diff --git a/OpenCL/m00050_a0.cl b/OpenCL/m00050_a0.cl index 8c7be295b..575b1b4dc 100644 --- a/OpenCL/m00050_a0.cl +++ b/OpenCL/m00050_a0.cl @@ -5,6 +5,8 @@ #define _MD5_ +#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -18,33 +20,31 @@ #include "OpenCL/common.c" #include "include/rp_kernel.h" #include "OpenCL/rp.c" +#include "OpenCL/simd.c" -#define COMPARE_S "OpenCL/check_single_comp4.c" -#define COMPARE_M "OpenCL/check_multi_comp4.c" - -static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], const u32 w3[4], u32 digest[4]) +static void md5_transform (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u32x digest[4]) { - u32 a = digest[0]; - u32 b = digest[1]; - u32 c = digest[2]; - u32 d = digest[3]; + u32x a = digest[0]; + u32x b = digest[1]; + u32x c = digest[2]; + u32x d = digest[3]; - u32 w0_t = w0[0]; - u32 w1_t = w0[1]; - u32 w2_t = w0[2]; - u32 w3_t = w0[3]; - u32 w4_t = w1[0]; - u32 w5_t = w1[1]; - u32 w6_t = w1[2]; - u32 w7_t = w1[3]; - u32 w8_t = w2[0]; - u32 w9_t = w2[1]; - u32 wa_t = w2[2]; - u32 wb_t = w2[3]; - u32 wc_t = w3[0]; - u32 wd_t = w3[1]; - u32 we_t = w3[2]; - u32 wf_t = w3[3]; + u32x w0_t = w0[0]; + u32x w1_t = w0[1]; + u32x w2_t = w0[2]; + u32x w3_t = w0[3]; + u32x w4_t = w1[0]; + u32x w5_t = w1[1]; + u32x w6_t = w1[2]; + u32x w7_t = w1[3]; + u32x w8_t = w2[0]; + u32x w9_t = w2[1]; + u32x wa_t = w2[2]; + u32x wb_t = w2[3]; + u32x wc_t = w3[0]; + u32x wd_t = w3[1]; + u32x we_t = w3[2]; + u32x wf_t = w3[3]; MD5_STEP (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00); MD5_STEP (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01); @@ -120,7 +120,7 @@ static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], co digest[3] += d; } -static void hmac_md5_pad (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[4], u32 opad[4]) +static void hmac_md5_pad (u32x w0[4], u32x w1[4], u32x w2[4], u32x w3[4], u32x ipad[4], u32x opad[4]) { w0[0] = w0[0] ^ 0x36363636; w0[1] = w0[1] ^ 0x36363636; @@ -171,7 +171,7 @@ static void hmac_md5_pad (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[4 md5_transform (w0, w1, w2, w3, opad); } -static void hmac_md5_run (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[4], u32 opad[4], u32 digest[4]) +static void hmac_md5_run (u32x w0[4], u32x w1[4], u32x w2[4], u32x w3[4], u32x ipad[4], u32x opad[4], u32x digest[4]) { digest[0] = ipad[0]; digest[1] = ipad[1]; @@ -275,72 +275,49 @@ __kernel void m00050_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * 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]; - - 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); + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); /** * pads */ - u32 w0_t[4]; + u32x w0_t[4]; w0_t[0] = w0[0]; w0_t[1] = w0[1]; w0_t[2] = w0[2]; w0_t[3] = w0[3]; - u32 w1_t[4]; + u32x w1_t[4]; w1_t[0] = w1[0]; w1_t[1] = w1[1]; w1_t[2] = w1[2]; w1_t[3] = w1[3]; - u32 w2_t[4]; + u32x w2_t[4]; w2_t[0] = 0; w2_t[1] = 0; w2_t[2] = 0; w2_t[3] = 0; - u32 w3_t[4]; + u32x w3_t[4]; w3_t[0] = 0; w3_t[1] = 0; w3_t[2] = 0; w3_t[3] = 0; - u32 ipad[4]; - u32 opad[4]; + u32x ipad[4]; + u32x opad[4]; hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad); @@ -361,16 +338,11 @@ __kernel void m00050_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu w3_t[2] = (64 + salt_len) * 8; w3_t[3] = 0; - u32 digest[4]; + u32x digest[4]; hmac_md5_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest); - const u32 r0 = digest[0]; - const u32 r1 = digest[3]; - const u32 r2 = digest[2]; - const u32 r3 = digest[1]; - - #include COMPARE_M + COMPARE_M_SIMD (digest[0], digest[3], digest[2], digest[1]); } } @@ -464,72 +436,49 @@ __kernel void m00050_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * 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]; - - 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); + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); /** * pads */ - u32 w0_t[4]; + u32x w0_t[4]; w0_t[0] = w0[0]; w0_t[1] = w0[1]; w0_t[2] = w0[2]; w0_t[3] = w0[3]; - u32 w1_t[4]; + u32x w1_t[4]; w1_t[0] = w1[0]; w1_t[1] = w1[1]; w1_t[2] = w1[2]; w1_t[3] = w1[3]; - u32 w2_t[4]; + u32x w2_t[4]; w2_t[0] = 0; w2_t[1] = 0; w2_t[2] = 0; w2_t[3] = 0; - u32 w3_t[4]; + u32x w3_t[4]; w3_t[0] = 0; w3_t[1] = 0; w3_t[2] = 0; w3_t[3] = 0; - u32 ipad[4]; - u32 opad[4]; + u32x ipad[4]; + u32x opad[4]; hmac_md5_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad); @@ -550,16 +499,11 @@ __kernel void m00050_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu w3_t[2] = (64 + salt_len) * 8; w3_t[3] = 0; - u32 digest[4]; + u32x digest[4]; hmac_md5_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest); - const u32 r0 = digest[0]; - const u32 r1 = digest[3]; - const u32 r2 = digest[2]; - const u32 r3 = digest[1]; - - #include COMPARE_S + COMPARE_S_SIMD (digest[0], digest[3], digest[2], digest[1]); } }