From b1a88da83e2c8afb628a19c6ad7fd815ec0629fb Mon Sep 17 00:00:00 2001 From: jsteube Date: Sun, 9 Jul 2017 18:20:28 +0200 Subject: [PATCH] Vectorized WinZip and added support for long passwords --- OpenCL/m13600.cl | 175 +++++++++++++++++++---------------------------- 1 file changed, 72 insertions(+), 103 deletions(-) diff --git a/OpenCL/m13600.cl b/OpenCL/m13600.cl index 5844b3e5c..1d2898fe0 100644 --- a/OpenCL/m13600.cl +++ b/OpenCL/m13600.cl @@ -117,7 +117,24 @@ __kernel void m13600_init (__global pw_t *pws, __global const kernel_rule_t *rul tmps[gid].opad[3] = sha1_hmac_ctx.opad.h[3]; tmps[gid].opad[4] = sha1_hmac_ctx.opad.h[4]; - sha1_hmac_update_global_swap (&sha1_hmac_ctx, esalt_bufs[digests_offset].salt_buf, esalt_bufs[digests_offset].salt_len); + w0[0] = swap32_S (esalt_bufs[digests_offset].salt_buf[0]); + w0[1] = swap32_S (esalt_bufs[digests_offset].salt_buf[1]); + w0[2] = swap32_S (esalt_bufs[digests_offset].salt_buf[2]); + w0[3] = swap32_S (esalt_bufs[digests_offset].salt_buf[3]); + w1[0] = 0; + w1[1] = 0; + w1[2] = 0; + w1[3] = 0; + w2[0] = 0; + w2[1] = 0; + w2[2] = 0; + w2[3] = 0; + w3[0] = 0; + w3[1] = 0; + w3[2] = 0; + w3[3] = 0; + + sha1_hmac_update_64 (&sha1_hmac_ctx, w0, w1, w2, w3, esalt_bufs[digests_offset].salt_len); const u32 mode = esalt_bufs[digests_offset].mode; @@ -143,8 +160,6 @@ __kernel void m13600_init (__global pw_t *pws, __global const kernel_rule_t *rul for (u32 i = iter_start, j = count_start; i < iter_stop; i++, j++) { - const u32 i5 = i * 5; - sha1_hmac_ctx_t sha1_hmac_ctx2 = sha1_hmac_ctx; w0[0] = j; @@ -168,6 +183,8 @@ __kernel void m13600_init (__global pw_t *pws, __global const kernel_rule_t *rul sha1_hmac_final (&sha1_hmac_ctx2); + const u32 i5 = i * 5; + tmps[gid].dgst[i5 + 0] = sha1_hmac_ctx2.opad.h[0]; tmps[gid].dgst[i5 + 1] = sha1_hmac_ctx2.opad.h[1]; tmps[gid].dgst[i5 + 2] = sha1_hmac_ctx2.opad.h[2]; @@ -186,23 +203,22 @@ __kernel void m13600_loop (__global pw_t *pws, __global const kernel_rule_t *rul { const u32 gid = get_global_id (0); - if (gid >= gid_max) return; + if ((gid * VECT_SIZE) >= gid_max) return; - u32 ipad[5]; + u32x ipad[5]; + u32x opad[5]; - ipad[0] = tmps[gid].ipad[0]; - ipad[1] = tmps[gid].ipad[1]; - ipad[2] = tmps[gid].ipad[2]; - ipad[3] = tmps[gid].ipad[3]; - ipad[4] = tmps[gid].ipad[4]; + ipad[0] = packv (tmps, ipad, gid, 0); + ipad[1] = packv (tmps, ipad, gid, 1); + ipad[2] = packv (tmps, ipad, gid, 2); + ipad[3] = packv (tmps, ipad, gid, 3); + ipad[4] = packv (tmps, ipad, gid, 4); - u32 opad[5]; - - opad[0] = tmps[gid].opad[0]; - opad[1] = tmps[gid].opad[1]; - opad[2] = tmps[gid].opad[2]; - opad[3] = tmps[gid].opad[3]; - opad[4] = tmps[gid].opad[4]; + opad[0] = packv (tmps, opad, gid, 0); + opad[1] = packv (tmps, opad, gid, 1); + opad[2] = packv (tmps, opad, gid, 2); + opad[3] = packv (tmps, opad, gid, 3); + opad[4] = packv (tmps, opad, gid, 4); const u32 mode = esalt_bufs[digests_offset].mode; @@ -230,28 +246,27 @@ __kernel void m13600_loop (__global pw_t *pws, __global const kernel_rule_t *rul { const u32 i5 = i * 5; - u32 dgst[5]; + u32x dgst[5]; + u32x out[5]; - dgst[0] = tmps[gid].dgst[i5 + 0]; - dgst[1] = tmps[gid].dgst[i5 + 1]; - dgst[2] = tmps[gid].dgst[i5 + 2]; - dgst[3] = tmps[gid].dgst[i5 + 3]; - dgst[4] = tmps[gid].dgst[i5 + 4]; + dgst[0] = packv (tmps, dgst, gid, i5 + 0); + dgst[1] = packv (tmps, dgst, gid, i5 + 1); + dgst[2] = packv (tmps, dgst, gid, i5 + 2); + dgst[3] = packv (tmps, dgst, gid, i5 + 3); + dgst[4] = packv (tmps, dgst, gid, i5 + 4); - u32 out[5]; + out[0] = packv (tmps, out, gid, i5 + 0); + out[1] = packv (tmps, out, gid, i5 + 1); + out[2] = packv (tmps, out, gid, i5 + 2); + out[3] = packv (tmps, out, gid, i5 + 3); + out[4] = packv (tmps, out, gid, i5 + 4); - out[0] = tmps[gid].out[i5 + 0]; - out[1] = tmps[gid].out[i5 + 1]; - out[2] = tmps[gid].out[i5 + 2]; - out[3] = tmps[gid].out[i5 + 3]; - out[4] = tmps[gid].out[i5 + 4]; - - for (u32 k = 0; k < loop_cnt; k++) + for (u32 j = 0; j < loop_cnt; j++) { - u32 w0[4]; - u32 w1[4]; - u32 w2[4]; - u32 w3[4]; + u32x w0[4]; + u32x w1[4]; + u32x w2[4]; + u32x w3[4]; w0[0] = dgst[0]; w0[1] = dgst[1]; @@ -270,7 +285,7 @@ __kernel void m13600_loop (__global pw_t *pws, __global const kernel_rule_t *rul w3[2] = 0; w3[3] = (64 + 20) * 8; - hmac_sha1_run (w0, w1, w2, w3, ipad, opad, dgst); + hmac_sha1_run_V (w0, w1, w2, w3, ipad, opad, dgst); out[0] ^= dgst[0]; out[1] ^= dgst[1]; @@ -279,17 +294,17 @@ __kernel void m13600_loop (__global pw_t *pws, __global const kernel_rule_t *rul out[4] ^= dgst[4]; } - tmps[gid].dgst[i5 + 0] = dgst[0]; - tmps[gid].dgst[i5 + 1] = dgst[1]; - tmps[gid].dgst[i5 + 2] = dgst[2]; - tmps[gid].dgst[i5 + 3] = dgst[3]; - tmps[gid].dgst[i5 + 4] = dgst[4]; + unpackv (tmps, dgst, gid, i5 + 0, dgst[0]); + unpackv (tmps, dgst, gid, i5 + 1, dgst[1]); + unpackv (tmps, dgst, gid, i5 + 2, dgst[2]); + unpackv (tmps, dgst, gid, i5 + 3, dgst[3]); + unpackv (tmps, dgst, gid, i5 + 4, dgst[4]); - tmps[gid].out[i5 + 0] = out[0]; - tmps[gid].out[i5 + 1] = out[1]; - tmps[gid].out[i5 + 2] = out[2]; - tmps[gid].out[i5 + 3] = out[3]; - tmps[gid].out[i5 + 4] = out[4]; + unpackv (tmps, out, gid, i5 + 0, out[0]); + unpackv (tmps, out, gid, i5 + 1, out[1]); + unpackv (tmps, out, gid, i5 + 2, out[2]); + unpackv (tmps, out, gid, i5 + 3, out[3]); + unpackv (tmps, out, gid, i5 + 4, out[4]); } } @@ -305,8 +320,6 @@ __kernel void m13600_comp (__global pw_t *pws, __global const kernel_rule_t *rul const u32 lid = get_local_id (0); - u32 key[8] = { 0 }; - const u32 mode = esalt_bufs[digests_offset].mode; u32 iter_start; @@ -325,8 +338,9 @@ __kernel void m13600_comp (__global pw_t *pws, __global const kernel_rule_t *rul break; } - u32 i, j; - for (i = iter_start, j = 0; i < iter_stop; i++, j++) + u32 key[8] = { 0 }; + + for (int i = iter_start, j = 0; i < iter_stop; i++, j++) { key[j] = tmps[gid].out[i]; } @@ -353,63 +367,18 @@ __kernel void m13600_comp (__global pw_t *pws, __global const kernel_rule_t *rul w3[2] = 0; w3[3] = 0; - u32 ipad[5]; - u32 opad[5]; + sha1_hmac_ctx_t ctx; - hmac_sha1_pad (w0, w1, w2, w3, ipad, opad); + sha1_hmac_init (&ctx, w0, w1, w2, w3); - int data_len = esalt_bufs[digests_offset].data_len; + sha1_hmac_update_global_swap (&ctx, esalt_bufs[digests_offset].data_buf, esalt_bufs[digests_offset].data_len); - int data_left; - int data_off; + sha1_hmac_final (&ctx); - for (data_left = data_len, data_off = 0; data_left >= 56; data_left -= 64, data_off += 16) - { - w0[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 0]); - w0[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 1]); - w0[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 2]); - w0[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 3]); - w1[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 4]); - w1[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 5]); - w1[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 6]); - w1[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 7]); - w2[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 8]); - w2[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 9]); - w2[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 10]); - w2[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 11]); - w3[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 12]); - w3[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 13]); - w3[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 14]); - w3[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 15]); - - sha1_transform (w0, w1, w2, w3, ipad); - } - - w0[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 0]); - w0[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 1]); - w0[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 2]); - w0[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 3]); - w1[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 4]); - w1[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 5]); - w1[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 6]); - w1[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 7]); - w2[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 8]); - w2[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 9]); - w2[2] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 10]); - w2[3] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 11]); - w3[0] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 12]); - w3[1] = swap32 (esalt_bufs[digests_offset].data_buf[data_off + 13]); - w3[2] = 0; - w3[3] = (64 + data_len) * 8; - - u32 digest[5]; - - hmac_sha1_run (w0, w1, w2, w3, ipad, opad, digest); - - const u32 r0 = swap32 (digest[0] & 0xffffffff); - const u32 r1 = swap32 (digest[1] & 0xffffffff); - const u32 r2 = swap32 (digest[2] & 0xffff0000); - const u32 r3 = swap32 (digest[3] & 0x00000000); + const u32 r0 = swap32_S (ctx.opad.h[0] & 0xffffffff); + const u32 r1 = swap32_S (ctx.opad.h[1] & 0xffffffff); + const u32 r2 = swap32_S (ctx.opad.h[2] & 0xffff0000); + const u32 r3 = swap32_S (ctx.opad.h[3] & 0x00000000); #define il_pos 0