poclbm110717.cl 46 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658
  1. // -ck modified kernel taken from Phoenix taken from poclbm, with aspects of
  2. // phatk and others.
  3. // Modified version copyright 2011 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. // The X is a placeholder for patching to suit hardware
  7. #define VECTORSX
  8. #ifdef VECTORS4
  9. typedef uint4 u;
  10. #elif defined VECTORS2
  11. typedef uint2 u;
  12. #else
  13. typedef uint u;
  14. #endif
  15. __constant uint K[64] = {
  16. 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
  17. 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
  18. 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
  19. 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
  20. 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
  21. 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
  22. 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
  23. 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
  24. };
  25. // This part is not from the stock poclbm kernel. It's part of an optimization
  26. // added in the Phoenix Miner.
  27. // Some AMD devices have a BFI_INT opcode, which behaves exactly like the
  28. // SHA-256 ch function, but provides it in exactly one instruction. If
  29. // detected, use it for ch. Otherwise, construct ch out of simpler logical
  30. // primitives.
  31. #define BFI_INTX
  32. #define BITALIGNX
  33. #ifdef BFI_INT
  34. // Well, slight problem... It turns out BFI_INT isn't actually exposed to
  35. // OpenCL (or CAL IL for that matter) in any way. However, there is
  36. // a similar instruction, BYTE_ALIGN_INT, which is exposed to OpenCL via
  37. // amd_bytealign, takes the same inputs, and provides the same output.
  38. // We can use that as a placeholder for BFI_INT and have the application
  39. // patch it after compilation.
  40. // This is the BFI_INT function
  41. #define ch(x, y, z) amd_bytealign(x, y, z)
  42. // Ma can also be implemented in terms of BFI_INT...
  43. #define Ma(x, y, z) amd_bytealign( (z^x), (y), (x) )
  44. #else
  45. #define ch(x, y, z) (z ^ (x & (y ^ z)))
  46. #define Ma(x, y, z) ((x & z) | (y & (x | z)))
  47. #endif
  48. #ifdef BITALIGN
  49. #pragma OPENCL EXTENSION cl_amd_media_ops : enable
  50. #define rotr(x, y) amd_bitalign((u)x, (u)x, (u)y)
  51. #else
  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 uint 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];
  68. u nonce;
  69. #ifdef VECTORS4
  70. nonce = base + (get_global_id(0)<<2) + (uint4)(0, 1, 2, 3);
  71. #elif defined VECTORS2
  72. nonce = base + (get_global_id(0)<<1) + (uint2)(0, 1);
  73. #else
  74. nonce = base + get_global_id(0);
  75. #endif
  76. W[3] = nonce + fw3;
  77. Vals[4] = fcty_e + nonce;
  78. Vals[0] = state0 + Vals[4];
  79. Vals[4] = Vals[4] + fcty_e2;
  80. Vals[3] = d1 + (rotr(Vals[0], 6) ^ rotr(Vals[0], 11) ^ rotr(Vals[0], 25)) + ch(Vals[0], b1, c1) + K[ 4] + 0x80000000;
  81. Vals[7] = h1 + Vals[3];
  82. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma2(g1, Vals[4], f1);
  83. Vals[2] = c1 + (rotr(Vals[7], 6) ^ rotr(Vals[7], 11) ^ rotr(Vals[7], 25)) + ch(Vals[7], Vals[0], b1) + K[ 5];
  84. Vals[6] = g1 + Vals[2];
  85. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma2(f1, Vals[3], Vals[4]);
  86. Vals[1] = b1 + (rotr(Vals[6], 6) ^ rotr(Vals[6], 11) ^ rotr(Vals[6], 25)) + ch(Vals[6], Vals[7], Vals[0]) + K[ 6];
  87. Vals[5] = f1 + Vals[1];
  88. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  89. 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];
  90. Vals[4] = Vals[4] + Vals[0];
  91. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  92. 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];
  93. Vals[3] = Vals[3] + Vals[7];
  94. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  95. 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];
  96. Vals[2] = Vals[2] + Vals[6];
  97. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  98. 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];
  99. Vals[1] = Vals[1] + Vals[5];
  100. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  101. 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];
  102. Vals[0] = Vals[0] + Vals[4];
  103. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  104. 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];
  105. Vals[7] = Vals[7] + Vals[3];
  106. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  107. 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];
  108. Vals[6] = Vals[6] + Vals[2];
  109. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  110. 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];
  111. Vals[5] = Vals[5] + Vals[1];
  112. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  113. 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;
  114. Vals[4] = Vals[4] + Vals[0];
  115. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  116. 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;
  117. Vals[3] = Vals[3] + Vals[7];
  118. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  119. 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;
  120. Vals[2] = Vals[2] + Vals[6];
  121. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  122. W[2] = (rotr(nonce, 7) ^ rotr(nonce, 18) ^ (nonce >> 3U)) + fw2;
  123. 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];
  124. Vals[1] = Vals[1] + Vals[5];
  125. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  126. 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];
  127. Vals[0] = Vals[0] + Vals[4];
  128. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  129. W[4] = (rotr(W[2], 17) ^ rotr(W[2], 19) ^ (W[2] >> 10U)) + 0x80000000;
  130. 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];
  131. Vals[7] = Vals[7] + Vals[3];
  132. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  133. W[5] = (rotr(W[3], 17) ^ rotr(W[3], 19) ^ (W[3] >> 10U));
