poclbm.cl 46 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683
  1. // This file is taken and modified from the public-domain poclbm project, and
  2. // we have therefore decided to keep it public-domain in Phoenix.
  3. // The X is a placeholder for patching to suit hardware
  4. #define VECTORSX
  5. #ifdef VECTORS4
  6. typedef uint4 u;
  7. #elif defined VECTORS2
  8. typedef uint2 u;
  9. #else
  10. typedef uint u;
  11. #endif
  12. __constant uint K[64] = {
  13. 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
  14. 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
  15. 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
  16. 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
  17. 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
  18. 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
  19. 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
  20. 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
  21. };
  22. // This part is not from the stock poclbm kernel. It's part of an optimization
  23. // added in the Phoenix Miner.
  24. // Some AMD devices have a BFI_INT opcode, which behaves exactly like the
  25. // SHA-256 ch function, but provides it in exactly one instruction. If
  26. // detected, use it for ch. Otherwise, construct ch out of simpler logical
  27. // primitives.
  28. #define BFI_INTX
  29. #ifdef BFI_INT
  30. #define BITALIGN
  31. // Well, slight problem... It turns out BFI_INT isn't actually exposed to
  32. // OpenCL (or CAL IL for that matter) in any way. However, there is
  33. // a similar instruction, BYTE_ALIGN_INT, which is exposed to OpenCL via
  34. // amd_bytealign, takes the same inputs, and provides the same output.
  35. // We can use that as a placeholder for BFI_INT and have the application
  36. // patch it after compilation.
  37. // This is the BFI_INT function
  38. #define ch(x, y, z) amd_bytealign(x, y, z)
  39. // Ma can also be implemented in terms of BFI_INT...
  40. #define Ma(x, y, z) amd_bytealign((y), (x | z), (z & x))
  41. #else
  42. #define ch(x, y, z) (z ^ (x & (y ^ z)))
  43. #define Ma(x, y, z) ((x & z) | (y & (x | z)))
  44. #endif
  45. #ifdef BITALIGN
  46. #pragma OPENCL EXTENSION cl_amd_media_ops : enable
  47. #define rotr(x, y) amd_bitalign((u)x, (u)x, (u)y)
  48. #else
  49. #define rotr(x, y) rotate((u)x, (u)(32-y))
  50. #endif
  51. // AMD's KernelAnalyzer throws errors compiling the kernel if we use
  52. // amd_bytealign on constants with vectors enabled, so we use this to avoid
  53. // problems. (this is used 4 times, and likely optimized out by the compiler.)
  54. #define Ma2(x, y, z) ((y & z) | (x & (y | z)))
  55. __kernel void search( const uint state0, const uint state1, const uint state2, const uint state3,
  56. const uint state4, const uint state5, const uint state6, const uint state7,
  57. const uint b1, const uint c1, const uint d1,
  58. const uint f1, const uint g1, const uint h1,
  59. const uint base,
  60. 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,
  61. __global uint * output)
  62. {
  63. u W[24];
  64. u Vals[8];
  65. u nonce;
  66. uint it;
  67. #ifdef VECTORS4
  68. nonce = ((base >> 2) + (get_global_id(0))<<2) + (uint4)(0, 1, 2, 3);
  69. #elif defined VECTORS2
  70. nonce = ((base >> 1) + (get_global_id(0))<<1) + (uint2)(0, 1);
  71. #else
  72. nonce = base + get_global_id(0);
  73. #endif
  74. W[3] = nonce + fw3;
  75. Vals[4] = fcty_e + nonce;
  76. Vals[0] = state0 + Vals[4];
  77. Vals[4] = Vals[4] + fcty_e2;
  78. Vals[3] = d1 + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], b1, c1) + K[ 4] + 0x80000000;
  79. Vals[7] = h1 + Vals[3];
  80. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma2(g1, Vals[4], f1);
  81. Vals[2] = c1 + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], b1) + K[ 5];
  82. Vals[6] = g1 + Vals[2];
  83. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma2(f1, Vals[3], Vals[4]);
  84. Vals[1] = b1 + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[ 6];
  85. Vals[5] = f1 + Vals[1];
  86. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  87. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[ 7];
  88. Vals[4] = Vals[4] + Vals[0];
  89. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  90. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[ 8];
  91. Vals[3] = Vals[3] + Vals[7];
  92. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  93. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[ 9];
