint fw0, const uint fw1, const uint fw2, const uint fw3, const uint fw15, const uint fw01r, const uint fcty_e, const uint fcty_e2,
  64. __global uint * output)
  65. {
  66. u W[24];
  67. //u Vals[8]; Now put at W[16] to be in same array
  68. #ifdef VECTORS4
  69. const u nonce = base + (uint)(get_local_id(0)) * 4u + (uint)(get_group_id(0)) * (WORKSIZE * 4u);
  70. #elif defined VECTORS2
  71. const u nonce = base + (uint)(get_local_id(0)) * 2u + (uint)(get_group_id(0)) * (WORKSIZE * 2u);
  72. #else
  73. const u nonce = base + get_local_id(0) + get_group_id(0) * (WORKSIZE);
  74. #endif
  75. W[20]=fcty_e;
  76. W[20]+=nonce;
  77. W[16]=W[20];
  78. W[16]+=state0;
  79. W[19]=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  80. W[19]+=d1;
  81. W[19]+=ch(W[16],b1,c1);
  82. W[19]+=K[4];
  83. W[19]+=0x80000000;
  84. W[23]=W[19];
  85. W[23]+=h1;
  86. W[20]+=fcty_e2;
  87. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  88. W[18]=c1;
  89. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  90. W[18]+=ch(W[23],W[16],b1);
  91. W[18]+=K[5];
  92. W[22]=W[18];
  93. W[22]+=g1;
  94. W[19]+=Ma2(g1,W[20],f1);
  95. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  96. W[17]=b1;
  97. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  98. W[17]+=ch(W[22],W[23],W[16]);
  99. W[17]+=K[6];
  100. W[21]=W[17];
  101. W[21]+=f1;
  102. W[18]+=Ma2(f1,W[19],W[20]);
  103. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  104. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  105. W[16]+=ch(W[21],W[22],W[23]);
  106. W[16]+=K[7];
  107. W[17]+=Ma(W[20],W[18],W[19]);
  108. W[20]+=W[16];
  109. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  110. W[16]+=Ma(W[19],W[17],W[18]);
  111. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  112. W[23]+=ch(W[20],W[21],W[22]);
  113. W[23]+=K[8];
  114. W[19]+=W[23];
  115. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  116. W[23]+=Ma(W[18],W[16],W[17]);
  117. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  118. W[22]+=ch(W[19],W[20],W[21]);
  119. W[22]+=K[9];
  120. W[18]+=W[22];
  121. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  122. W[22]+=Ma(W[17],W[23],W[16]);
  123. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  124. W[21]+=ch(W[18],W[19],W[20]);
  125. W[21]+=K[10];
  126. W[17]+=W[21];
  127. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  128. W[21]+=Ma(W[16],W[22],W[23]);
  129. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  130. W[20]+=ch(W[17],W[18],W[19]);
  131. W[20]+=K[11];
  132. W[16]+=W[20];
  133. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  134. W[20]+=Ma(W[23],W[21],W[22]);
  135. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  136. W[19]+=ch(W[16],W[17],W[18]);
  137. W[19]+=K[12];
  138. W[23]+=W[19];
  139. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  140. W[19]+=Ma(W[22],W[20],W[21]);
  141. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  142. W[18]+=ch(W[23],W[16],W[17]);
  143. W[18]+=K[13];
  144. W[22]+=W[18];
  145. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  146. W[18]+=Ma(W[21],W[19],W[20]);
  147. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  148. W[17]+=ch(W[22],W[23],W[16]);
  149. W[17]+=K[14];
  150. W[21]+=W[17];
  151. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  152. W[17]+=Ma(W[20],W[18],W[19]);
  153. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  154. W[16]+=ch(W[21],W[22],W[23]);
  155. W[16]+=K[15];
  156. W[16]+=0x00000280U;
  157. W[20]+=W[16];
  158. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  159. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  160. W[23]+=ch(W[20],W[21],W[22]);
  161. W[23]+=K[16];
  162. W[23]+=fw0;
  163. W[16]+=Ma(W[19],W[17],W[18]);
  164. W[19]+=W[23];
  165. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  166. W[23]+=Ma(W[18],W[16],W[17]);
  167. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  168. W[22]+=ch(W[19],W[20],W[21]);
  169. W[22]+=K[17];
  170. W[22]+=fw1;
  171. W[18]+=W[22];
  172. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  173. W[2]=(rotr(nonce,7)^rotr(nonce,18)^(nonce>>3U));
  174. W[2]+=fw2;
  175. W[21]+=W[2];
  176. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  177. W[21]+=ch(W[18],W[19],W[20]);
  178. W[21]+=K[18];
  179. W[22]+=Ma(W[17],W[23],W[16]);
  180. W[17]+=W[21];
  181. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  182. W[21]+=Ma(W[16],W[22],W[23]);
  183. W[3]=nonce;
  184. W[3]+=fw3;
  185. W[20]+=W[3];
  186. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  187. W[20]+=ch(W[17],W[18],W[19]);
  188. W[20]+=K[19];
  189. W[16]+=W[20];
  190. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  191. W[4]=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  192. W[4]+=0x80000000;
  193. W[19]+=W[4];
  194. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  195. W[19]+=ch(W[16],W[17],W[18]);
  196. W[19]+=K[20];
  197. W[20]+=Ma(W[23],W[21],W[22]);
  198. W[23]+=W[19];
  199. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  200. W[19]+=Ma(W[22],W[20],W[21]);
  201. W[5]=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  202. W[18]+=W[5];
  203. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  204. W[18]+=ch(W[23],W[16],W[17]);
  205. W[18]+=K[21];
  206. W[22]+=W[18];
  207. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  208. W[6]=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  209. W[6]+=0x00000280U;