  134. 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];
  135. Vals[6] = Vals[6] + Vals[2];
  136. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  137. W[6] = (rotr(W[4], 17) ^ rotr(W[4], 19) ^ (W[4] >> 10U)) + 0x00000280U;
  138. 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];
  139. Vals[5] = Vals[5] + Vals[1];
  140. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  141. W[7] = (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U)) + fw0;
  142. 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];
  143. Vals[4] = Vals[4] + Vals[0];
  144. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  145. W[8] = (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U)) + fw1;
  146. 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];
  147. Vals[3] = Vals[3] + Vals[7];
  148. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  149. W[9] = W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  150. 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];
  151. Vals[2] = Vals[2] + Vals[6];
  152. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  153. W[10] = W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  154. 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];
  155. Vals[1] = Vals[1] + Vals[5];
  156. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  157. W[11] = W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  158. 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];
  159. Vals[0] = Vals[0] + Vals[4];
  160. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  161. W[12] = W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  162. 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];
  163. Vals[7] = Vals[7] + Vals[3];
  164. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  165. W[13] = W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  166. 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];
  167. Vals[6] = Vals[6] + Vals[2];
  168. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  169. W[14] = 0x00a00055U + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  170. 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];
  171. Vals[5] = Vals[5] + Vals[1];
  172. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  173. W[15] = fw15 + W[8] + (rotr(W[13], 17) ^ rotr(W[13], 19) ^ (W[13] >> 10U));
  174. 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];
  175. Vals[4] = Vals[4] + Vals[0];
  176. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  177. W[0] = fw01r + W[9] + (rotr(W[14], 17) ^ rotr(W[14], 19) ^ (W[14] >> 10U));
  178. 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];
  179. Vals[3] = Vals[3] + Vals[7];
  180. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  181. 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));
  182. 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];
  183. Vals[2] = Vals[2] + Vals[6];
  184. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  185. 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));
  186. 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];
  187. Vals[1] = Vals[1] + Vals[5];
  188. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  189. 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));
  190. 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];
  191. Vals[0] = Vals[0] + Vals[4];
  192. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  193. 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));
  194. 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];
  195. Vals[7] = Vals[7] + Vals[3];
  196. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  197. 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));
  198. 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];
  199. Vals[6] = Vals[6] + Vals[2];
  200. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  201. 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));
  202. 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];
  203. Vals[5] = Vals[5] + Vals[1];
  204. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  205. 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));
  206. 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];
  207. Vals[4] = Vals[4] + Vals[0];
  208. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  209. 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));
  210. 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];
  211. Vals[3] = Vals[3] + Vals[7];
  212. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  213. 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));
  214. 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];
  215. Vals[2] = Vals[2] + Vals[6];
  216. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  217. 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));
  218. 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];
  219. Vals[1] = Vals[1] + Vals[5];
  220. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  221. 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));
  222. 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];
  223. Vals[0] = Vals[0] + Vals[4];
  224. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  225. 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));
  226. 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];
  227. Vals[7] = Vals[7] + Vals[3];
  228. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  229. 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));
  230. 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];
  231. Vals[6] = Vals[6] + Vals[2];
  232. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  233. 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));
  234. 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];
  235. Vals[5] = Vals[5] + Vals[1];
  236. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  237. 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));