  94. Vals[2] = Vals[2] + Vals[6];
  95. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  96. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[10];
  97. Vals[1] = Vals[1] + Vals[5];
  98. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  99. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[11];
  100. Vals[0] = Vals[0] + Vals[4];
  101. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  102. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[12];
  103. Vals[7] = Vals[7] + Vals[3];
  104. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  105. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[13];
  106. Vals[6] = Vals[6] + Vals[2];
  107. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  108. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[14];
  109. Vals[5] = Vals[5] + Vals[1];
  110. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  111. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[15] + 0x00000280U;
  112. Vals[4] = Vals[4] + Vals[0];
  113. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  114. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[16] + fw0;
  115. Vals[3] = Vals[3] + Vals[7];
  116. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  117. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[17] + fw1;
  118. Vals[2] = Vals[2] + Vals[6];
  119. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  120. W[2] = (rotr(nonce, 7) ^ rotr(nonce, 18) ^ (nonce >> 3U)) + fw2;
  121. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[18] + W[2];
  122. Vals[1] = Vals[1] + Vals[5];
  123. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  124. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[19] + W[3];
  125. Vals[0] = Vals[0] + Vals[4];
  126. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  127. W[4] = (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U)) + 0x80000000;
  128. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[20] + W[4];
  129. Vals[7] = Vals[7] + Vals[3];
  130. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  131. W[5] = (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  132. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[21] + W[5];
  133. Vals[6] = Vals[6] + Vals[2];
  134. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  135. W[6] = (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U)) + 0x00000280U;
  136. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[22] + W[6];
  137. Vals[5] = Vals[5] + Vals[1];
  138. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  139. W[7] = (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U)) + fw0;
  140. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[23] + W[7];
  141. Vals[4] = Vals[4] + Vals[0];
  142. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  143. W[8] = (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U)) + fw1;
  144. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[24] + W[8];
  145. Vals[3] = Vals[3] + Vals[7];
  146. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  147. W[9] = W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  148. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[25] + W[9];
  149. Vals[2] = Vals[2] + Vals[6];
  150. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  151. W[10] = W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  152. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[26] + W[10];
  153. Vals[1] = Vals[1] + Vals[5];
  154. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  155. W[11] = W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  156. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[27] + W[11];
  157. Vals[0] = Vals[0] + Vals[4];
  158. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  159. W[12] = W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  160. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[28] + W[12];
  161. Vals[7] = Vals[7] + Vals[3];
  162. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  163. W[13] = W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  164. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[29] + W[13];
  165. Vals[6] = Vals[6] + Vals[2];
  166. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  167. W[14] = 0x00a00055U + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  168. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[30] + W[14];
  169. Vals[5] = Vals[5] + Vals[1];
