2 * Author......: Jens Steube <jens.steube@gmail.com>
12 __device__ static void generate_pw (u32 pw_buf[16], cs_t *root_css_buf, cs_t *markov_css_buf, const u32 pw_l_len, const u32 pw_r_len, const u32 mask80, const u32 bits14, const u32 bits15, u64 val)
31 cs_t *cs = &root_css_buf[pw_r_len];
36 for (i = 0, j = pw_r_len; i < pw_l_len; i++, j++)
38 const u32 len = cs->cs_len;
40 const u64 next = val / len;
41 const u64 pos = val % len;
45 const u32 key = cs->cs_buf[pos];
47 const u32 jd4 = j / 4;
48 const u32 jm4 = j % 4;
50 pw_buf[jd4] |= key << (jm4 * 8);
52 cs = &markov_css_buf[(j * CHARSIZ) + key];
55 const u32 jd4 = j / 4;
56 const u32 jm4 = j % 4;
58 pw_buf[jd4] |= (0xff << (jm4 * 8)) & mask80;
60 if (bits14) pw_buf[14] = (pw_l_len + pw_r_len) * 8;
61 if (bits15) pw_buf[15] = (pw_l_len + pw_r_len) * 8;
64 extern "C" __global__ void __launch_bounds__ (256, 1) l_markov (pw_t *pws_buf_l, cs_t *root_css_buf, cs_t *markov_css_buf, const u64 off, const u32 pw_l_len, const u32 pw_r_len, const u32 mask80, const u32 bits14, const u32 bits15, const u32 gid_max)
66 const u32 gid = (blockIdx.x * blockDim.x) + threadIdx.x;
68 if (gid >= gid_max) return;
70 const u32 gid2 = gid * 2;
75 generate_pw (pw_buf0, root_css_buf, markov_css_buf, pw_l_len, pw_r_len, mask80, bits14, bits15, off + gid2 + 0);
76 generate_pw (pw_buf1, root_css_buf, markov_css_buf, pw_l_len, pw_r_len, mask80, bits14, bits15, off + gid2 + 1);
79 for (int i = 0; i < 16; i++)
81 pws_buf_l[gid].i[i].x = pw_buf0[i];
82 pws_buf_l[gid].i[i].y = pw_buf1[i];
85 pws_buf_l[gid].pw_len = pw_l_len + pw_r_len;
88 extern "C" __global__ void __launch_bounds__ (256, 1) r_markov (bf_t *pws_buf_r, cs_t *root_css_buf, cs_t *markov_css_buf, const u64 off, const u32 pw_r_len, const u32 mask80, const u32 bits14, const u32 bits15, const u32 gid_max)
90 const u32 gid = (blockIdx.x * blockDim.x) + threadIdx.x;
92 if (gid >= gid_max) return;
94 const u32 gid2 = gid * 2;
98 generate_pw (pw_buf, root_css_buf, markov_css_buf, pw_r_len, 0, 0, 0, 0, off + gid2 + 0);
100 pws_buf_r[gid2 + 0].i = pw_buf[0];
102 generate_pw (pw_buf, root_css_buf, markov_css_buf, pw_r_len, 0, 0, 0, 0, off + gid2 + 1);
104 pws_buf_r[gid2 + 1].i = pw_buf[0];
107 extern "C" __global__ void __launch_bounds__ (256, 1) C_markov (comb_t *pws_buf, cs_t *root_css_buf, cs_t *markov_css_buf, const u64 off, const u32 pw_len, const u32 mask80, const u32 bits14, const u32 bits15, const u32 gid_max)
109 const u32 gid = (blockIdx.x * blockDim.x) + threadIdx.x;
111 if (gid >= gid_max) return;
113 const u32 gid2 = gid * 2;
117 generate_pw (pw_buf, root_css_buf, markov_css_buf, pw_len, 0, mask80, bits14, bits15, off + gid2 + 0);
119 pws_buf[gid2 + 0].i[ 0] = pw_buf[ 0];
120 pws_buf[gid2 + 0].i[ 1] = pw_buf[ 1];
121 pws_buf[gid2 + 0].i[ 2] = pw_buf[ 2];
122 pws_buf[gid2 + 0].i[ 3] = pw_buf[ 3];
123 pws_buf[gid2 + 0].i[ 4] = pw_buf[ 4];
124 pws_buf[gid2 + 0].i[ 5] = pw_buf[ 5];
125 pws_buf[gid2 + 0].i[ 6] = pw_buf[ 6];
126 pws_buf[gid2 + 0].i[ 7] = pw_buf[ 7];
128 pws_buf[gid2 + 0].pw_len = pw_len;
130 generate_pw (pw_buf, root_css_buf, markov_css_buf, pw_len, 0, mask80, bits14, bits15, off + gid2 + 1);
132 pws_buf[gid2 + 1].i[ 0] = pw_buf[ 0];
133 pws_buf[gid2 + 1].i[ 1] = pw_buf[ 1];
134 pws_buf[gid2 + 1].i[ 2] = pw_buf[ 2];
135 pws_buf[gid2 + 1].i[ 3] = pw_buf[ 3];
136 pws_buf[gid2 + 1].i[ 4] = pw_buf[ 4];
137 pws_buf[gid2 + 1].i[ 5] = pw_buf[ 5];
138 pws_buf[gid2 + 1].i[ 6] = pw_buf[ 6];
139 pws_buf[gid2 + 1].i[ 7] = pw_buf[ 7];
141 pws_buf[gid2 + 1].pw_len = pw_len;