X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm08600_a3.cl;h=30bf6bf29bafd7ca79d1e1cae3316f4b676091bb;hb=161a6eb4bc643d8e636e96eda613f5137d30da59;hp=23fab4aab49291e6c056a378f69f48ba2d125901;hpb=e4465d7dcb88426532f6b0488cf9bc1cd3b06cc6;p=hashcat.git diff --git a/OpenCL/m08600_a3.cl b/OpenCL/m08600_a3.cl index 23fab4a..30bf6bf 100644 --- a/OpenCL/m08600_a3.cl +++ b/OpenCL/m08600_a3.cl @@ -1,24 +1,21 @@ /** - * Author......: Jens Steube + * Authors.....: Jens Steube + * Gabriele Gristina + * magnum + * * License.....: MIT */ #define _LOTUS5_ -#include "include/constants.h" -#include "include/kernel_vendor.h" +#define NEW_SIMD_CODE -#define DGST_R0 0 -#define DGST_R1 1 -#define DGST_R2 2 -#define DGST_R3 3 - -#include "include/kernel_functions.c" -#include "types_ocl.c" -#include "common.c" - -#define COMPARE_S "check_single_comp4.c" -#define COMPARE_M "check_multi_comp4.c" +#include "inc_vendor.cl" +#include "inc_hash_constants.h" +#include "inc_hash_functions.cl" +#include "inc_types.cl" +#include "inc_common.cl" +#include "inc_simd.cl" __constant u32 lotus_magic_table[256] = { @@ -56,65 +53,76 @@ __constant u32 lotus_magic_table[256] = 0x29, 0x39, 0xb9, 0xe9, 0x4c, 0xff, 0x43, 0xab, }; -#define BOX(S,i) (S)[(i)] - -static void lotus_mix (u32 *in, __local u32 s_lotus_magic_table[256]) +#if VECT_SIZE == 1 +#define BOX1(S,i) (S)[(i)] +#elif VECT_SIZE == 2 +#define BOX1(S,i) (u32x) ((S)[(i).s0], (S)[(i).s1]) +#elif VECT_SIZE == 4 +#define BOX1(S,i) (u32x) ((S)[(i).s0], (S)[(i).s1], (S)[(i).s2], (S)[(i).s3]) +#elif VECT_SIZE == 8 +#define BOX1(S,i) (u32x) ((S)[(i).s0], (S)[(i).s1], (S)[(i).s2], (S)[(i).s3], (S)[(i).s4], (S)[(i).s5], (S)[(i).s6], (S)[(i).s7]) +#elif VECT_SIZE == 16 +#define BOX1(S,i) (u32x) ((S)[(i).s0], (S)[(i).s1], (S)[(i).s2], (S)[(i).s3], (S)[(i).s4], (S)[(i).s5], (S)[(i).s6], (S)[(i).s7], (S)[(i).s8], (S)[(i).s9], (S)[(i).sa], (S)[(i).sb], (S)[(i).sc], (S)[(i).sd], (S)[(i).se], (S)[(i).sf]) +#endif + +void lotus_mix (u32x *in, __local u32 *s_lotus_magic_table) { - u32 p = 0; + u32x p = 0; for (int i = 0; i < 18; i++) { u32 s = 48; - #pragma unroll 12 for (int j = 0; j < 12; j++) { - u32 tmp_in = in[j]; - u32 tmp_out = 0; + u32x tmp_in = in[j]; + u32x tmp_out = 0; - p = (p + s--) & 0xff; p = ((tmp_in >> 0) & 0xff) ^ BOX (s_lotus_magic_table, p); tmp_out |= p << 0; - p = (p + s--) & 0xff; p = ((tmp_in >> 8) & 0xff) ^ BOX (s_lotus_magic_table, p); tmp_out |= p << 8; - p = (p + s--) & 0xff; p = ((tmp_in >> 16) & 0xff) ^ BOX (s_lotus_magic_table, p); tmp_out |= p << 16; - p = (p + s--) & 0xff; p = ((tmp_in >> 24) & 0xff) ^ BOX (s_lotus_magic_table, p); tmp_out |= p << 24; + p = (p + s--) & 0xff; p = ((tmp_in >> 0) & 0xff) ^ BOX1 (s_lotus_magic_table, p); tmp_out |= p << 0; + p = (p + s--) & 0xff; p = ((tmp_in >> 8) & 0xff) ^ BOX1 (s_lotus_magic_table, p); tmp_out |= p << 8; + p = (p + s--) & 0xff; p = ((tmp_in >> 16) & 0xff) ^ BOX1 (s_lotus_magic_table, p); tmp_out |= p << 16; + p = (p + s--) & 0xff; p = ((tmp_in >> 24) & 0xff) ^ BOX1 (s_lotus_magic_table, p); tmp_out |= p << 24; in[j] = tmp_out; } } } -static void lotus_transform_password (u32 in[4], u32 out[4], __local u32 s_lotus_magic_table[256]) +void lotus_transform_password (u32x in[4], u32x out[4], __local u32 *s_lotus_magic_table) { - u32 t = out[3] >> 24; + u32x t = out[3] >> 24; - u32 c; + u32x c; - #pragma unroll 4 + #ifdef _unroll + #pragma unroll + #endif for (int i = 0; i < 4; i++) { - t ^= (in[i] >> 0) & 0xff; c = BOX (s_lotus_magic_table, t); out[i] ^= c << 0; t = ((out[i] >> 0) & 0xff); - t ^= (in[i] >> 8) & 0xff; c = BOX (s_lotus_magic_table, t); out[i] ^= c << 8; t = ((out[i] >> 8) & 0xff); - t ^= (in[i] >> 16) & 0xff; c = BOX (s_lotus_magic_table, t); out[i] ^= c << 16; t = ((out[i] >> 16) & 0xff); - t ^= (in[i] >> 24) & 0xff; c = BOX (s_lotus_magic_table, t); out[i] ^= c << 24; t = ((out[i] >> 24) & 0xff); + t ^= (in[i] >> 0) & 0xff; c = BOX1 (s_lotus_magic_table, t); out[i] ^= c << 0; t = ((out[i] >> 0) & 0xff); + t ^= (in[i] >> 8) & 0xff; c = BOX1 (s_lotus_magic_table, t); out[i] ^= c << 8; t = ((out[i] >> 8) & 0xff); + t ^= (in[i] >> 16) & 0xff; c = BOX1 (s_lotus_magic_table, t); out[i] ^= c << 16; t = ((out[i] >> 16) & 0xff); + t ^= (in[i] >> 24) & 0xff; c = BOX1 (s_lotus_magic_table, t); out[i] ^= c << 24; t = ((out[i] >> 24) & 0xff); } } -static void pad (u32 w[4], const u32 len) +void pad (u32 w[4], const u32 len) { const u32 val = 16 - len; const u32 mask1 = val << 24; const u32 mask2 = val << 16 - | val << 24; + | val << 24; const u32 mask3 = val << 8 - | val << 16 - | val << 24; + | val << 16 + | val << 24; const u32 mask4 = val << 0 - | val << 8 - | val << 16 - | val << 24; + | val << 8 + | val << 16 + | val << 24; switch (len) { @@ -177,9 +185,9 @@ static void pad (u32 w[4], const u32 len) } } -static void mdtransform_norecalc (u32 state[4], u32 block[4], __local u32 s_lotus_magic_table[256]) +void mdtransform_norecalc (u32x state[4], u32x block[4], __local u32 *s_lotus_magic_table) { - u32 x[12]; + u32x x[12]; x[ 0] = state[0]; x[ 1] = state[1]; @@ -202,35 +210,28 @@ static void mdtransform_norecalc (u32 state[4], u32 block[4], __local u32 s_lotu state[3] = x[3]; } -static void mdtransform (u32 state[4], u32 checksum[4], u32 block[4], __local u32 s_lotus_magic_table[256]) +void mdtransform (u32x state[4], u32x checksum[4], u32x block[4], __local u32 *s_lotus_magic_table) { mdtransform_norecalc (state, block, s_lotus_magic_table); lotus_transform_password (block, checksum, s_lotus_magic_table); } -static void domino_big_md (const u32 saved_key[16], const u32 size, u32 state[4], __local u32 s_lotus_magic_table[256]) +void domino_big_md (const u32x saved_key[4], const u32 size, u32x state[4], __local u32 *s_lotus_magic_table) { - u32 checksum[4]; + u32x checksum[4]; checksum[0] = 0; checksum[1] = 0; checksum[2] = 0; checksum[3] = 0; - u32 block[4]; - - block[0] = saved_key[0]; - block[1] = saved_key[1]; - block[2] = saved_key[2]; - block[3] = saved_key[3]; - - mdtransform (state, checksum, block, s_lotus_magic_table); + mdtransform (state, checksum, saved_key, s_lotus_magic_table); mdtransform_norecalc (state, checksum, s_lotus_magic_table); } -static void m08600m (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 pw_len, __global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset) +void m08600m (__local u32 *s_lotus_magic_table, u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset) { /** * modifier @@ -240,25 +241,10 @@ static void m08600m (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 const u32 lid = get_local_id (0); /** - * padding + * base */ - if (pw_len < 16) - { - pad (&w[ 0], pw_len & 0xf); - } - else if (pw_len < 32) - { - pad (&w[ 4], pw_len & 0xf); - } - else if (pw_len < 48) - { - pad (&w[ 8], pw_len & 0xf); - } - else if (pw_len < 64) - { - pad (&w[12], pw_len & 0xf); - } + pad (&w[ 0], pw_len); /** * loop @@ -266,50 +252,33 @@ static void m08600m (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 u32 w0l = w[0]; - for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) { - const u32 w0r = words_buf_r[il_pos]; - - const u32 w0 = w0l | w0r; - - u32 w_tmp[16]; - - w_tmp[ 0] = w0; - w_tmp[ 1] = w[ 1]; - w_tmp[ 2] = w[ 2]; - w_tmp[ 3] = w[ 3]; - w_tmp[ 4] = w[ 4]; - w_tmp[ 5] = w[ 5]; - w_tmp[ 6] = w[ 6]; - w_tmp[ 7] = w[ 7]; - w_tmp[ 8] = w[ 8]; - w_tmp[ 9] = w[ 9]; - w_tmp[10] = w[10]; - w_tmp[11] = w[11]; - w_tmp[12] = w[12]; - w_tmp[13] = w[13]; - w_tmp[14] = w[14]; - w_tmp[15] = w[15]; - - u32 state[4]; + const u32x w0r = words_buf_r[il_pos / VECT_SIZE]; + + const u32x w0lr = w0l | w0r; + + u32x w_t[4]; + + w_t[0] = w0lr; + w_t[1] = w[ 1]; + w_t[2] = w[ 2]; + w_t[3] = w[ 3]; + + u32x state[4]; state[0] = 0; state[1] = 0; state[2] = 0; state[3] = 0; - domino_big_md (w_tmp, pw_len, state, s_lotus_magic_table); - - const u32 r0 = state[0]; - const u32 r1 = state[1]; - const u32 r2 = state[2]; - const u32 r3 = state[3]; + domino_big_md (w_t, pw_len, state, s_lotus_magic_table); - #include COMPARE_M + COMPARE_M_SIMD (state[0], state[1], state[2], state[3]); } } -static void m08600s (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 pw_len, __global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset) +void