  170. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  171. W[15] = fw15 + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  172. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[31] + W[15];
  173. Vals[4] = Vals[4] + Vals[0];
  174. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  175. W[0] = fw01r + W[9] + (rotr(W[14], 17) ^ rotr(W[14], 19) ^ (W[14] >> 10U));
  176. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[32] + W[0];
  177. Vals[3] = Vals[3] + Vals[7];
  178. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  179. W[1] = fw1 + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + W[10] + (rotr(W[15], 17) ^ rotr(W[15], 19) ^ (W[15] >> 10U));
  180. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[33] + W[1];
  181. Vals[2] = Vals[2] + Vals[6];
  182. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  183. W[2] = W[2] + (rotr(W[3], 7) ^ rotr(W[3], 18) ^ (W[3] >> 3U)) + W[11] + (rotr(W[0], 17) ^ rotr(W[0], 19) ^ (W[0] >> 10U));
  184. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[34] + W[2];
  185. Vals[1] = Vals[1] + Vals[5];
  186. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  187. W[3] = W[3] + (rotr(W[4], 7) ^ rotr(W[4], 18) ^ (W[4] >> 3U)) + W[12] + (rotr(W[1], 17) ^ rotr(W[1], 19) ^ (W[1] >> 10U));
  188. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[35] + W[3];
  189. Vals[0] = Vals[0] + Vals[4];
  190. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  191. W[4] = W[4] + (rotr(W[5], 7) ^ rotr(W[5], 18) ^ (W[5] >> 3U)) + W[13] + (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U));
  192. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[36] + W[4];
  193. Vals[7] = Vals[7] + Vals[3];
  194. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  195. W[5] = W[5] + (rotr(W[6], 7) ^ rotr(W[6], 18) ^ (W[6] >> 3U)) + W[14] + (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  196. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[37] + W[5];
  197. Vals[6] = Vals[6] + Vals[2];
  198. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  199. W[6] = W[6] + (rotr(W[7], 7) ^ rotr(W[7], 18) ^ (W[7] >> 3U)) + W[15] + (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U));
  200. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[38] + W[6];
  201. Vals[5] = Vals[5] + Vals[1];
  202. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  203. W[7] = W[7] + (rotr(W[8], 7) ^ rotr(W[8], 18) ^ (W[8] >> 3U)) + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  204. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[39] + W[7];
  205. Vals[4] = Vals[4] + Vals[0];
  206. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  207. W[8] = W[8] + (rotr(W[9], 7) ^ rotr(W[9], 18) ^ (W[9] >> 3U)) + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  208. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[40] + W[8];
  209. Vals[3] = Vals[3] + Vals[7];
  210. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  211. W[9] = W[9] + (rotr(W[10], 7) ^ rotr(W[10], 18) ^ (W[10] >> 3U)) + W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  212. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[41] + W[9];
  213. Vals[2] = Vals[2] + Vals[6];
  214. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  215. W[10] = W[10] + (rotr(W[11], 7) ^ rotr(W[11], 18) ^ (W[11] >> 3U)) + W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  216. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[42] + W[10];
  217. Vals[1] = Vals[1] + Vals[5];
  218. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  219. W[11] = W[11] + (rotr(W[12], 7) ^ rotr(W[12], 18) ^ (W[12] >> 3U)) + W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  220. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[43] + W[11];
  221. Vals[0] = Vals[0] + Vals[4];
  222. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  223. W[12] = W[12] + (rotr(W[13], 7) ^ rotr(W[13], 18) ^ (W[13] >> 3U)) + W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  224. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[44] + W[12];
  225. Vals[7] = Vals[7] + Vals[3];
  226. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  227. W[13] = W[13] + (rotr(W[14], 7) ^ rotr(W[14], 18) ^ (W[14] >> 3U)) + W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  228. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[45] + W[13];
  229. Vals[6] = Vals[6] + Vals[2];
  230. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  231. W[14] = W[14] + (rotr(W[15], 7) ^ rotr(W[15], 18) ^ (W[15] >> 3U)) + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  232. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[46] + W[14];
