diff --git a/OpenCL/m10410_a0.cl b/OpenCL/m10410_a0.cl index 041bd0cd2..845834048 100644 --- a/OpenCL/m10410_a0.cl +++ b/OpenCL/m10410_a0.cl @@ -7,6 +7,9 @@ #define _MD5_ +//too much register pressure +//#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -20,9 +23,7 @@ #include "OpenCL/common.c" #include "include/rp_kernel.h" #include "OpenCL/rp.c" - -#define COMPARE_S "OpenCL/check_single_comp4.c" -#define COMPARE_M "OpenCL/check_multi_comp4.c" +#include "OpenCL/simd.c" __constant u32 padding[8] = { @@ -89,6 +90,7 @@ static void rc4_init_16 (__local RC4_KEY *rc4_key, const u32 data[4]) static u8 rc4_next_16 (__local RC4_KEY *rc4_key, u8 i, u8 j, __constant u32 *in, u32 out[4]) { + #pragma unroll 4 for (u32 k = 0; k < 4; k++) { u32 xor4 = 0; @@ -154,14 +156,12 @@ __kernel void m10410_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, if (gid >= gid_max) return; u32 pw_buf0[4]; + u32 pw_buf1[4]; pw_buf0[0] = pws[gid].i[ 0]; pw_buf0[1] = pws[gid].i[ 1]; pw_buf0[2] = pws[gid].i[ 2]; pw_buf0[3] = pws[gid].i[ 3]; - - u32 pw_buf1[4]; - pw_buf1[0] = pws[gid].i[ 4]; pw_buf1[1] = pws[gid].i[ 5]; pw_buf1[2] = pws[gid].i[ 6]; @@ -170,7 +170,7 @@ __kernel void m10410_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, const u32 pw_len = pws[gid].pw_len; /** - * key + * shared */ __local RC4_KEY rc4_keys[64]; @@ -180,59 +180,26 @@ __kernel void m10410_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, * loop */ - for (u32 il_pos = 0; il_pos < il_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) { - u32 w0[4]; - - w0[0] = pw_buf0[0]; - w0[1] = pw_buf0[1]; - w0[2] = pw_buf0[2]; - w0[3] = pw_buf0[3]; - - u32 w1[4]; - - w1[0] = pw_buf1[0]; - w1[1] = pw_buf1[1]; - w1[2] = pw_buf1[2]; - w1[3] = pw_buf1[3]; - - u32 w2[4]; - - w2[0] = 0; - w2[1] = 0; - w2[2] = 0; - w2[3] = 0; + u32x w0[4] = { 0 }; + u32x w1[4] = { 0 }; + u32x w2[4] = { 0 }; + u32x w3[4] = { 0 }; - u32 w3[4]; + apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); - w3[0] = 0; - w3[1] = 0; - w3[2] = 0; - w3[3] = 0; + /** + * pdf + */ - const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); - - // now the RC4 part - - u32 key[4]; - - key[0] = w0[0]; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; - - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4]; rc4_next_16 (rc4_key, 0, 0, padding, out); - const u32 r0 = out[0]; - const u32 r1 = out[1]; - const u32 r2 = out[2]; - const u32 r3 = out[3]; - - #include COMPARE_M + COMPARE_M_SIMD (out[0], out[1], out[2], out[3]); } } @@ -261,14 +228,12 @@ __kernel void m10410_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, if (gid >= gid_max) return; u32 pw_buf0[4]; + u32 pw_buf1[4]; pw_buf0[0] = pws[gid].i[ 0]; pw_buf0[1] = pws[gid].i[ 1]; pw_buf0[2] = pws[gid].i[ 2]; pw_buf0[3] = pws[gid].i[ 3]; - - u32 pw_buf1[4]; - pw_buf1[0] = pws[gid].i[ 4]; pw_buf1[1] = pws[gid].i[ 5]; pw_buf1[2] = pws[gid].i[ 6]; @@ -276,6 +241,13 @@ __kernel void m10410_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, const u32 pw_len = pws[gid].pw_len; + /** + * shared + */ + + __local RC4_KEY rc4_keys[64]; + __local RC4_KEY *rc4_key = &rc4_keys[lid]; + /** * digest */ @@ -288,70 +260,30 @@ __kernel void m10410_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, digests_buf[digests_offset].digest_buf[DGST_R3] }; - /** - * key - */ - - __local RC4_KEY rc4_keys[64]; - __local RC4_KEY *rc4_key = &rc4_keys[lid]; - /** * loop */ - for (u32 il_pos = 0; il_pos < il_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) { - u32 w0[4]; + u32x w0[4] = { 0 }; + u32x w1[4] = { 0 }; + u32x w2[4] = { 0 }; + u32x w3[4] = { 0 }; - w0[0] = pw_buf0[0]; - w0[1] = pw_buf0[1]; - w0[2] = pw_buf0[2]; - w0[3] = pw_buf0[3]; + apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); - u32 w1[4]; + /** + * pdf + */ - w1[0] = pw_buf1[0]; - w1[1] = pw_buf1[1]; - w1[2] = pw_buf1[2]; - w1[3] = pw_buf1[3]; - - u32 w2[4]; - - w2[0] = 0; - w2[1] = 0; - w2[2] = 0; - w2[3] = 0; - - u32 w3[4]; - - w3[0] = 0; - w3[1] = 0; - w3[2] = 0; - w3[3] = 0; - - const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); - - // now the RC4 part - - u32 key[4]; - - key[0] = w0[0]; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; - - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4]; rc4_next_16 (rc4_key, 0, 0, padding, out); - const u32 r0 = out[0]; - const u32 r1 = out[1]; - const u32 r2 = out[2]; - const u32 r3 = out[3]; - - #include COMPARE_S + COMPARE_S_SIMD (out[0], out[1], out[2], out[3]); } } diff --git a/OpenCL/m10410_a1.cl b/OpenCL/m10410_a1.cl index ed8ea3899..221519122 100644 --- a/OpenCL/m10410_a1.cl +++ b/OpenCL/m10410_a1.cl @@ -7,6 +7,9 @@ #define _MD5_ +//too much register pressure +//#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -18,9 +21,7 @@ #include "include/kernel_functions.c" #include "OpenCL/types_ocl.c" #include "OpenCL/common.c" - -#define COMPARE_S "OpenCL/check_single_comp4.c" -#define COMPARE_M "OpenCL/check_multi_comp4.c" +#include "OpenCL/simd.c" __constant u32 padding[8] = { @@ -87,6 +88,7 @@ static void rc4_init_16 (__local RC4_KEY *rc4_key, const u32 data[4]) static u8 rc4_next_16 (__local RC4_KEY *rc4_key, u8 i, u8 j, __constant u32 *in, u32 out[4]) { + #pragma unroll 4 for (u32 k = 0; k < 4; k++) { u32 xor4 = 0; @@ -151,113 +153,121 @@ __kernel void m10410_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, if (gid >= gid_max) return; - u32 wordl0[4]; - - wordl0[0] = pws[gid].i[ 0]; - wordl0[1] = pws[gid].i[ 1]; - wordl0[2] = pws[gid].i[ 2]; - wordl0[3] = pws[gid].i[ 3]; - - u32 wordl1[4]; - - wordl1[0] = pws[gid].i[ 4]; - wordl1[1] = pws[gid].i[ 5]; - wordl1[2] = pws[gid].i[ 6]; - wordl1[3] = pws[gid].i[ 7]; - - u32 wordl2[4]; - - wordl2[0] = 0; - wordl2[1] = 0; - wordl2[2] = 0; - wordl2[3] = 0; + u32 pw_buf0[4]; + u32 pw_buf1[4]; - u32 wordl3[4]; - - wordl3[0] = 0; - wordl3[1] = 0; - wordl3[2] = 0; - wordl3[3] = 0; + pw_buf0[0] = pws[gid].i[0]; + pw_buf0[1] = pws[gid].i[1]; + pw_buf0[2] = pws[gid].i[2]; + pw_buf0[3] = pws[gid].i[3]; + pw_buf1[0] = pws[gid].i[4]; + pw_buf1[1] = pws[gid].i[5]; + pw_buf1[2] = pws[gid].i[6]; + pw_buf1[3] = pws[gid].i[7]; const u32 pw_l_len = pws[gid].pw_len; - if (combs_mode == COMBINATOR_MODE_BASE_RIGHT) - { - switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len); - } - /** - * key + * shared */ __local RC4_KEY rc4_keys[64]; + __local RC4_KEY *rc4_key = &rc4_keys[lid]; /** - * loop + * U_buf */ - for (u32 il_pos = 0; il_pos < il_cnt; il_pos++) - { - u32 wordr0[4]; + u32 o_buf[8]; - wordr0[0] = combs_buf[il_pos].i[0]; - wordr0[1] = combs_buf[il_pos].i[1]; - wordr0[2] = combs_buf[il_pos].i[2]; - wordr0[3] = combs_buf[il_pos].i[3]; + o_buf[0] = pdf_bufs[salt_pos].o_buf[0]; + o_buf[1] = pdf_bufs[salt_pos].o_buf[1]; + o_buf[2] = pdf_bufs[salt_pos].o_buf[2]; + o_buf[3] = pdf_bufs[salt_pos].o_buf[3]; + o_buf[4] = pdf_bufs[salt_pos].o_buf[4]; + o_buf[5] = pdf_bufs[salt_pos].o_buf[5]; + o_buf[6] = pdf_bufs[salt_pos].o_buf[6]; + o_buf[7] = pdf_bufs[salt_pos].o_buf[7]; - u32 wordr1[4]; + u32 P = pdf_bufs[salt_pos].P; - wordr1[0] = combs_buf[il_pos].i[4]; - wordr1[1] = combs_buf[il_pos].i[5]; - wordr1[2] = combs_buf[il_pos].i[6]; - wordr1[3] = combs_buf[il_pos].i[7]; + u32 id_buf[4]; - u32 wordr2[4]; + id_buf[0] = pdf_bufs[salt_pos].id_buf[0]; + id_buf[1] = pdf_bufs[salt_pos].id_buf[1]; + id_buf[2] = pdf_bufs[salt_pos].id_buf[2]; + id_buf[3] = pdf_bufs[salt_pos].id_buf[3]; - wordr2[0] = 0; - wordr2[1] = 0; - wordr2[2] = 0; - wordr2[3] = 0; - - u32 wordr3[4]; + /** + * loop + */ - wordr3[0] = 0; - wordr3[1] = 0; - wordr3[2] = 0; - wordr3[3] = 0; + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) + { + const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos); + + const u32x pw_len = pw_l_len + pw_r_len; + + /** + * concat password candidate + */ + + u32x wordl0[4] = { 0 }; + u32x wordl1[4] = { 0 }; + u32x wordl2[4] = { 0 }; + u32x wordl3[4] = { 0 }; + + wordl0[0] = pw_buf0[0]; + wordl0[1] = pw_buf0[1]; + wordl0[2] = pw_buf0[2]; + wordl0[3] = pw_buf0[3]; + wordl1[0] = pw_buf1[0]; + wordl1[1] = pw_buf1[1]; + wordl1[2] = pw_buf1[2]; + wordl1[3] = pw_buf1[3]; + + u32x wordr0[4] = { 0 }; + u32x wordr1[4] = { 0 }; + u32x wordr2[4] = { 0 }; + u32x wordr3[4] = { 0 }; + + wordr0[0] = ix_create_combt (combs_buf, il_pos, 0); + wordr0[1] = ix_create_combt (combs_buf, il_pos, 1); + wordr0[2] = ix_create_combt (combs_buf, il_pos, 2); + wordr0[3] = ix_create_combt (combs_buf, il_pos, 3); + wordr1[0] = ix_create_combt (combs_buf, il_pos, 4); + wordr1[1] = ix_create_combt (combs_buf, il_pos, 5); + wordr1[2] = ix_create_combt (combs_buf, il_pos, 6); + wordr1[3] = ix_create_combt (combs_buf, il_pos, 7); if (combs_mode == COMBINATOR_MODE_BASE_LEFT) { - switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len); + switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len); + } + else + { + switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len); } - u32 w0[2]; + u32x w0[4]; w0[0] = wordl0[0] | wordr0[0]; w0[1] = wordl0[1] | wordr0[1]; + w0[2] = 0; + w0[3] = 0; - // now the RC4 part + /** + * pdf + */ - u32 key[4]; - - key[0] = w0[0]; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; - - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4]; rc4_next_16 (rc4_key, 0, 0, padding, out); - const u32 r0 = out[0]; - const u32 r1 = out[1]; - const u32 r2 = out[2]; - const