  210. W[17]+=W[6];
  211. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  212. W[17]+=ch(W[22],W[23],W[16]);
  213. W[17]+=K[22];
  214. W[18]+=Ma(W[21],W[19],W[20]);
  215. W[21]+=W[17];
  216. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  217. W[17]+=Ma(W[20],W[18],W[19]);
  218. W[7]=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  219. W[7]+=fw0;
  220. W[16]+=W[7];
  221. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  222. W[16]+=ch(W[21],W[22],W[23]);
  223. W[16]+=K[23];
  224. W[20]+=W[16];
  225. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  226. W[8]=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  227. W[8]+=fw1;
  228. W[23]+=W[8];
  229. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  230. W[23]+=ch(W[20],W[21],W[22]);
  231. W[23]+=K[24];
  232. W[16]+=Ma(W[19],W[17],W[18]);
  233. W[19]+=W[23];
  234. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  235. W[23]+=Ma(W[18],W[16],W[17]);
  236. W[9]=W[2];
  237. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  238. W[22]+=W[9];
  239. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  240. W[22]+=ch(W[19],W[20],W[21]);
  241. W[22]+=K[25];
  242. W[18]+=W[22];
  243. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  244. W[10]=W[3];
  245. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  246. W[21]+=W[10];
  247. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  248. W[21]+=ch(W[18],W[19],W[20]);
  249. W[21]+=K[26];
  250. W[22]+=Ma(W[17],W[23],W[16]);
  251. W[17]+=W[21];
  252. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  253. W[21]+=Ma(W[16],W[22],W[23]);
  254. W[11]=W[4];
  255. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  256. W[20]+=W[11];
  257. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  258. W[20]+=ch(W[17],W[18],W[19]);
  259. W[20]+=K[27];
  260. W[16]+=W[20];
  261. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  262. W[12]=W[5];
  263. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  264. W[19]+=W[12];
  265. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  266. W[19]+=ch(W[16],W[17],W[18]);
  267. W[19]+=K[28];
  268. W[20]+=Ma(W[23],W[21],W[22]);
  269. W[23]+=W[19];
  270. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  271. W[19]+=Ma(W[22],W[20],W[21]);
  272. W[13]=W[6];
  273. W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
  274. W[18]+=W[13];
  275. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  276. W[18]+=ch(W[23],W[16],W[17]);
  277. W[18]+=K[29];
  278. W[22]+=W[18];
  279. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  280. W[14]=0x00a00055U;
  281. W[14]+=W[7];
  282. W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
  283. W[17]+=W[14];
  284. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  285. W[17]+=ch(W[22],W[23],W[16]);
  286. W[17]+=K[30];
  287. W[18]+=Ma(W[21],W[19],W[20]);
  288. W[21]+=W[17];
  289. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  290. W[17]+=Ma(W[20],W[18],W[19]);
  291. W[15]=fw15;
  292. W[15]+=W[8];
  293. W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
  294. W[16]+=W[15];
  295. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  296. W[16]+=ch(W[21],W[22],W[23]);
  297. W[16]+=K[31];
  298. W[20]+=W[16];
  299. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  300. W[0]=fw01r;
  301. W[0]+=W[9];
  302. W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
  303. W[23]+=W[0];
  304. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  305. W[23]+=ch(W[20],W[21],W[22]);
  306. W[23]+=K[32];
  307. W[16]+=Ma(W[19],W[17],W[18]);
  308. W[19]+=W[23];
  309. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  310. W[23]+=Ma(W[18],W[16],W[17]);
  311. W[1]=fw1;
  312. W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
  313. W[1]+=W[10];
  314. W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
  315. W[22]+=W[1];
  316. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  317. W[22]+=ch(W[19],W[20],W[21]);
  318. W[22]+=K[33];
  319. W[18]+=W[22];
  320. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  321. W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
  322. W[2]+=W[11];
  323. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  324. W[21]+=ch(W[18],W[19],W[20]);
  325. W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
  326. W[21]+=K[34];
  327. W[21]+=W[2];
  328. W[22]+=Ma(W[17],W[23],W[16]);
  329. W[17]+=W[21];
  330. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  331. W[21]+=Ma(W[16],W[22],W[23]);
  332. W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
  333. W[3]+=W[12];
  334. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  335. W[20]+=ch(W[17],W[18],W[19]);
  336. W[20]+=K[35];
  337. W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
  338. W[20]+=W[3];
  339. W[16]+=W[20];
  340. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  341. W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
  342. W[4]+=W[13];
  343. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  344. W[19]+=ch(W[16],W[17],W[18]);
  345. W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  346. W[19]+=K[36];
  347. W[19]+=W[4];
  348. W[20]+=Ma(W[23],W[21],W[22]);
  349. W[23]+=W[19];
  350. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  351. W[19]+=Ma(W[22],W[20],W[21]);
  352. W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
  353. W[5]+=W[14];
  354. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  355. W[18]+=ch(W[23],W[16],W[17]);