  233. Vals[5] = Vals[5] + Vals[1];
  234. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  235. W[15] = W[15] + (rotr(W[0], 7) ^ rotr(W[0], 18) ^ (W[0] >> 3U)) + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  236. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[47] + W[15];
  237. Vals[4] = Vals[4] + Vals[0];
  238. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  239. W[0] = W[0] + (rotr(W[1], 7) ^ rotr(W[1], 18) ^ (W[1] >> 3U)) + W[9] + (rotr(W[14], 17) ^ rotr(W[14], 19) ^ (W[14] >> 10U));
  240. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[48] + W[0];
  241. Vals[3] = Vals[3] + Vals[7];
  242. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  243. W[1] = W[1] + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + W[10] + (rotr(W[15], 17) ^ rotr(W[15], 19) ^ (W[15] >> 10U));
  244. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[49] + W[1];
  245. Vals[2] = Vals[2] + Vals[6];
  246. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  247. W[2] = W[2] + (rotr(W[3], 7) ^ rotr(W[3], 18) ^ (W[3] >> 3U)) + W[11] + (rotr(W[0], 17) ^ rotr(W[0], 19) ^ (W[0] >> 10U));
  248. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[50] + W[2];
  249. Vals[1] = Vals[1] + Vals[5];
  250. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  251. W[3] = W[3] + (rotr(W[4], 7) ^ rotr(W[4], 18) ^ (W[4] >> 3U)) + W[12] + (rotr(W[1], 17) ^ rotr(W[1], 19) ^ (W[1] >> 10U));
  252. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[51] + W[3];
  253. Vals[0] = Vals[0] + Vals[4];
  254. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  255. W[4] = W[4] + (rotr(W[5], 7) ^ rotr(W[5], 18) ^ (W[5] >> 3U)) + W[13] + (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U));
  256. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[52] + W[4];
  257. Vals[7] = Vals[7] + Vals[3];
  258. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  259. W[5] = W[5] + (rotr(W[6], 7) ^ rotr(W[6], 18) ^ (W[6] >> 3U)) + W[14] + (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  260. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[53] + W[5];
  261. Vals[6] = Vals[6] + Vals[2];
  262. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  263. W[6] = W[6] + (rotr(W[7], 7) ^ rotr(W[7], 18) ^ (W[7] >> 3U)) + W[15] + (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U));
  264. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[54] + W[6];
  265. Vals[5] = Vals[5] + Vals[1];
  266. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  267. W[7] = W[7] + (rotr(W[8], 7) ^ rotr(W[8], 18) ^ (W[8] >> 3U)) + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  268. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[55] + W[7];
  269. Vals[4] = Vals[4] + Vals[0];
  270. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  271. W[8] = W[8] + (rotr(W[9], 7) ^ rotr(W[9], 18) ^ (W[9] >> 3U)) + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  272. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[56] + W[8];
  273. Vals[3] = Vals[3] + Vals[7];
  274. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  275. W[9] = W[9] + (rotr(W[10], 7) ^ rotr(W[10], 18) ^ (W[10] >> 3U)) + W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  276. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[57] + W[9];
  277. Vals[2] = Vals[2] + Vals[6];
  278. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  279. W[10] = W[10] + (rotr(W[11], 7) ^ rotr(W[11], 18) ^ (W[11] >> 3U)) + W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  280. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[58] + W[10];
  281. Vals[1] = Vals[1] + Vals[5];
  282. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  283. W[11] = W[11] + (rotr(W[12], 7) ^ rotr(W[12], 18) ^ (W[12] >> 3U)) + W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  284. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[59] + W[11];
  285. Vals[0] = Vals[0] + Vals[4];
  286. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  287. W[12] = W[12] + (rotr(W[13], 7) ^ rotr(W[13], 18) ^ (W[13] >> 3U)) + W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  288. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[60] + W[12];
  289. Vals[7] = Vals[7] + Vals[3];
  290. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  291. W[13] = W[13] + (rotr(W[14], 7) ^ rotr(W[14], 18) ^ (W[14] >> 3U)) + W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  292. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[61] + W[13];
