diff --git a/OpenCL/m01500_a0.cl b/OpenCL/m01500_a0.cl index 097bf92a9..9f061b996 100644 --- a/OpenCL/m01500_a0.cl +++ b/OpenCL/m01500_a0.cl @@ -7,6 +7,8 @@ #define _DES_ +#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" #define PERM_OP(a,b,tt,n,m) \ { \ @@ -340,11 +340,19 @@ __constant u32 c_skb[8][64] = }, }; +#if VECT_SIZE == 1 #define BOX(i,n,S) (S)[(n)][(i)] +#elif VECT_SIZE == 2 +#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1]) +#elif VECT_SIZE == 4 +#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3]) +#elif VECT_SIZE == 8 +#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7]) +#endif -static void _des_crypt_keysetup (u32 c, u32 d, u32 Kc[16], u32 Kd[16], __local u32 (*s_skb)[64]) +static void _des_crypt_keysetup (u32x c, u32x d, u32x Kc[16], u32x Kd[16], __local u32 (*s_skb)[64]) { - u32 tt; + u32x tt; PERM_OP (d, c, tt, 4, 0x0f0f0f0f); HPERM_OP (c, tt, 2, 0xcccc0000); @@ -360,7 +368,7 @@ static void _des_crypt_keysetup (u32 c, u32 d, u32 Kc[16], u32 Kd[16], __local u c = c & 0x0fffffff; - #pragma unroll + #pragma unroll 16 for (u32 i = 0; i < 16; i++) { if ((i < 2) || (i == 8) || (i == 15)) @@ -377,52 +385,52 @@ static void _des_crypt_keysetup (u32 c, u32 d, u32 Kc[16], u32 Kd[16], __local u c = c & 0x0fffffff; d = d & 0x0fffffff; - const u32 c00 = (c >> 0) & 0x0000003f; - const u32 c06 = (c >> 6) & 0x00383003; - const u32 c07 = (c >> 7) & 0x0000003c; - const u32 c13 = (c >> 13) & 0x0000060f; - const u32 c20 = (c >> 20) & 0x00000001; + const u32x c00 = (c >> 0) & 0x0000003f; + const u32x c06 = (c >> 6) & 0x00383003; + const u32x c07 = (c >> 7) & 0x0000003c; + const u32x c13 = (c >> 13) & 0x0000060f; + const u32x c20 = (c >> 20) & 0x00000001; - u32 s = BOX (((c00 >> 0) & 0xff), 0, s_skb) - | BOX (((c06 >> 0) & 0xff) - |((c07 >> 0) & 0xff), 1, s_skb) - | BOX (((c13 >> 0) & 0xff) - |((c06 >> 8) & 0xff), 2, s_skb) - | BOX (((c20 >> 0) & 0xff) - |((c13 >> 8) & 0xff) - |((c06 >> 16) & 0xff), 3, s_skb); + u32x s = BOX (((c00 >> 0) & 0xff), 0, s_skb) + | BOX (((c06 >> 0) & 0xff) + |((c07 >> 0) & 0xff), 1, s_skb) + | BOX (((c13 >> 0) & 0xff) + |((c06 >> 8) & 0xff), 2, s_skb) + | BOX (((c20 >> 0) & 0xff) + |((c13 >> 8) & 0xff) + |((c06 >> 16) & 0xff), 3, s_skb); - const u32 d00 = (d >> 0) & 0x00003c3f; - const u32 d07 = (d >> 7) & 0x00003f03; - const u32 d21 = (d >> 21) & 0x0000000f; - const u32 d22 = (d >> 22) & 0x00000030; + const u32x d00 = (d >> 0) & 0x00003c3f; + const u32x d07 = (d >> 7) & 0x00003f03; + const u32x d21 = (d >> 21) & 0x0000000f; + const u32x d22 = (d >> 22) & 0x00000030; - u32 t = BOX (((d00 >> 0) & 0xff), 4, s_skb) - | BOX (((d07 >> 0) & 0xff) - |((d00 >> 8) & 0xff), 5, s_skb) - | BOX (((d07 >> 8) & 0xff), 6, s_skb) - | BOX (((d21 >> 0) & 0xff) - |((d22 >> 0) & 0xff), 7, s_skb); + u32x t = BOX (((d00 >> 0) & 0xff), 4, s_skb) + | BOX (((d07 >> 0) & 0xff) + |((d00 >> 8) & 0xff), 5, s_skb) + | BOX (((d07 >> 8) & 0xff), 6, s_skb) + | BOX (((d21 >> 0) & 0xff) + |((d22 >> 0) & 0xff), 7, s_skb); Kc[i] = ((t << 16) | (s & 0x0000ffff)); Kd[i] = ((s >> 16) | (t & 0xffff0000)); } } -static