  356. W[18]+=K[37];
  357. W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  358. W[18]+=W[5];
  359. W[22]+=W[18];
  360. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  361. W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
  362. W[6]+=W[15];
  363. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  364. W[17]+=ch(W[22],W[23],W[16]);
  365. W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  366. W[17]+=K[38];
  367. W[17]+=W[6];
  368. W[18]+=Ma(W[21],W[19],W[20]);
  369. W[21]+=W[17];
  370. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  371. W[17]+=Ma(W[20],W[18],W[19]);
  372. W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
  373. W[7]+=W[0];
  374. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  375. W[16]+=ch(W[21],W[22],W[23]);
  376. W[16]+=K[39];
  377. W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  378. W[16]+=W[7];
  379. W[20]+=W[16];
  380. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  381. W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
  382. W[8]+=W[1];
  383. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  384. W[23]+=ch(W[20],W[21],W[22]);
  385. W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  386. W[23]+=K[40];
  387. W[23]+=W[8];
  388. W[16]+=Ma(W[19],W[17],W[18]);
  389. W[19]+=W[23];
  390. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  391. W[23]+=Ma(W[18],W[16],W[17]);
  392. W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
  393. W[9]+=W[2];
  394. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  395. W[22]+=ch(W[19],W[20],W[21]);
  396. W[22]+=K[41];
  397. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  398. W[22]+=W[9];
  399. W[18]+=W[22];
  400. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  401. W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
  402. W[10]+=W[3];
  403. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  404. W[21]+=ch(W[18],W[19],W[20]);
  405. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  406. W[21]+=K[42];
  407. W[21]+=W[10];
  408. W[22]+=Ma(W[17],W[23],W[16]);
  409. W[17]+=W[21];
  410. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  411. W[21]+=Ma(W[16],W[22],W[23]);
  412. W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
  413. W[11]+=W[4];
  414. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  415. W[20]+=ch(W[17],W[18],W[19]);
  416. W[20]+=K[43];
  417. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  418. W[20]+=W[11];
  419. W[16]+=W[20];
  420. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  421. W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
  422. W[12]+=W[5];
  423. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  424. W[19]+=ch(W[16],W[17],W[18]);
  425. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  426. W[19]+=K[44];
  427. W[19]+=W[12];
  428. W[20]+=Ma(W[23],W[21],W[22]);
  429. W[23]+=W[19];
  430. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  431. W[19]+=Ma(W[22],W[20],W[21]);
  432. W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
  433. W[13]+=W[6];
  434. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  435. W[18]+=ch(W[23],W[16],W[17]);
  436. W[18]+=K[45];
  437. W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
  438. W[18]+=W[13];
  439. W[22]+=W[18];
  440. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  441. W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
  442. W[14]+=W[7];
  443. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  444. W[17]+=ch(W[22],W[23],W[16]);
  445. W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
  446. W[17]+=K[46];
  447. W[17]+=W[14];
  448. W[18]+=Ma(W[21],W[19],W[20]);
  449. W[21]+=W[17];
  450. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  451. W[17]+=Ma(W[20],W[18],W[19]);
  452. W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
  453. W[15]+=W[8];
  454. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  455. W[16]+=ch(W[21],W[22],W[23]);
  456. W[16]+=K[47];
  457. W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
  458. W[16]+=W[15];
  459. W[20]+=W[16];
  460. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  461. W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
  462. W[0]+=W[9];
  463. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  464. W[23]+=ch(W[20],W[21],W[22]);
  465. W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
  466. W[23]+=K[48];
  467. W[23]+=W[0];
  468. W[16]+=Ma(W[19],W[17],W[18]);
  469. W[19]+=W[23];
  470. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  471. W[23]+=Ma(W[18],W[16],W[17]);
  472. W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
  473. W[1]+=W[10];
  474. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  475. W[22]+=ch(W[19],W[20],W[21]);
  476. W[22]+=K[49];
  477. W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
  478. W[22]+=W[1];
  479. W[18]+=W[22];
  480. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  481. W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
  482. W[2]+=W[11];
  483. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  484. W[21]+=ch(W[18],W[19],W[20]);
  485. W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
  486. W[21]+=K[50];
  487. W[21]+=W[2];
  488. W[22]+=Ma(W[17],W[23],W[16]);
  489. W[17]+=W[21];
  490. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  491. W[21]+=Ma(W[16],W[22],W[23]);
  492. W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