  293. Vals[6] = Vals[6] + Vals[2];
  294. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  295. W[14] = W[14] + (rotr(W[15], 7) ^ rotr(W[15], 18) ^ (W[15] >> 3U)) + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  296. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[62] + W[14];
  297. Vals[5] = Vals[5] + Vals[1];
  298. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  299. W[15] = W[15] + (rotr(W[0], 7) ^ rotr(W[0], 18) ^ (W[0] >> 3U)) + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  300. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[63] + W[15];
  301. Vals[4] = Vals[4] + Vals[0];
  302. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  303. W[0] = Vals[0] + state0;
  304. W[1] = Vals[1] + state1;
  305. W[2] = Vals[2] + state2;
  306. W[3] = Vals[3] + state3;
  307. W[4] = Vals[4] + state4;
  308. W[5] = Vals[5] + state5;
  309. W[6] = Vals[6] + state6;
  310. W[7] = Vals[7] + state7;
  311. Vals[7] = 0xb0edbdd0 + K[ 0] + W[0];
  312. Vals[3] = 0xa54ff53a + Vals[7];
  313. Vals[7] = Vals[7] + 0x08909ae5U;
  314. Vals[6] = 0x1f83d9abU + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + (0x9b05688cU ^ (Vals[3] & 0xca0b3af3U)) + K[ 1] + W[1];
  315. Vals[2] = 0x3c6ef372U + Vals[6];
  316. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma2(0xbb67ae85U, Vals[7], 0x6a09e667U);
  317. Vals[5] = 0x9b05688cU + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], 0x510e527fU) + K[ 2] + W[2];
  318. Vals[1] = 0xbb67ae85U + Vals[5];
  319. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma2(0x6a09e667U, Vals[6], Vals[7]);
  320. Vals[4] = 0x510e527fU + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[ 3] + W[3];
  321. Vals[0] = 0x6a09e667U + Vals[4];
  322. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  323. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[ 4] + W[4];
  324. Vals[7] = Vals[7] + Vals[3];
  325. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  326. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[ 5] + W[5];
  327. Vals[6] = Vals[6] + Vals[2];
  328. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  329. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[ 6] + W[6];
  330. Vals[5] = Vals[5] + Vals[1];
  331. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  332. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[ 7] + W[7];
  333. Vals[4] = Vals[4] + Vals[0];
  334. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  335. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[ 8] + 0x80000000;
  336. Vals[3] = Vals[3] + Vals[7];
  337. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  338. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[ 9];
  339. Vals[2] = Vals[2] + Vals[6];
  340. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  341. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[10];
  342. Vals[1] = Vals[1] + Vals[5];
  343. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  344. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[11];
  345. Vals[0] = Vals[0] + Vals[4];
  346. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  347. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[12];
  348. Vals[7] = Vals[7] + Vals[3];
  349. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  350. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[13];
  351. Vals[6] = Vals[6] + Vals[2];
  352. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  353. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[14];
  354. Vals[5] = Vals[5] + Vals[1];
  355. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  356. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[15] + 0x00000100U;
  357. Vals[4] = Vals[4] + Vals[0];
  358. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  359. W[0] = W[0] + (rotr(W[1], 7) ^ rotr(W[1], 18) ^ (W[1] >> 3U));
  360. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[16] + W[0];
  361. Vals[3] = Vals[3] + Vals[7];
  362. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  363. W[1] = W[1] + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + 0x00a00000U;
  364. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[17] + W[1];
  365. Vals[2] = Vals[2] + Vals[6];
  366. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  367. W[2] = W[2] + (rotr(W[3], 7) ^ rotr(W[3], 18) ^ (W[3] >> 3U)) + (rotr(W[0], 17) ^ rotr(W[0], 19) ^ (W[0] >> 10U));
  368. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[18] + W[2];