m08600s (__local u32 *s_lotus_magic_table, u32 w[16], const u32 pw_len, __global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset) { /** * modifier @@ -322,22 +291,7 @@ static void m08600s (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 * base */ - if (pw_len < 16) - { - pad (&w[ 0], pw_len & 0xf); - } - else if (pw_len < 32) - { - pad (&w[ 4], pw_len & 0xf); - } - else if (pw_len < 48) - { - pad (&w[ 8], pw_len & 0xf); - } - else if (pw_len < 64) - { - pad (&w[12], pw_len & 0xf); - } + pad (&w[0], pw_len); /** * digest @@ -357,50 +311,33 @@ static void m08600s (__local u32 s_lotus_magic_table[256], u32 w[16], const u32 u32 w0l = w[0]; - for (u32 il_pos = 0; il_pos < bfs_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) { - const u32 w0r = words_buf_r[il_pos]; - - const u32 w0 = w0l | w0r; - - u32 w_tmp[16]; - - w_tmp[ 0] = w0; - w_tmp[ 1] = w[ 1]; - w_tmp[ 2] = w[ 2]; - w_tmp[ 3] = w[ 3]; - w_tmp[ 4] = w[ 4]; - w_tmp[ 5] = w[ 5]; - w_tmp[ 6] = w[ 6]; - w_tmp[ 7] = w[ 7]; - w_tmp[ 8] = w[ 8]; - w_tmp[ 9] = w[ 9]; - w_tmp[10] = w[10]; - w_tmp[11] = w[11]; - w_tmp[12] = w[12]; - w_tmp[13] = w[13]; - w_tmp[14] = w[14]; - w_tmp[15] = w[15]; - - u32 state[4]; + const u32x w0r = words_buf_r[il_pos / VECT_SIZE]; + + const u32x w0lr = w0l | w0r; + + u32x w_t[4]; + + w_t[0] = w0lr; + w_t[1] = w[ 1]; + w_t[2] = w[ 2]; + w_t[3] = w[ 3]; + + u32x state[4]; state[0] = 0; state[1] = 0; state[2] = 0; state[3] = 0; - domino_big_md (w_tmp, pw_len, state, s_lotus_magic_table); + domino_big_md (w_t, pw_len, state, s_lotus_magic_table); - const u32 r0 = state[0]; - const u32 r1 = state[1]; - const u32 r2 = state[2]; - const u32 r3 = state[3]; - - #include COMPARE_S + COMPARE_S_SIMD (state[0], state[1], state[2], state[3]); } } -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m04 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +__kernel void m08600_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) { /** * base @@ -408,6 +345,26 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m04 (__glo const u32 gid = get_global_id (0); const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ + + __local u32 s_lotus_magic_table[256]; + + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } + + barrier (CLK_LOCAL_MEM_FENCE); + + if (gid >= gid_max) return; + + /** + * base + */ u32 w[16]; @@ -431,38 +388,41 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m04 (__glo const u32 pw_len = pws[gid].pw_len; /** - * sbox + * main */ - const u32 lid4 = lid * 4; + m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); +} + +__kernel void m08600_m08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +{ + /** + * base + */ + + const u32 gid = get_global_id (0); + const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ __local u32 s_lotus_magic_table[256]; - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } barrier (CLK_LOCAL_MEM_FENCE); if (gid >= gid_max) return; - /** - * main - */ - - m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); -} - -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m08 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) -{ /** * base */ - const u32 gid = get_global_id (0); - const u32 lid = get_local_id (0); - u32 w[16]; w[ 0] = pws[gid].i[ 0]; @@ -485,38 +445,41 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m08 (__glo const u32 pw_len = pws[gid].pw_len; /** - * sbox + * main + */ + + m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); +} + +__kernel void m08600_m16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +{ + /** + * base */ - const u32 lid4 = lid * 4; + const u32 gid = get_global_id (0); + const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ __local u32 s_lotus_magic_table[256]; - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } barrier (CLK_LOCAL_MEM_FENCE); if (gid >= gid_max) return; - /** - * main - */ - - m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); -} - -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m16 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) -{ /** * base */ - const u32 gid = get_global_id (0); - const u32 lid = get_local_id (0); - u32 w[16]; w[ 0] = pws[gid].i[ 0]; @@ -539,38 +502,41 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_m16 (__glo const u32 pw_len = pws[gid].pw_len; /** - * sbox + * main */ - const u32 lid4 = lid * 4; + m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); +} + +__kernel void m08600_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +{ + /** + * base + */ + + const u32 gid = get_global_id (0); + const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ __local u32 s_lotus_magic_table[256]; - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } barrier (CLK_LOCAL_MEM_FENCE); if (gid >= gid_max) return; - /** - * main - */ - - m08600m (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); -} - -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s04 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) -{ /** * base */ - const u32 gid = get_global_id (0); - const u32 lid = get_local_id (0); - u32 w[16]; w[ 0] = pws[gid].i[ 0]; @@ -593,38 +559,41 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s04 (__glo const u32 pw_len = pws[gid].pw_len; /** - * sbox + * main + */ + + m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); +} + +__kernel void m08600_s08 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +{ + /** + * base */ - const u32 lid4 = lid * 4; + const u32 gid = get_global_id (0); + const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ __local u32 s_lotus_magic_table[256]; - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } barrier (CLK_LOCAL_MEM_FENCE); if (gid >= gid_max) return; - /** - * main - */ - - m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); -} - -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s08 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) -{ /** * base */ - const u32 gid = get_global_id (0); - const u32 lid = get_local_id (0); - u32 w[16]; w[ 0] = pws[gid].i[ 0]; @@ -647,38 +616,41 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s08 (__glo const u32 pw_len = pws[gid].pw_len; /** - * sbox + * main */ - const u32 lid4 = lid * 4; + m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); +} + +__kernel void m08600_s16 (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32x * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) +{ + /** + * base + */ + + const u32 gid = get_global_id (0); + const u32 lid = get_local_id (0); + const u32 lsz = get_local_size (0); + + /** + * sbox + */ __local u32 s_lotus_magic_table[256]; - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; + for (u32 i = lid; i < 256; i += lsz) + { + s_lotus_magic_table[i] = lotus_magic_table[i]; + } barrier (CLK_LOCAL_MEM_FENCE); if (gid >= gid_max) return; - /** - * main - */ - - m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); -} - -__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s16 (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __constant u32 * words_buf_r, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 bfs_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max) -{ /** * base */ - const u32 gid = get_global_id (0); - const u32 lid = get_local_id (0); - u32 w[16]; w[ 0] = pws[gid].i[ 0]; @@ -700,26 +672,9 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m08600_s16 (__glo const u32 pw_len = pws[gid].pw_len; - /** - * sbox - */ - - const u32 lid4 = lid * 4; - - __local u32 s_lotus_magic_table[256]; - - s_lotus_magic_table[lid4 + 0] = lotus_magic_table[lid4 + 0]; - s_lotus_magic_table[lid4 + 1] = lotus_magic_table[lid4 + 1]; - s_lotus_magic_table[lid4 + 2] = lotus_magic_table[lid4 + 2]; - s_lotus_magic_table[lid4 + 3] = lotus_magic_table[lid4 + 3]; - - barrier (CLK_LOCAL_MEM_FENCE); - - if (gid >= gid_max) return; - /** * main */ - m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, bfs_cnt, digests_cnt, digests_offset); + m08600s (s_lotus_magic_table, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_scryptV0_buf, d_scryptV1_buf, d_scryptV2_buf, d_scryptV3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, salt_pos, loop_pos, loop_cnt, il_cnt, digests_cnt, digests_offset); }