  493. W[3]+=W[12];
  494. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  495. W[20]+=ch(W[17],W[18],W[19]);
  496. W[20]+=K[51];
  497. W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
  498. W[20]+=W[3];
  499. W[16]+=W[20];
  500. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  501. W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
  502. W[4]+=W[13];
  503. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  504. W[19]+=ch(W[16],W[17],W[18]);
  505. W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  506. W[19]+=K[52];
  507. W[19]+=W[4];
  508. W[20]+=Ma(W[23],W[21],W[22]);
  509. W[23]+=W[19];
  510. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  511. W[19]+=Ma(W[22],W[20],W[21]);
  512. W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
  513. W[5]+=W[14];
  514. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  515. W[18]+=ch(W[23],W[16],W[17]);
  516. W[18]+=K[53];
  517. W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  518. W[18]+=W[5];
  519. W[22]+=W[18];
  520. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  521. W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
  522. W[6]+=W[15];
  523. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  524. W[17]+=ch(W[22],W[23],W[16]);
  525. W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  526. W[17]+=K[54];
  527. W[17]+=W[6];
  528. W[18]+=Ma(W[21],W[19],W[20]);
  529. W[21]+=W[17];
  530. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  531. W[17]+=Ma(W[20],W[18],W[19]);
  532. W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
  533. W[7]+=W[0];
  534. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  535. W[16]+=ch(W[21],W[22],W[23]);
  536. W[16]+=K[55];
  537. W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  538. W[16]+=W[7];
  539. W[20]+=W[16];
  540. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  541. W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
  542. W[8]+=W[1];
  543. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  544. W[23]+=ch(W[20],W[21],W[22]);
  545. W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  546. W[23]+=K[56];
  547. W[23]+=W[8];
  548. W[16]+=Ma(W[19],W[17],W[18]);
  549. W[19]+=W[23];
  550. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  551. W[23]+=Ma(W[18],W[16],W[17]);
  552. W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
  553. W[9]+=W[2];
  554. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  555. W[22]+=ch(W[19],W[20],W[21]);
  556. W[22]+=K[57];
  557. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  558. W[22]+=W[9];
  559. W[18]+=W[22];
  560. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  561. W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
  562. W[10]+=W[3];
  563. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  564. W[21]+=ch(W[18],W[19],W[20]);
  565. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  566. W[21]+=K[58];
  567. W[21]+=W[10];
  568. W[22]+=Ma(W[17],W[23],W[16]);
  569. W[17]+=W[21];
  570. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  571. W[21]+=Ma(W[16],W[22],W[23]);
  572. W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
  573. W[11]+=W[4];
  574. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  575. W[20]+=ch(W[17],W[18],W[19]);
  576. W[20]+=K[59];
  577. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  578. W[20]+=W[11];
  579. W[16]+=W[20];
  580. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  581. W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
  582. W[12]+=W[5];
  583. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  584. W[19]+=ch(W[16],W[17],W[18]);
  585. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  586. W[19]+=K[60];
  587. W[19]+=W[12];
  588. W[20]+=Ma(W[23],W[21],W[22]);
  589. W[23]+=W[19];
  590. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  591. W[19]+=Ma(W[22],W[20],W[21]);
  592. W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
  593. W[13]+=W[6];
  594. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  595. W[18]+=ch(W[23],W[16],W[17]);
  596. W[18]+=K[61];
  597. W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
  598. W[18]+=W[13];
  599. W[22]+=W[18];
  600. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  601. W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
  602. W[14]+=W[7];
  603. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  604. W[17]+=ch(W[22],W[23],W[16]);
  605. W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
  606. W[17]+=K[62];
  607. W[17]+=W[14];
  608. W[18]+=Ma(W[21],W[19],W[20]);
  609. W[21]+=W[17];
  610. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  611. W[17]+=Ma(W[20],W[18],W[19]);
  612. W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
  613. W[15]+=W[8];
  614. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  615. W[16]+=ch(W[21],W[22],W[23]);
  616. W[16]+=K[63];
  617. W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
  618. W[16]+=W[15];
  619. W[20]+=W[16];
  620. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  621. W[16]+=Ma(W[19],W[17],W[18]);
  622. W[0]=W[16];
  623. W[7]=state7;
  624. W[7]+=W[23];
  625. W[23]=0xb0edbdd0;
  626. W[23]+=K[0];
  627. W[0]+=state0;
  628. W[23]+=W[0];
  629. W[3]=state3;
  630. W[3]+=W[19];
  631. W[19]=0xa54ff53a;
  632. W[19]+=W[23];
  633. W[1]=W[17];
  634. W[1]+=state1;
  635. W[6]=state6;
  636. W[6]+=W[22];
  637. W[22]=0x1f83d9abU;