  369. Vals[1] = Vals[1] + Vals[5];
  370. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  371. W[3] = W[3] + (rotr(W[4], 7) ^ rotr(W[4], 18) ^ (W[4] >> 3U)) + (rotr(W[1], 17) ^ rotr(W[1], 19) ^ (W[1] >> 10U));
  372. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[19] + W[3];
  373. Vals[0] = Vals[0] + Vals[4];
  374. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  375. W[4] = W[4] + (rotr(W[5], 7) ^ rotr(W[5], 18) ^ (W[5] >> 3U)) + (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U));
  376. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[20] + W[4];
  377. Vals[7] = Vals[7] + Vals[3];
  378. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  379. W[5] = W[5] + (rotr(W[6], 7) ^ rotr(W[6], 18) ^ (W[6] >> 3U)) + (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  380. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[21] + W[5];
  381. Vals[6] = Vals[6] + Vals[2];
  382. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  383. W[6] = W[6] + (rotr(W[7], 7) ^ rotr(W[7], 18) ^ (W[7] >> 3U)) + 0x00000100U + (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U));
  384. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[22] + W[6];
  385. Vals[5] = Vals[5] + Vals[1];
  386. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  387. W[7] = W[7] + 0x11002000U + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  388. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[23] + W[7];
  389. Vals[4] = Vals[4] + Vals[0];
  390. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  391. W[8] = 0x80000000 + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  392. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[24] + W[8];
  393. Vals[3] = Vals[3] + Vals[7];
  394. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  395. W[9] = W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  396. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[25] + W[9];
  397. Vals[2] = Vals[2] + Vals[6];
  398. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  399. W[10] = W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  400. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[26] + W[10];
  401. Vals[1] = Vals[1] + Vals[5];
  402. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  403. W[11] = W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  404. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[27] + W[11];
  405. Vals[0] = Vals[0] + Vals[4];
  406. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  407. W[12] = W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  408. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[28] + W[12];
  409. Vals[7] = Vals[7] + Vals[3];
  410. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  411. W[13] = W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  412. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[29] + W[13];
  413. Vals[6] = Vals[6] + Vals[2];
  414. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  415. W[14] = 0x00400022U + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  416. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[30] + W[14];
  417. Vals[5] = Vals[5] + Vals[1];
  418. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  419. W[15] = 0x00000100U + (rotr(W[0], 7) ^ rotr(W[0], 18) ^ (W[0] >> 3U)) + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  420. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[31] + W[15];
  421. Vals[4] = Vals[4] + Vals[0];
  422. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  423. W[0] = W[0] + (rotr(W[1], 7) ^ rotr(W[1], 18) ^ (W[1] >> 3U)) + W[9] + (rotr(W[14], 17) ^ rotr(W[14], 19) ^ (W[14] >> 10U));
  424. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[32] + W[0];
  425. Vals[3] = Vals[3] + Vals[7];
  426. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  427. W[1] = W[1] + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + W[10] + (rotr(W[15], 17) ^ rotr(W[15], 19) ^ (W[15] >> 10U));
  428. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[33] + W[1];
  429. Vals[2] = Vals[2] + Vals[6];
  430. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  431. W[2] = W[2] + (rotr(W[3], 7) ^ rotr(W[3], 18) ^ (W[3] >> 3U)) + W[11] + (rotr(W[0], 17) ^ rotr(W[0], 19) ^ (W[0] >> 10U));
  432. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[34] + W[2];
  433. Vals[1] = Vals[1] + Vals[5];
  434. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  435. W[3] = W[3] + (rotr(W[4], 7) ^ rotr(W[4], 18) ^ (W[4] >> 3U)) + W[12] + (rotr(W[1], 17) ^ rotr(W[1], 19) ^ (W[1] >> 10U));