  238. 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];
  239. Vals[4] = Vals[4] + Vals[0];
  240. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  241. 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));
  242. 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];
  243. Vals[3] = Vals[3] + Vals[7];
  244. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  245. 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));
  246. 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];
  247. Vals[2] = Vals[2] + Vals[6];
  248. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  249. 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));
  250. 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];
  251. Vals[1] = Vals[1] + Vals[5];
  252. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  253. 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));
  254. 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];
  255. Vals[0] = Vals[0] + Vals[4];
  256. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  257. 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));
  258. 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];
  259. Vals[7] = Vals[7] + Vals[3];
  260. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  261. 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));
  262. 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];
  263. Vals[6] = Vals[6] + Vals[2];
  264. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  265. 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));
  266. 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];
  267. Vals[5] = Vals[5] + Vals[1];
  268. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  269. 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));
  270. 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];
  271. Vals[4] = Vals[4] + Vals[0];
  272. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  273. 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));
  274. 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];
  275. Vals[3] = Vals[3] + Vals[7];
  276. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  277. 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));
  278. 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];
  279. Vals[2] = Vals[2] + Vals[6];
  280. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  281. 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));
  282. 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];
  283. Vals[1] = Vals[1] + Vals[5];
  284. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  285. 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));
  286. 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];
  287. Vals[0] = Vals[0] + Vals[4];
  288. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  289. 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));
  290. 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];
  291. Vals[7] = Vals[7] + Vals[3];
  292. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  293. 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));
  294. 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];
  295. Vals[6] = Vals[6] + Vals[2];
  296. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  297. 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));
  298. 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];
  299. Vals[5] = Vals[5] + Vals[1];
  300. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  301. 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));
  302. 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];
  303. Vals[4] = Vals[4] + Vals[0];
  304. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  305. W[0] = Vals[0] + state0;
  306. W[1] = Vals[1] + state1;
  307. W[2] = Vals[2] + state2;
  308. W[3] = Vals[3] + state3;
  309. W[4] = Vals[4] + state4;
  310. W[5] = Vals[5] + state5;
  311. W[6] = Vals[6] + state6;
  312. W[7] = Vals[7] + state7;
  313. Vals[7] = 0xb0edbdd0 + K[ 0] + W[0];
  314. Vals[3] = 0xa54ff53a + Vals[7];
  315. Vals[7] = Vals[7] + 0x08909ae5U;
  316. Vals[6] = 0x1f83d9abU + (rotr(Vals[3], 6) ^ rotr(Vals[3], 11) ^ rotr(Vals[3], 25)) + (0x9b05688cU ^ (Vals[3] & 0xca0b3af3U)) + K[ 1] + W[1];
  317. Vals[2] = 0x3c6ef372U + Vals[6];
  318. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma2(0xbb67ae85U, Vals[7], 0x6a09e667U);
  319. 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];
  320. Vals[1] = 0xbb67ae85U + Vals[5];
  321. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma2(0x6a09e667U, Vals[6], Vals[7]);
  322. 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];
  323. Vals[0] = 0x6a09e667U + Vals[4];
  324. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  325. 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];
  326. Vals[7] = Vals[7] + Vals[3];
  327. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  328. 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];
  329. Vals[6] = Vals[6] + Vals[2];
  330. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  331. 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];
  332. Vals[5] = Vals[5] + Vals[1];
  333. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  334. 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];
  335. Vals[4] = Vals[4] + Vals[0];
  336. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  337. 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;
  338. Vals[3] = Vals[3] + Vals[7];
  339. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  340. 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];
  341. Vals[2] = Vals[2] + Vals[6];
  342. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  343. 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];
  344. Vals[1] = Vals[1] + Vals[5];
  345. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  346. 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];