  638. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  639. W[22]+=(0x9b05688cU^(W[19]&0xca0b3af3U));
  640. W[22]+=K[1];
  641. W[2]=state2;
  642. W[2]+=W[18];
  643. W[18]=0x3c6ef372U;
  644. W[22]+=W[1];
  645. W[18]+=W[22];
  646. W[23]+=0x08909ae5U;
  647. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  648. W[5]=state5;
  649. W[5]+=W[21];
  650. W[21]=0x9b05688cU;
  651. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  652. W[21]+=ch(W[18],W[19],0x510e527fU);
  653. W[21]+=K[2];
  654. W[21]+=W[2];
  655. W[17]=0xbb67ae85U;
  656. W[17]+=W[21];
  657. W[22]+=Ma2(0xbb67ae85U,W[23],0x6a09e667U);
  658. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  659. W[4]=state4;
  660. W[4]+=W[20];
  661. W[20]=0x510e527fU;
  662. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  663. W[20]+=ch(W[17],W[18],W[19]);
  664. W[20]+=K[3];
  665. W[20]+=W[3];
  666. W[16]=W[20];
  667. W[16]+=0x6a09e667U;
  668. W[21]+=Ma2(0x6a09e667U,W[22],W[23]);
  669. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  670. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  671. W[19]+=ch(W[16],W[17],W[18]);
  672. W[19]+=K[4];
  673. W[19]+=W[4];
  674. W[20]+=Ma(W[23],W[21],W[22]);
  675. W[23]+=W[19];
  676. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  677. W[19]+=Ma(W[22],W[20],W[21]);
  678. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  679. W[18]+=ch(W[23],W[16],W[17]);
  680. W[18]+=K[5];
  681. W[18]+=W[5];
  682. W[22]+=W[18];
  683. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  684. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  685. W[17]+=ch(W[22],W[23],W[16]);
  686. W[17]+=K[6];
  687. W[17]+=W[6];
  688. W[18]+=Ma(W[21],W[19],W[20]);
  689. W[21]+=W[17];
  690. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  691. W[17]+=Ma(W[20],W[18],W[19]);
  692. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  693. W[16]+=ch(W[21],W[22],W[23]);
  694. W[16]+=K[7];
  695. W[16]+=W[7];
  696. W[20]+=W[16];
  697. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  698. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  699. W[23]+=ch(W[20],W[21],W[22]);
  700. W[23]+=K[8];
  701. W[23]+=0x80000000;
  702. W[16]+=Ma(W[19],W[17],W[18]);
  703. W[19]+=W[23];
  704. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  705. W[23]+=Ma(W[18],W[16],W[17]);
  706. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  707. W[22]+=ch(W[19],W[20],W[21]);
  708. W[22]+=K[9];
  709. W[18]+=W[22];
  710. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  711. W[22]+=Ma(W[17],W[23],W[16]);
  712. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  713. W[21]+=ch(W[18],W[19],W[20]);
  714. W[21]+=K[10];
  715. W[17]+=W[21];
  716. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  717. W[21]+=Ma(W[16],W[22],W[23]);
  718. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  719. W[20]+=ch(W[17],W[18],W[19]);
  720. W[20]+=K[11];
  721. W[16]+=W[20];
  722. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  723. W[20]+=Ma(W[23],W[21],W[22]);
  724. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  725. W[19]+=ch(W[16],W[17],W[18]);
  726. W[19]+=K[12];
  727. W[23]+=W[19];
  728. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  729. W[19]+=Ma(W[22],W[20],W[21]);
  730. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  731. W[18]+=ch(W[23],W[16],W[17]);
  732. W[18]+=K[13];
  733. W[22]+=W[18];
  734. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  735. W[18]+=Ma(W[21],W[19],W[20]);
  736. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  737. W[17]+=ch(W[22],W[23],W[16]);
  738. W[17]+=K[14];
  739. W[21]+=W[17];
  740. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  741. W[17]+=Ma(W[20],W[18],W[19]);
  742. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  743. W[16]+=ch(W[21],W[22],W[23]);
  744. W[16]+=K[15];
  745. W[16]+=0x00000100U;
  746. W[20]+=W[16];
  747. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  748. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  749. W[23]+=ch(W[20],W[21],W[22]);
  750. W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
  751. W[23]+=K[16];
  752. W[23]+=W[0];
  753. W[16]+=Ma(W[19],W[17],W[18]);
  754. W[19]+=W[23];
  755. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  756. W[23]+=Ma(W[18],W[16],W[17]);
  757. W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
  758. W[1]+=0x00a00000U;
  759. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  760. W[22]+=ch(W[19],W[20],W[21]);
  761. W[22]+=K[17];
  762. W[22]+=W[1];
  763. W[18]+=W[22];
  764. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  765. W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
  766. W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
  767. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  768. W[21]+=ch(W[18],W[19],W[20]);
  769. W[21]+=K[18];
  770. W[21]+=W[2];
  771. W[22]+=Ma(W[17],W[23],W[16]);
  772. W[17]+=W[21];
  773. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  774. W[21]+=Ma(W[16],W[22],W[23]);
  775. W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
  776. W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
  777. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  778. W[20]+=ch(W[17],W[18],W[19]);
  779. W[20]+=K[19];
  780. W[20]+=W[3];
  781. W[16]+=W[20];
  782. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  783. W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