u32 r3 = out[3]; - - #include COMPARE_M + COMPARE_M_SIMD (out[0], out[1], out[2], out[3]); } } @@ -285,40 +295,51 @@ __kernel void m10410_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, if (gid >= gid_max) return; - u32 wordl0[4]; + u32 pw_buf0[4]; + u32 pw_buf1[4]; + + pw_buf0[0] = pws[gid].i[0]; + pw_buf0[1] = pws[gid].i[1]; + pw_buf0[2] = pws[gid].i[2]; + pw_buf0[3] = pws[gid].i[3]; + pw_buf1[0] = pws[gid].i[4]; + pw_buf1[1] = pws[gid].i[5]; + pw_buf1[2] = pws[gid].i[6]; + pw_buf1[3] = pws[gid].i[7]; + + const u32 pw_l_len = pws[gid].pw_len; - wordl0[0] = pws[gid].i[ 0]; - wordl0[1] = pws[gid].i[ 1]; - wordl0[2] = pws[gid].i[ 2]; - wordl0[3] = pws[gid].i[ 3]; + /** + * shared + */ - u32 wordl1[4]; + __local RC4_KEY rc4_keys[64]; - wordl1[0] = pws[gid].i[ 4]; - wordl1[1] = pws[gid].i[ 5]; - wordl1[2] = pws[gid].i[ 6]; - wordl1[3] = pws[gid].i[ 7]; + __local RC4_KEY *rc4_key = &rc4_keys[lid]; - u32 wordl2[4]; + /** + * U_buf + */ - wordl2[0] = 0; - wordl2[1] = 0; - wordl2[2] = 0; - wordl2[3] = 0; + u32 o_buf[8]; - u32 wordl3[4]; + o_buf[0] = pdf_bufs[salt_pos].o_buf[0]; + o_buf[1] = pdf_bufs[salt_pos].o_buf[1]; + o_buf[2] = pdf_bufs[salt_pos].o_buf[2]; + o_buf[3] = pdf_bufs[salt_pos].o_buf[3]; + o_buf[4] = pdf_bufs[salt_pos].o_buf[4]; + o_buf[5] = pdf_bufs[salt_pos].o_buf[5]; + o_buf[6] = pdf_bufs[salt_pos].o_buf[6]; + o_buf[7] = pdf_bufs[salt_pos].o_buf[7]; - wordl3[0] = 0; - wordl3[1] = 0; - wordl3[2] = 0; - wordl3[3] = 0; + u32 P = pdf_bufs[salt_pos].P; - const u32 pw_l_len = pws[gid].pw_len; + u32 id_buf[4]; - if (combs_mode == COMBINATOR_MODE_BASE_RIGHT) - { - switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len); - } + id_buf[0] = pdf_bufs[salt_pos].id_buf[0]; + id_buf[1] = pdf_bufs[salt_pos].id_buf[1]; + id_buf[2] = pdf_bufs[salt_pos].id_buf[2]; + id_buf[3] = pdf_bufs[salt_pos].id_buf[3]; /** * digest @@ -332,78 +353,75 @@ __kernel void m10410_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, digests_buf[digests_offset].digest_buf[DGST_R3] }; - /** - * key - */ - - __local RC4_KEY rc4_keys[64]; - __local RC4_KEY *rc4_key = &rc4_keys[lid]; - /** * loop */ - for (u32 il_pos = 0; il_pos < il_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE) { - u32 wordr0[4]; - - wordr0[0] = combs_buf[il_pos].i[0]; - wordr0[1] = combs_buf[il_pos].i[1]; - wordr0[2] = combs_buf[il_pos].i[2]; - wordr0[3] = combs_buf[il_pos].i[3]; - - u32 wordr1[4]; - - wordr1[0] = combs_buf[il_pos].i[4]; - wordr1[1] = combs_buf[il_pos].i[5]; - wordr1[2] = combs_buf[il_pos].i[6]; - wordr1[3] = combs_buf[il_pos].i[7]; - - u32 wordr2[4]; - - wordr2[0] = 0; - wordr2[1] = 0; - wordr2[2] = 0; - wordr2[3] = 0; - - u32 wordr3[4]; - - wordr3[0] = 0; - wordr3[1] = 0; - wordr3[2] = 0; - wordr3[3] = 0; + const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos); + + const u32x pw_len = pw_l_len + pw_r_len; + + /** + * concat password candidate + */ + + u32x wordl0[4] = { 0 }; + u32x wordl1[4] = { 0 }; + u32x wordl2[4] = { 0 }; + u32x wordl3[4] = { 0 }; + + wordl0[0] = pw_buf0[0]; + wordl0[1] = pw_buf0[1]; + wordl0[2] = pw_buf0[2]; + wordl0[3] = pw_buf0[3]; + wordl1[0] = pw_buf1[0]; + wordl1[1] = pw_buf1[1]; + wordl1[2] = pw_buf1[2]; + wordl1[3] = pw_buf1[3]; + + u32x wordr0[4] = { 0 }; + u32x wordr1[4] = { 0 }; + u32x wordr2[4] = { 0 }; + u32x wordr3[4] = { 0 }; + + wordr0[0] = ix_create_combt (combs_buf, il_pos, 0); + wordr0[1] = ix_create_combt (combs_buf, il_pos, 1); + wordr0[2] = ix_create_combt (combs_buf, il_pos, 2); + wordr0[3] = ix_create_combt (combs_buf, il_pos, 3); + wordr1[0] = ix_create_combt (combs_buf, il_pos, 4); + wordr1[1] = ix_create_combt (combs_buf, il_pos, 5); + wordr1[2] = ix_create_combt (combs_buf, il_pos, 6); + wordr1[3] = ix_create_combt (combs_buf, il_pos, 7); if (combs_mode == COMBINATOR_MODE_BASE_LEFT) { - switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len); + switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len); + } + else + { + switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len); } - u32 w0[2]; + u32x w0[4]; w0[0] = wordl0[0] | wordr0[0]; w0[1] = wordl0[1] | wordr0[1]; + w0[2] = 0; + w0[3] = 0; - // now the RC4 part + /** + * pdf + */ - u32 key[4]; - - key[0] = w0[0]; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; - - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4]; rc4_next_16 (rc4_key, 0, 0, padding, out); - const u32 r0 = out[0]; - const u32 r1 = out[1]; - const u32 r2 = out[2]; - const u32 r3 = out[3]; - - #include COMPARE_S + COMPARE_S_SIMD (out[0], out[1], out[2], out[3]); } } diff --git a/OpenCL/m10410_a3.cl b/OpenCL/m10410_a3.cl index 990987ec3..459d3f4b4 100644 --- a/OpenCL/m10410_a3.cl +++ b/OpenCL/m10410_a3.cl @@ -7,6 +7,9 @@ #define _MD5_ +//too much register pressure +//#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -85,6 +88,7 @@ static void rc4_init_16 (__local RC4_KEY *rc4_key, const u32 data[4]) static u8 rc4_next_16 (__local RC4_KEY *rc4_key, u8 i, u8 j, __constant u32 *in, u32 out[4]) { + #pragma unroll 4 for (u32 k = 0; k < 4; k++) { u32 xor4 = 0; @@ -142,6 +146,10 @@ static void m10410m (__local RC4_KEY *rc4_keys, u32 w0[4], u32 w1[4], u32 w2[4], const u32 gid = get_global_id (0); const u32 lid = get_local_id (0); + /** + * shared + */ + __local RC4_KEY *rc4_key = &rc4_keys[lid]; /** @@ -156,16 +164,13 @@ static void m10410m (__local RC4_KEY *rc4_keys, u32 w0[4], u32 w1[4], u32 w2[4], const u32 w0lr = w0l | w0r; - // now the RC4 part - - u32 key[4]; + w0[0] = w0lr; - key[0] = w0lr; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; + /** + * pdf + */ - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4]; @@ -184,6 +189,10 @@ static void m10410s (__local RC4_KEY *rc4_keys, u32 w0[4], u32 w1[4], u32 w2[4], const u32 gid = get_global_id (0); const u32 lid = get_local_id (0); + /** + * shared + */ + __local RC4_KEY *rc4_key = &rc4_keys[lid]; /** @@ -210,16 +219,13 @@ static void m10410s (__local RC4_KEY *rc4_keys, u32 w0[4], u32 w1[4], u32 w2[4], const u32 w0lr = w0l | w0r; - // now the RC4 part - - u32 key[4]; + w0[0] = w0lr; - key[0] = w0lr; - key[1] = w0[1]; - key[2] = 0; - key[3] = 0; + /** + * pdf + */ - rc4_init_16 (rc4_key, key); + rc4_init_16 (rc4_key, w0); u32 out[4];