  347. Vals[0] = Vals[0] + Vals[4];
  348. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  349. 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];
  350. Vals[7] = Vals[7] + Vals[3];
  351. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  352. 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];
  353. Vals[6] = Vals[6] + Vals[2];
  354. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  355. 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];
  356. Vals[5] = Vals[5] + Vals[1];
  357. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  358. 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;
  359. Vals[4] = Vals[4] + Vals[0];
  360. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  361. W[0] = W[0] + (rotr(W[1], 7) ^ rotr(W[1], 18) ^ (W[1] >> 3U));
  362. 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];
  363. Vals[3] = Vals[3] + Vals[7];
  364. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  365. W[1] = W[1] + (rotr(W[2], 7) ^ rotr(W[2], 18) ^ (W[2] >> 3U)) + 0x00a00000U;
  366. 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];
  367. Vals[2] = Vals[2] + Vals[6];
  368. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  369. 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));
  370. 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];
  371. Vals[1] = Vals[1] + Vals[5];
  372. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  373. 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));
  374. 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];
  375. Vals[0] = Vals[0] + Vals[4];
  376. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  377. 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));
  378. 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];
  379. Vals[7] = Vals[7] + Vals[3];
  380. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  381. 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));
  382. 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];
  383. Vals[6] = Vals[6] + Vals[2];
  384. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  385. 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));
  386. 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];
  387. Vals[5] = Vals[5] + Vals[1];
  388. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  389. W[7] = W[7] + 0x11002000U + W[0] + (rotr(W[5], 17) ^ rotr(W[5], 19) ^ (W[5] >> 10U));
  390. 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];
  391. Vals[4] = Vals[4] + Vals[0];
  392. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  393. W[8] = 0x80000000 + W[1] + (rotr(W[6], 17) ^ rotr(W[6], 19) ^ (W[6] >> 10U));
  394. 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];
  395. Vals[3] = Vals[3] + Vals[7];
  396. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  397. W[9] = W[2] + (rotr(W[7], 17) ^ rotr(W[7], 19) ^ (W[7] >> 10U));
  398. 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];
  399. Vals[2] = Vals[2] + Vals[6];
  400. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  401. W[10] = W[3] + (rotr(W[8], 17) ^ rotr(W[8], 19) ^ (W[8] >> 10U));
  402. 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];
  403. Vals[1] = Vals[1] + Vals[5];
  404. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  405. W[11] = W[4] + (rotr(W[9], 17) ^ rotr(W[9], 19) ^ (W[9] >> 10U));
  406. 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];
  407. Vals[0] = Vals[0] + Vals[4];
  408. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  409. W[12] = W[5] + (rotr(W[10], 17) ^ rotr(W[10], 19) ^ (W[10] >> 10U));
  410. 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];
  411. Vals[7] = Vals[7] + Vals[3];
  412. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  413. W[13] = W[6] + (rotr(W[11], 17) ^ rotr(W[11], 19) ^ (W[11] >> 10U));
  414. 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];
  415. Vals[6] = Vals[6] + Vals[2];
  416. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  417. W[14] = 0x00400022U + W[7] + (rotr(W[12], 17) ^ rotr(W[12], 19) ^ (W[12] >> 10U));
  418. 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];
  419. Vals[5] = Vals[5] + Vals[1];
  420. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  421. 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));
  422. 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];
  423. Vals[4] = Vals[4] + Vals[0];
  424. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  425. 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));
  426. 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];
  427. Vals[3] = Vals[3] + Vals[7];
  428. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  429. 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));
  430. 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];
  431. Vals[2] = Vals[2] + Vals[6];
  432. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  433. 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));
  434. 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];
  435. Vals[1] = Vals[1] + Vals[5];
  436. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  437. 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));
  438. 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];
  439. Vals[0] = Vals[0] + Vals[4];
  440. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  441. 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));
  442. 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];
  443. Vals[7] = Vals[7] + Vals[3];
  444. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  445. 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));
  446. 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];
  447. Vals[6] = Vals[6] + Vals[2];
  448. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  449. 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));
  450. 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];
  451. Vals[5] = Vals[5] + Vals[1];