  784. W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  785. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  786. W[19]+=ch(W[16],W[17],W[18]);
  787. W[19]+=K[20];
  788. W[19]+=W[4];
  789. W[20]+=Ma(W[23],W[21],W[22]);
  790. W[23]+=W[19];
  791. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  792. W[19]+=Ma(W[22],W[20],W[21]);
  793. W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
  794. W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  795. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  796. W[18]+=ch(W[23],W[16],W[17]);
  797. W[18]+=K[21];
  798. W[18]+=W[5];
  799. W[22]+=W[18];
  800. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  801. W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
  802. W[6]+=0x00000100U;
  803. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  804. W[17]+=ch(W[22],W[23],W[16]);
  805. W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  806. W[17]+=K[22];
  807. W[17]+=W[6];
  808. W[18]+=Ma(W[21],W[19],W[20]);
  809. W[21]+=W[17];
  810. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  811. W[17]+=Ma(W[20],W[18],W[19]);
  812. W[7]+=0x11002000U;
  813. W[7]+=W[0];
  814. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  815. W[16]+=ch(W[21],W[22],W[23]);
  816. W[16]+=K[23];
  817. W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  818. W[16]+=W[7];
  819. W[20]+=W[16];
  820. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  821. W[8]=0x80000000;
  822. W[8]+=W[1];
  823. W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  824. W[23]+=W[8];
  825. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  826. W[23]+=ch(W[20],W[21],W[22]);
  827. W[23]+=K[24];
  828. W[16]+=Ma(W[19],W[17],W[18]);
  829. W[19]+=W[23];
  830. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  831. W[23]+=Ma(W[18],W[16],W[17]);
  832. W[9]=W[2];
  833. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  834. W[22]+=W[9];
  835. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  836. W[22]+=ch(W[19],W[20],W[21]);
  837. W[22]+=K[25];
  838. W[18]+=W[22];
  839. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  840. W[10]=W[3];
  841. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  842. W[21]+=W[10];
  843. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  844. W[21]+=ch(W[18],W[19],W[20]);
  845. W[21]+=K[26];
  846. W[22]+=Ma(W[17],W[23],W[16]);
  847. W[17]+=W[21];
  848. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  849. W[21]+=Ma(W[16],W[22],W[23]);
  850. W[11]=W[4];
  851. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  852. W[20]+=W[11];
  853. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  854. W[20]+=ch(W[17],W[18],W[19]);
  855. W[20]+=K[27];
  856. W[16]+=W[20];
  857. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  858. W[12]=W[5];
  859. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  860. W[19]+=W[12];
  861. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  862. W[19]+=ch(W[16],W[17],W[18]);
  863. W[19]+=K[28];
  864. W[20]+=Ma(W[23],W[21],W[22]);
  865. W[23]+=W[19];
  866. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  867. W[19]+=Ma(W[22],W[20],W[21]);
  868. W[13]=W[6];
  869. W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
  870. W[18]+=W[13];
  871. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  872. W[18]+=ch(W[23],W[16],W[17]);
  873. W[18]+=K[29];
  874. W[22]+=W[18];
  875. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  876. W[14]=0x00400022U;
  877. W[14]+=W[7];
  878. W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
  879. W[17]+=W[14];
  880. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  881. W[17]+=ch(W[22],W[23],W[16]);
  882. W[17]+=K[30];
  883. W[18]+=Ma(W[21],W[19],W[20]);
  884. W[21]+=W[17];
  885. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  886. W[17]+=Ma(W[20],W[18],W[19]);
  887. W[15]=0x00000100U;
  888. W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
  889. W[15]+=W[8];
  890. W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
  891. W[16]+=W[15];
  892. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  893. W[16]+=ch(W[21],W[22],W[23]);
  894. W[16]+=K[31];
  895. W[20]+=W[16];
  896. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  897. W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
  898. W[0]+=W[9];
  899. W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
  900. W[23]+=W[0];
  901. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  902. W[23]+=ch(W[20],W[21],W[22]);
  903. W[23]+=K[32];
  904. W[16]+=Ma(W[19],W[17],W[18]);
  905. W[19]+=W[23];
  906. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  907. W[23]+=Ma(W[18],W[16],W[17]);
  908. W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
  909. W[1]+=W[10];
  910. W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
  911. W[22]+=W[1];
  912. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  913. W[22]+=ch(W[19],W[20],W[21]);
  914. W[22]+=K[33];
  915. W[18]+=W[22];
  916. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  917. W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
  918. W[2]+=W[11];
  919. W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
  920. W[21]+=W[2];
  921. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  922. W[21]+=ch(W[18],W[19],W[20]);
  923. W[21]+=K[34];
  924. W[22]+=Ma(W[17],W[23],W[16]);
  925. W[17]+=W[21];
  926. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  927. W[21]+=Ma(W[16],W[22],W[23]);
  928. W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
