From ae33777e73d34b1c19e32ca347868d4dbbd317c3 Mon Sep 17 00:00:00 2001 From: Jens Steube Date: Sat, 6 Feb 2016 11:17:00 +0100 Subject: [PATCH] Converted to new SIMD: -m 8600 -a 0 --- OpenCL/m08600_a0.cl | 160 ++++++++++++++++---------------------------- 1 file changed, 56 insertions(+), 104 deletions(-) diff --git a/OpenCL/m08600_a0.cl b/OpenCL/m08600_a0.cl index bae3f4c86..df78e9dca 100644 --- a/OpenCL/m08600_a0.cl +++ b/OpenCL/m08600_a0.cl @@ -7,6 +7,8 @@ #define _LOTUS5_ +#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -20,9 +22,7 @@ #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 lotus_magic_table[256] = { @@ -60,11 +60,19 @@ __constant u32 lotus_magic_table[256] = 0x29, 0x39, 0xb9, 0xe9, 0x4c, 0xff, 0x43, 0xab, }; -#define BOX(S,i) (S)[(i)] +#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]) +#endif -static void lotus_mix (u32 *in, __local u32 *s_lotus_magic_table) +static void lotus_mix (u32x *in, __local u32 *s_lotus_magic_table) { - u32 p = 0; + u32x p = 0; for (int i = 0; i < 18; i++) { @@ -73,49 +81,49 @@ static void lotus_mix (u32 *in, __local u32 *s_lotus_magic_table) #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) +static 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 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) +static void pad (u32x w[4], const u32 len) { const u32 val = 16 - len; - const u32 mask1 = val << 24; + const u32x mask1 = val << 24; - const u32 mask2 = val << 16 + const u32x mask2 = val << 16 | val << 24; - const u32 mask3 = val << 8 + const u32x mask3 = val << 8 | val << 16 | val << 24; - const u32 mask4 = val << 0 + const u32x mask4 = val << 0 | val << 8 | val << 16 | val << 24; @@ -181,9 +189,9 @@ 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) +static 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]; @@ -206,23 +214,23 @@ static void mdtransform_norecalc (u32 state[4], u32 block[4], __local u32 *s_lot state[3] = x[3]; } -static void mdtransform (u32 state[4], u32 checksum[4], u32 block[4], __local u32 *s_lotus_magic_table) +static 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) +static void domino_big_md (const u32x saved_key[16], const u32x 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]; + u32x block[4]; block[0] = saved_key[0]; block[1] = saved_key[1]; @@ -283,39 +291,16 @@ __kernel void m08600_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, * loop */ - for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < rules_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]; + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); - 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; - - 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); - - u32 w[16]; + u32x w[16]; w[ 0] = w0[0]; w[ 1] = w0[1]; @@ -334,7 +319,7 @@ __kernel void m08600_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, w[14] = 0; w[15] = 0; - u32 state[4]; + u32x state[4]; state[0] = 0; state[1] = 0; @@ -364,12 +349,7 @@ __kernel void m08600_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, domino_big_md (w, out_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_M + COMPARE_M_SIMD (state[0], state[1], state[2], state[3]); } } @@ -442,39 +422,16 @@ __kernel void m08600_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, * loop */ - for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < rules_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]; + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); - 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; - - 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); - - u32 w[16]; + u32x w[16]; w[ 0] = w0[0]; w[ 1] = w0[1]; @@ -493,7 +450,7 @@ __kernel void m08600_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, w[14] = 0; w[15] = 0; - u32 state[4]; + u32x state[4]; state[0] = 0; state[1] = 0; @@ -523,12 +480,7 @@ __kernel void m08600_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf, domino_big_md (w, out_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]); } }