  436. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[35] + W[3];
  437. Vals[0] = Vals[0] + Vals[4];
  438. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  439. W[4] = W[4] + (rotr(W[5], 7) ^ rotr(W[5], 18) ^ (W[5] >> 3U)) + W[13] + (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U));
  440. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[36] + W[4];
  441. Vals[7] = Vals[7] + Vals[3];
  442. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  443. W[5] = W[5] + (rotr(W[6], 7) ^ rotr(W[6], 18) ^ (W[6] >> 3U)) + W[14] + (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  444. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[37] + W[5];
  445. Vals[6] = Vals[6] + Vals[2];
  446. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  447. W[6] = W[6] + (rotr(W[7], 7) ^ rotr(W[7], 18) ^ (W[7] >> 3U)) + W[15] + (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U));
  448. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[38] + W[6];
  449. Vals[5] = Vals[5] + Vals[1];
  450. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  451. W[7] = W[7] + (rotr(W[8], 7) ^ rotr(W[8], 18) ^ (W[8] >> 3U)) + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  452. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[39] + W[7];
  453. Vals[4] = Vals[4] + Vals[0];
  454. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  455. W[8] = W[8] + (rotr(W[9], 7) ^ rotr(W[9], 18) ^ (W[9] >> 3U)) + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  456. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[40] + W[8];
  457. Vals[3] = Vals[3] + Vals[7];
  458. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  459. W[9] = W[9] + (rotr(W[10], 7) ^ rotr(W[10], 18) ^ (W[10] >> 3U)) + W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  460. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[41] + W[9];
  461. Vals[2] = Vals[2] + Vals[6];
  462. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  463. W[10] = W[10] + (rotr(W[11], 7) ^ rotr(W[11], 18) ^ (W[11] >> 3U)) + W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  464. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[42] + W[10];
  465. Vals[1] = Vals[1] + Vals[5];
  466. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  467. W[11] = W[11] + (rotr(W[12], 7) ^ rotr(W[12], 18) ^ (W[12] >> 3U)) + W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  468. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[43] + W[11];
  469. Vals[0] = Vals[0] + Vals[4];
  470. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  471. W[12] = W[12] + (rotr(W[13], 7) ^ rotr(W[13], 18) ^ (W[13] >> 3U)) + W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  472. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[44] + W[12];
  473. Vals[7] = Vals[7] + Vals[3];
  474. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  475. W[13] = W[13] + (rotr(W[14], 7) ^ rotr(W[14], 18) ^ (W[14] >> 3U)) + W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  476. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[45] + W[13];
  477. Vals[6] = Vals[6] + Vals[2];
  478. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  479. W[14] = W[14] + (rotr(W[15], 7) ^ rotr(W[15], 18) ^ (W[15] >> 3U)) + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  480. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[46] + W[14];
  481. Vals[5] = Vals[5] + Vals[1];
  482. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  483. W[15] = W[15] + (rotr(W[0], 7) ^ rotr(W[0], 18) ^ (W[0] >> 3U)) + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  484. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[47] + W[15];
  485. Vals[4] = Vals[4] + Vals[0];
  486. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  487. W[0] = W[0] + (rotr(W[1], 7) ^ rotr(W[1], 18) ^ (W[1] >> 3U)) + W[9] + (rotr(W[14], 17) ^ rotr(W[14], 19) ^ (W[14] >> 10U));
  488. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[48] + W[0];
  489. Vals[3] = Vals[3] + Vals[7];
  490. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  491. W[1] = W[1] + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + W[10] + (rotr(W[15], 17) ^ rotr(W[15], 19) ^ (W[15] >> 10U));
  492. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[49] + W[1];
  493. Vals[2] = Vals[2] + Vals[6];
  494. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  495. W[2] = W[2] + (rotr(W[3], 7) ^ rotr(W[3], 18) ^ (W[3] >> 3U)) + W[11] + (rotr(W[0], 17) ^ rotr(W[0], 19) ^ (W[0] >> 10U));