  929. W[3]+=W[12];
  930. W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
  931. W[20]+=W[3];
  932. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  933. W[20]+=ch(W[17],W[18],W[19]);
  934. W[20]+=K[35];
  935. W[16]+=W[20];
  936. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  937. W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
  938. W[4]+=W[13];
  939. W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  940. W[19]+=W[4];
  941. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  942. W[19]+=ch(W[16],W[17],W[18]);
  943. W[19]+=K[36];
  944. W[20]+=Ma(W[23],W[21],W[22]);
  945. W[23]+=W[19];
  946. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  947. W[19]+=Ma(W[22],W[20],W[21]);
  948. W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
  949. W[5]+=W[14];
  950. W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  951. W[18]+=W[5];
  952. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  953. W[18]+=ch(W[23],W[16],W[17]);
  954. W[18]+=K[37];
  955. W[22]+=W[18];
  956. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  957. W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
  958. W[6]+=W[15];
  959. W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  960. W[17]+=W[6];
  961. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  962. W[17]+=ch(W[22],W[23],W[16]);
  963. W[17]+=K[38];
  964. W[18]+=Ma(W[21],W[19],W[20]);
  965. W[21]+=W[17];
  966. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  967. W[17]+=Ma(W[20],W[18],W[19]);
  968. W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
  969. W[7]+=W[0];
  970. W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  971. W[16]+=W[7];
  972. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  973. W[16]+=ch(W[21],W[22],W[23]);
  974. W[16]+=K[39];
  975. W[20]+=W[16];
  976. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  977. W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
  978. W[8]+=W[1];
  979. W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  980. W[23]+=W[8];
  981. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  982. W[23]+=ch(W[20],W[21],W[22]);
  983. W[23]+=K[40];
  984. W[16]+=Ma(W[19],W[17],W[18]);
  985. W[19]+=W[23];
  986. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  987. W[23]+=Ma(W[18],W[16],W[17]);
  988. W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
  989. W[9]+=W[2];
  990. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  991. W[22]+=W[9];
  992. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  993. W[22]+=ch(W[19],W[20],W[21]);
  994. W[22]+=K[41];
  995. W[18]+=W[22];
  996. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  997. W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
  998. W[10]+=W[3];
  999. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  1000. W[21]+=W[10];
  1001. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  1002. W[21]+=ch(W[18],W[19],W[20]);
  1003. W[21]+=K[42];
  1004. W[22]+=Ma(W[17],W[23],W[16]);
  1005. W[17]+=W[21];
  1006. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  1007. W[21]+=Ma(W[16],W[22],W[23]);
  1008. W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
  1009. W[11]+=W[4];
  1010. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  1011. W[20]+=W[11];
  1012. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  1013. W[20]+=ch(W[17],W[18],W[19]);
  1014. W[20]+=K[43];
  1015. W[16]+=W[20];
  1016. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  1017. W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
  1018. W[12]+=W[5];
  1019. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  1020. W[19]+=W[12];
  1021. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  1022. W[19]+=ch(W[16],W[17],W[18]);
  1023. W[19]+=K[44];
  1024. W[20]+=Ma(W[23],W[21],W[22]);
  1025. W[23]+=W[19];
  1026. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  1027. W[19]+=Ma(W[22],W[20],W[21]);
  1028. W[13]+=(rotr(W[14],7)^rotr(W[14],18)^(W[14]>>3U));
  1029. W[13]+=W[6];
  1030. W[13]+=(rotr(W[11],17)^rotr(W[11],19)^(W[11]>>10U));
  1031. W[18]+=W[13];
  1032. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  1033. W[18]+=ch(W[23],W[16],W[17]);
  1034. W[18]+=K[45];
  1035. W[22]+=W[18];
  1036. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  1037. W[14]+=(rotr(W[15],7)^rotr(W[15],18)^(W[15]>>3U));
  1038. W[14]+=W[7];
  1039. W[14]+=(rotr(W[12],17)^rotr(W[12],19)^(W[12]>>10U));
  1040. W[17]+=W[14];
  1041. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  1042. W[17]+=ch(W[22],W[23],W[16]);
  1043. W[17]+=K[46];
  1044. W[18]+=Ma(W[21],W[19],W[20]);
  1045. W[21]+=W[17];
  1046. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  1047. W[17]+=Ma(W[20],W[18],W[19]);
  1048. W[15]+=(rotr(W[0],7)^rotr(W[0],18)^(W[0]>>3U));
  1049. W[15]+=W[8];
  1050. W[15]+=(rotr(W[13],17)^rotr(W[13],19)^(W[13]>>10U));
  1051. W[16]+=W[15];
  1052. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  1053. W[16]+=ch(W[21],W[22],W[23]);
  1054. W[16]+=K[47];
  1055. W[20]+=W[16];
  1056. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  1057. W[0]+=(rotr(W[1],7)^rotr(W[1],18)^(W[1]>>3U));
  1058. W[0]+=W[9];
  1059. W[0]+=(rotr(W[14],17)^rotr(W[14],19)^(W[14]>>10U));
  1060. W[23]+=W[0];
  1061. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  1062. W[23]+=ch(W[20],W[21],W[22]);
  1063. W[23]+=K[48];
  1064. W[16]+=Ma(W[19],W[17],W[18]);
  1065. W[19]+=W[23];