void _des_crypt_encrypt (u32 iv[2], u32 mask, u32 Kc[16], u32 Kd[16], __local u32 (*s_SPtrans)[64]) +static void _des_crypt_encrypt (u32x iv[2], u32 mask, u32x Kc[16], u32x Kd[16], __local u32 (*s_SPtrans)[64]) { const u32 E1 = (mask >> 2) & 0x3f0; const u32 E0 = mask & 0x3f; - u32 r = 0; - u32 l = 0; + u32x r = 0; + u32x l = 0; for (u32 i = 0; i < 25; i++) { for (u32 j = 0; j < 16; j += 2) { - u32 t; - u32 u; + u32x t; + u32x u; t = r ^ (r >> 16); u = t & E0; @@ -465,7 +473,7 @@ static void _des_crypt_encrypt (u32 iv[2], u32 mask, u32 Kc[16], u32 Kd[16], __l | BOX (((t >> 24) & 0x3f), 7, s_SPtrans); } - u32 tt; + u32x tt; tt = l; l = r; @@ -497,6 +505,13 @@ __kernel void m01500_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu pw_buf0[2] = 0; pw_buf0[3] = 0; + u32 pw_buf1[4]; + + pw_buf1[0] = 0; + pw_buf1[1] = 0; + pw_buf1[2] = 0; + pw_buf1[3] = 0; + const u32 pw_len = pws[gid].pw_len; /** @@ -541,60 +556,35 @@ __kernel void m01500_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * main */ - 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] = 0; - w0[3] = 0; - - u32 w1[4]; - - w1[0] = 0; - w1[1] = 0; - w1[2] = 0; - w1[3] = 0; - - 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; - - u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); + u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); out_len = (out_len >= 8) ? 8 : out_len; - u32 data[2]; + u32x data[2]; data[0] = (w0[0] << 1) & 0xfefefefe; data[1] = (w0[1] << 1) & 0xfefefefe; - u32 Kc[16]; - u32 Kd[16]; + u32x Kc[16]; + u32x Kd[16]; _des_crypt_keysetup (data[0], data[1], Kc, Kd, s_skb); - u32 iv[2]; + u32x iv[2]; _des_crypt_encrypt (iv, mask, Kc, Kd, s_SPtrans); - const u32 r0 = iv[0]; - const u32 r1 = iv[1]; - const u32 r2 = 0; - const u32 r3 = 0; + u32x c = 0; + u32x d = 0; - #include COMPARE_M + COMPARE_M_SIMD (iv[0], iv[1], c, d); } } @@ -627,6 +617,13 @@ __kernel void m01500_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu pw_buf0[2] = 0; pw_buf0[3] = 0; + u32 pw_buf1[4]; + + pw_buf1[0] = 0; + pw_buf1[1] = 0; + pw_buf1[2] = 0; + pw_buf1[3] = 0; + const u32 pw_len = pws[gid].pw_len; /** @@ -683,60 +680,35 @@ __kernel void m01500_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * main */ - 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] = 0; - w0[3] = 0; - - u32 w1[4]; - - w1[0] = 0; - w1[1] = 0; - w1[2] = 0; - w1[3] = 0; - - 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; - - u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); + u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); out_len = (out_len >= 8) ? 8 : out_len; - u32 data[2]; + u32x data[2]; data[0] = (w0[0] << 1) & 0xfefefefe; data[1] = (w0[1] << 1) & 0xfefefefe; - u32 Kc[16]; - u32 Kd[16]; + u32x Kc[16]; + u32x Kd[16]; _des_crypt_keysetup (data[0], data[1], Kc, Kd, s_skb); - u32 iv[2]; + u32x iv[2]; _des_crypt_encrypt (iv, mask, Kc, Kd, s_SPtrans); - const u32 r0 = iv[0]; - const u32 r1 = iv[1]; - const u32 r2 = 0; - const u32 r3 = 0; + u32x c = 0; + u32x d = 0; - #include COMPARE_S + COMPARE_S_SIMD (iv[0], iv[1], c, d); } }