  496. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[50] + W[2];
  497. Vals[1] = Vals[1] + Vals[5];
  498. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  499. W[3] = W[3] + (rotr(W[4], 7) ^ rotr(W[4], 18) ^ (W[4] >> 3U)) + W[12] + (rotr(W[1], 17) ^ rotr(W[1], 19) ^ (W[1] >> 10U));
  500. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[51] + W[3];
  501. Vals[0] = Vals[0] + Vals[4];
  502. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  503. W[4] = W[4] + (rotr(W[5], 7) ^ rotr(W[5], 18) ^ (W[5] >> 3U)) + W[13] + (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U));
  504. Vals[3] = Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[52] + W[4];
  505. Vals[7] = Vals[7] + Vals[3];
  506. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  507. W[5] = W[5] + (rotr(W[6], 7) ^ rotr(W[6], 18) ^ (W[6] >> 3U)) + W[14] + (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  508. Vals[2] = Vals[2] + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], Vals[1]) + K[53] + W[5];
  509. Vals[6] = Vals[6] + Vals[2];
  510. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  511. W[6] = W[6] + (rotr(W[7], 7) ^ rotr(W[7], 18) ^ (W[7] >> 3U)) + W[15] + (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U));
  512. Vals[1] = Vals[1] + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[54] + W[6];
  513. Vals[5] = Vals[5] + Vals[1];
  514. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  515. W[7] = W[7] + (rotr(W[8], 7) ^ rotr(W[8], 18) ^ (W[8] >> 3U)) + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  516. Vals[0] = Vals[0] + (rotr(Vals[5], 6) ^ rotr(Vals[5], 11) ^ rotr(Vals[5], 25)) + ch(Vals[5], Vals[6], Vals[7]) + K[55] + W[7];
  517. Vals[4] = Vals[4] + Vals[0];
  518. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  519. W[8] = W[8] + (rotr(W[9], 7) ^ rotr(W[9], 18) ^ (W[9] >> 3U)) + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  520. Vals[7] = Vals[7] + (rotr(Vals[4], 6) ^ rotr(Vals[4], 11) ^ rotr(Vals[4], 25)) + ch(Vals[4], Vals[5], Vals[6]) + K[56] + W[8];
  521. Vals[3] = Vals[3] + Vals[7];
  522. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  523. W[9] = W[9] + (rotr(W[10], 7) ^ rotr(W[10], 18) ^ (W[10] >> 3U)) + W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  524. Vals[6] = Vals[6] + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + ch(Vals[3], Vals[4], Vals[5]) + K[57] + W[9];
  525. Vals[2] = Vals[2] + Vals[6];
  526. W[10] = W[10] + (rotr(W[11], 7) ^ rotr(W[11], 18) ^ (W[11] >> 3U)) + W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  527. Vals[5] = Vals[5] + (rotr(Vals[2], 6) ^ rotr(Vals[2], 11) ^ rotr(Vals[2], 25)) + ch(Vals[2], Vals[3], Vals[4]) + K[58] + W[10];
  528. Vals[1] = Vals[1] + Vals[5];
  529. W[11] = W[11] + (rotr(W[12], 7) ^ rotr(W[12], 18) ^ (W[12] >> 3U)) + W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  530. Vals[4] = Vals[4] + (rotr(Vals[1], 6) ^ rotr(Vals[1], 11) ^ rotr(Vals[1], 25)) + ch(Vals[1], Vals[2], Vals[3]) + K[59] + W[11];
  531. Vals[0] = Vals[0] + Vals[4];
  532. W[12] = W[12] + (rotr(W[13], 7) ^ rotr(W[13], 18) ^ (W[13] >> 3U)) + W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  533. Vals[7] = Vals[7] + Vals[3] + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], Vals[1], Vals[2]) + K[60] + W[12];
  534. Vals[7]+=0x5be0cd19U;
  535. #if defined(VECTORS4) || defined(VECTORS2)
  536. if (Vals[7].x == 0)
  537. {
  538. for (it = 0; it != 127; it++) {
  539. if (!output[it]) {
  540. output[it] = nonce.x;
  541. output[127] = 1;
  542. break;
  543. }
  544. }
  545. }
  546. if (Vals[7].y == 0)
  547. {
  548. for (it = 0; it != 127; it++) {
  549. if (!output[it]) {
  550. output[it] = nonce.y;
  551. output[127] = 1;
  552. break;
  553. }
  554. }
  555. }
  556. #ifdef VECTORS4
  557. if (Vals[7].z == 0)
  558. {
  559. for (it = 0; it != 127; it++) {
  560. if (!output[it]) {
  561. output[it] = nonce.z;
  562. output[127] = 1;
  563. break;
  564. }
  565. }
  566. }
  567. if (Vals[7].w == 0)
  568. {
  569. for (it = 0; it != 127; it++) {
  570. if (!output[it]) {
  571. output[it] = nonce.w;
  572. output[127] = 1;
  573. break;
  574. }
  575. }
  576. }
  577. #endif
  578. #else
  579. if (Vals[7] == 0)
  580. {
  581. for (it = 0; it != 127; it++) {
  582. if (!output[it]) {
  583. output[it] = nonce;
  584. output[127] = 1;
  585. break;
  586. }
  587. }
  588. }
  589. #endif
  590. }