  1066. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  1067. W[23]+=Ma(W[18],W[16],W[17]);
  1068. W[1]+=(rotr(W[2],7)^rotr(W[2],18)^(W[2]>>3U));
  1069. W[1]+=W[10];
  1070. W[1]+=(rotr(W[15],17)^rotr(W[15],19)^(W[15]>>10U));
  1071. W[22]+=W[1];
  1072. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  1073. W[22]+=ch(W[19],W[20],W[21]);
  1074. W[22]+=K[49];
  1075. W[18]+=W[22];
  1076. W[22]+=(rotr(W[23],2)^rotr(W[23],13)^rotr(W[23],22));
  1077. W[2]+=(rotr(W[3],7)^rotr(W[3],18)^(W[3]>>3U));
  1078. W[2]+=W[11];
  1079. W[2]+=(rotr(W[0],17)^rotr(W[0],19)^(W[0]>>10U));
  1080. W[21]+=W[2];
  1081. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  1082. W[21]+=ch(W[18],W[19],W[20]);
  1083. W[21]+=K[50];
  1084. W[22]+=Ma(W[17],W[23],W[16]);
  1085. W[17]+=W[21];
  1086. W[21]+=(rotr(W[22],2)^rotr(W[22],13)^rotr(W[22],22));
  1087. W[21]+=Ma(W[16],W[22],W[23]);
  1088. W[3]+=(rotr(W[4],7)^rotr(W[4],18)^(W[4]>>3U));
  1089. W[3]+=W[12];
  1090. W[3]+=(rotr(W[1],17)^rotr(W[1],19)^(W[1]>>10U));
  1091. W[20]+=W[3];
  1092. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  1093. W[20]+=ch(W[17],W[18],W[19]);
  1094. W[20]+=K[51];
  1095. W[16]+=W[20];
  1096. W[20]+=(rotr(W[21],2)^rotr(W[21],13)^rotr(W[21],22));
  1097. W[4]+=(rotr(W[5],7)^rotr(W[5],18)^(W[5]>>3U));
  1098. W[4]+=W[13];
  1099. W[4]+=(rotr(W[2],17)^rotr(W[2],19)^(W[2]>>10U));
  1100. W[19]+=W[4];
  1101. W[19]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  1102. W[19]+=ch(W[16],W[17],W[18]);
  1103. W[19]+=K[52];
  1104. W[20]+=Ma(W[23],W[21],W[22]);
  1105. W[23]+=W[19];
  1106. W[19]+=(rotr(W[20],2)^rotr(W[20],13)^rotr(W[20],22));
  1107. W[19]+=Ma(W[22],W[20],W[21]);
  1108. W[5]+=(rotr(W[6],7)^rotr(W[6],18)^(W[6]>>3U));
  1109. W[5]+=W[14];
  1110. W[5]+=(rotr(W[3],17)^rotr(W[3],19)^(W[3]>>10U));
  1111. W[18]+=W[5];
  1112. W[18]+=(rotr(W[23],6)^rotr(W[23],11)^rotr(W[23],25));
  1113. W[18]+=ch(W[23],W[16],W[17]);
  1114. W[18]+=K[53];
  1115. W[22]+=W[18];
  1116. W[18]+=(rotr(W[19],2)^rotr(W[19],13)^rotr(W[19],22));
  1117. W[6]+=(rotr(W[7],7)^rotr(W[7],18)^(W[7]>>3U));
  1118. W[6]+=W[15];
  1119. W[6]+=(rotr(W[4],17)^rotr(W[4],19)^(W[4]>>10U));
  1120. W[17]+=W[6];
  1121. W[17]+=(rotr(W[22],6)^rotr(W[22],11)^rotr(W[22],25));
  1122. W[17]+=ch(W[22],W[23],W[16]);
  1123. W[17]+=K[54];
  1124. W[18]+=Ma(W[21],W[19],W[20]);
  1125. W[21]+=W[17];
  1126. W[17]+=(rotr(W[18],2)^rotr(W[18],13)^rotr(W[18],22));
  1127. W[17]+=Ma(W[20],W[18],W[19]);
  1128. W[7]+=(rotr(W[8],7)^rotr(W[8],18)^(W[8]>>3U));
  1129. W[7]+=W[0];
  1130. W[7]+=(rotr(W[5],17)^rotr(W[5],19)^(W[5]>>10U));
  1131. W[16]+=W[7];
  1132. W[16]+=(rotr(W[21],6)^rotr(W[21],11)^rotr(W[21],25));
  1133. W[16]+=ch(W[21],W[22],W[23]);
  1134. W[16]+=K[55];
  1135. W[20]+=W[16];
  1136. W[16]+=(rotr(W[17],2)^rotr(W[17],13)^rotr(W[17],22));
  1137. W[8]+=(rotr(W[9],7)^rotr(W[9],18)^(W[9]>>3U));
  1138. W[8]+=W[1];
  1139. W[8]+=(rotr(W[6],17)^rotr(W[6],19)^(W[6]>>10U));
  1140. W[23]+=W[8];
  1141. W[23]+=(rotr(W[20],6)^rotr(W[20],11)^rotr(W[20],25));
  1142. W[23]+=ch(W[20],W[21],W[22]);
  1143. W[23]+=K[56];
  1144. W[16]+=Ma(W[19],W[17],W[18]);
  1145. W[19]+=W[23];
  1146. W[23]+=(rotr(W[16],2)^rotr(W[16],13)^rotr(W[16],22));
  1147. W[23]+=Ma(W[18],W[16],W[17]);
  1148. W[9]+=(rotr(W[10],7)^rotr(W[10],18)^(W[10]>>3U));
  1149. W[9]+=W[2];
  1150. W[9]+=(rotr(W[7],17)^rotr(W[7],19)^(W[7]>>10U));
  1151. W[22]+=W[9];
  1152. W[22]+=(rotr(W[19],6)^rotr(W[19],11)^rotr(W[19],25));
  1153. W[22]+=ch(W[19],W[20],W[21]);
  1154. W[22]+=K[57];
  1155. W[10]+=(rotr(W[11],7)^rotr(W[11],18)^(W[11]>>3U));
  1156. W[10]+=W[3];
  1157. W[10]+=(rotr(W[8],17)^rotr(W[8],19)^(W[8]>>10U));
  1158. W[21]+=W[10];
  1159. W[18]+=W[22];
  1160. W[21]+=(rotr(W[18],6)^rotr(W[18],11)^rotr(W[18],25));
  1161. W[21]+=ch(W[18],W[19],W[20]);
  1162. W[21]+=K[58];
  1163. W[11]+=(rotr(W[12],7)^rotr(W[12],18)^(W[12]>>3U));
  1164. W[11]+=W[4];
  1165. W[11]+=(rotr(W[9],17)^rotr(W[9],19)^(W[9]>>10U));
  1166. W[20]+=W[11];
  1167. W[17]+=W[21];
  1168. W[20]+=(rotr(W[17],6)^rotr(W[17],11)^rotr(W[17],25));
  1169. W[20]+=ch(W[17],W[18],W[19]);
  1170. W[20]+=K[59];
  1171. W[12]+=(rotr(W[13],7)^rotr(W[13],18)^(W[13]>>3U));
  1172. W[12]+=W[5];
  1173. W[12]+=(rotr(W[10],17)^rotr(W[10],19)^(W[10]>>10U));
  1174. W[23]+=W[12];
  1175. W[16]+=W[20];
  1176. W[23]+=W[19];
  1177. W[23]+=(rotr(W[16],6)^rotr(W[16],11)^rotr(W[16],25));
  1178. W[23]+=ch(W[16],W[17],W[18]);
  1179. //W[23]+=K[60]; diffed from 0xA41F32E7
  1180. #define FOUND (0x80)
  1181. #define NFLAG (0x7F)
  1182. #if defined(VECTORS4)
  1183. W[23] ^= 0x136032ED;
  1184. bool result = W[23].x & W[23].y & W[23].z & W[23].w;
  1185. if (!result) {
  1186. if (!W[23].x)
  1187. output[FOUND] = output[NFLAG & nonce.x] = nonce.x;
  1188. if (!W[23].y)
  1189. output[FOUND] = output[NFLAG & nonce.y] = nonce.y;
  1190. if (!W[23].z)
  1191. output[FOUND] = output[NFLAG & nonce.z] = nonce.z;
  1192. if (!W[23].w)
  1193. output[FOUND] = output[NFLAG & nonce.w] = nonce.w;
  1194. }
  1195. #elif defined(VECTORS2)
  1196. W[23] ^= 0x136032ED;
  1197. bool result = W[23].x & W[23].y;
  1198. if (!result) {
  1199. if (!W[23].x)
  1200. output[FOUND] = output[NFLAG & nonce.x] = nonce.x;
  1201. if (!W[23].y)
  1202. output[FOUND] = output[NFLAG & nonce.y] = nonce.y;
  1203. }
  1204. #else
  1205. if (W[23] == 0x136032ED)
  1206. output[FOUND] = output[NFLAG & nonce] = nonce;
  1207. #endif
  1208. }