  452. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  453. 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));
  454. 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];
  455. Vals[4] = Vals[4] + Vals[0];
  456. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  457. 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));
  458. 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];
  459. Vals[3] = Vals[3] + Vals[7];
  460. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  461. 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));
  462. 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];
  463. Vals[2] = Vals[2] + Vals[6];
  464. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  465. 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));
  466. 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];
  467. Vals[1] = Vals[1] + Vals[5];
  468. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  469. 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));
  470. 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];
  471. Vals[0] = Vals[0] + Vals[4];
  472. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  473. 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));
  474. 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];
  475. Vals[7] = Vals[7] + Vals[3];
  476. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  477. 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));
  478. 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];
  479. Vals[6] = Vals[6] + Vals[2];
  480. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  481. 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));
  482. 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];
  483. Vals[5] = Vals[5] + Vals[1];
  484. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  485. 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));
  486. 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];
  487. Vals[4] = Vals[4] + Vals[0];
  488. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  489. 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));
  490. 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];
  491. Vals[3] = Vals[3] + Vals[7];
  492. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  493. 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));
  494. 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];
  495. Vals[2] = Vals[2] + Vals[6];
  496. Vals[6] = Vals[6] + (rotr(Vals[7], 2) ^ rotr(Vals[7], 13) ^ rotr(Vals[7], 22)) + Ma(Vals[1], Vals[7], Vals[0]);
  497. 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));
  498. 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];
  499. Vals[1] = Vals[1] + Vals[5];
  500. Vals[5] = Vals[5] + (rotr(Vals[6], 2) ^ rotr(Vals[6], 13) ^ rotr(Vals[6], 22)) + Ma(Vals[0], Vals[6], Vals[7]);
  501. 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));
  502. 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];
  503. Vals[0] = Vals[0] + Vals[4];
  504. Vals[4] = Vals[4] + (rotr(Vals[5], 2) ^ rotr(Vals[5], 13) ^ rotr(Vals[5], 22)) + Ma(Vals[7], Vals[5], Vals[6]);
  505. 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));
  506. 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];
  507. Vals[7] = Vals[7] + Vals[3];
  508. Vals[3] = Vals[3] + (rotr(Vals[4], 2) ^ rotr(Vals[4], 13) ^ rotr(Vals[4], 22)) + Ma(Vals[6], Vals[4], Vals[5]);
  509. 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));
  510. 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];
  511. Vals[6] = Vals[6] + Vals[2];
  512. Vals[2] = Vals[2] + (rotr(Vals[3], 2) ^ rotr(Vals[3], 13) ^ rotr(Vals[3], 22)) + Ma(Vals[5], Vals[3], Vals[4]);
  513. 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));
  514. 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];
  515. Vals[5] = Vals[5] + Vals[1];
  516. Vals[1] = Vals[1] + (rotr(Vals[2], 2) ^ rotr(Vals[2], 13) ^ rotr(Vals[2], 22)) + Ma(Vals[4], Vals[2], Vals[3]);
  517. 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));
  518. 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];
  519. Vals[4] = Vals[4] + Vals[0];
  520. Vals[0] = Vals[0] + (rotr(Vals[1], 2) ^ rotr(Vals[1], 13) ^ rotr(Vals[1], 22)) + Ma(Vals[3], Vals[1], Vals[2]);
  521. 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));
  522. 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];
  523. Vals[3] = Vals[3] + Vals[7];
  524. Vals[7] = Vals[7] + (rotr(Vals[0], 2) ^ rotr(Vals[0], 13) ^ rotr(Vals[0], 22)) + Ma(Vals[2], Vals[0], Vals[1]);
  525. 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));
  526. 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];
  527. Vals[2] = Vals[2] + Vals[6];
  528. 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));
  529. 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];
  530. Vals[1] = Vals[1] + Vals[5];
  531. 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));
  532. 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];
  533. Vals[0] = Vals[0] + Vals[4];
  534. 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));
  535. 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];
  536. Vals[7]+=0x5be0cd19U;
  537. #define MAXBUFFERS (4095)
  538. #define NFLAG (0xFFFUL)
  539. #if defined(VECTORS4) || defined(VECTORS2)
  540. if (Vals[7].x == 0)
  541. {
  542. output[MAXBUFFERS] = output[NFLAG & nonce.x] = nonce.x;
  543. }
  544. if (Vals[7].y == 0)
  545. {
  546. output[MAXBUFFERS] = output[NFLAG & nonce.y] = nonce.y;
  547. }
  548. #ifdef VECTORS4
  549. if (Vals[7].z == 0)
  550. {
  551. output[MAXBUFFERS] = output[NFLAG & nonce.z] = nonce.z;
  552. }
  553. if (Vals[7].w == 0)
  554. {
  555. output[MAXBUFFERS] = output[NFLAG & nonce.w] = nonce.w;
  556. }
  557. #endif
  558. #else
  559. if (Vals[7] == 0)
  560. {
  561. output[MAXBUFFERS] = output[NFLAG & nonce] = nonce;
  562. }
  563. #endif
  564. }