-
Notifications
You must be signed in to change notification settings - Fork 37
/
SSE.cpp
212 lines (180 loc) · 9.76 KB
/
SSE.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
#include "SSE.hpp"
#include "Hash.hpp"
#include "Utils.hpp"
SSE::SSE(Shared* const sh) : shared(sh),
Text{
{ /*APM:*/ {sh,1 << 15,20,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}} , /* APM: contexts, steps */
{ /*APM1:*/ {sh,1 << 16,7}, {sh,1 << 16,6}, {sh,1 << 16,6}}, /* APM1: contexts, rate */
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
},
Image{
// color:
{ { /*APM:*/ {sh,1 << 11,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}},
{ /*APM1:*/ {sh,1 << 16,7}, {sh,1 << 16,7}} ,
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
},
// palette:
{ { /*APM:*/ {sh,1 << 11,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}},
{ /*APM1:*/ {sh,1 << 16,5}, {sh,1 << 16,6}},
{ /*ApmPostA: */ sh,1},
{ /*ApmPostB: */ sh,1}
},
// gray:
{ { /*APM:*/ {sh,1 << 11,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}} ,
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
}
},
Audio{
{ /*APM:*/ {sh,1 << 14,24,1023} },
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
},
Jpeg{
{ /*APM:*/ {sh,0x1000 + 1,24,1023} },
{ /*ApmPostA: */ sh,1},
{ /*ApmPostB: */ sh,1}
},
DEC{
{ /*APM:*/ {sh,25*26,20,1023} },
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
},
x86_64{
{ /*APM:*/ {sh,1 << 14,20,1023}, {sh,1 << 16,16,1023}, {sh,1 << 16,16,1023} },
{ /*APM1:*/ {sh,1 << 13,7}, {sh,1 << 16,7}, {sh,1 << 16,7} },
{ /*ApmPostA: */ sh,1},
{ /*ApmPostB: */ sh,1}
},
Generic {
{ /*APM:*/ {sh,1<<8,20,1023}, {sh,1 << 8,24,1023}, {sh,1 << 16,24,1023}, {sh,1 << 16,24,1023}}, /* APM: contexts, steps */
{ /*APM1:*/ {sh,1 << 16,7}, {sh,1 << 16,7}, {sh,1 << 16,7}},
{ /*ApmPostA: */ sh,8},
{ /*ApmPostB: */ sh,8}
}
{}
uint32_t SSE::p(const uint32_t pr_orig) {
INJECT_SHARED_c0
INJECT_SHARED_bpos
INJECT_SHARED_c4
INJECT_SHARED_blockPos
INJECT_SHARED_blockType
assert(shared->State.NormalModel.order <= 7);
assert(shared->State.WordModel.order <= 31);
assert(shared->State.Text.order <= 15);
assert(shared->State.x86_64.state <= 255);
assert(shared->State.Audio <= 255);
assert(shared->State.JPEG.state <= 4095 + 1);
//uint32_t misses4 = shared->State.misses & 15u;
uint32_t misses = shared->State.misses << ((8 - bpos) & 7); //byte-aligned
misses = (misses & 0xffffff00) | (misses & 0xff) >> ((8 - bpos) & 7);
uint32_t misses3 =
((misses & 0x1) != 0) |
((misses & 0xfe) != 0) << 1 |
((misses & 0xff00) != 0) << 2;
const BlockType normalizedBlockType =
blockType == BlockType::JPEG && shared->State.JPEG.state == 0 ? BlockType::DEFAULT :
blockType;
switch(normalizedBlockType) {
case BlockType::TEXT:
case BlockType::TEXT_EOL: {
uint32_t pr0 = Text.APMs[0].p(pr_orig, static_cast<uint32_t>(c0) << 7 | (shared->State.Text.mask & 0x0F) | misses3 << 4); //15
uint32_t pr1 = Text.APMs[1].p(pr_orig, finalize64(hash(bpos, misses3 & 3, c4 & 0xFFFF, shared->State.Text.mask >> 4), 16)); //16
uint32_t pr2 = Text.APMs[2].p(pr_orig, finalize64(hash(c0, shared->State.Match.expectedByte << 2 | shared->State.Match.length2), 16)); //16
uint32_t pr3 = Text.APMs[3].p(pr_orig, finalize64(hash(c0, c4 & 0xFFFF, shared->State.Text.firstLetter), 16)); //16
uint32_t prA = (pr_orig + pr1 + pr2 + pr3 + 2) >> 2;
uint32_t pr4 = Text.APM1s[0].p(prA, (shared->State.WordModel.order>>2)<<13 | shared->State.Match.expectedByte << 5 | shared->State.Match.length2 << 3 | (shared->State.Text.order>>1)); //16
uint32_t pr5 = Text.APM1s[1].p(pr0, finalize64(hash(c0, c4 & 0x00FFFFFF), 16)); //16
uint32_t pr6 = Text.APM1s[2].p(pr0, finalize64(hash(c0, c4), 16)); //16
uint32_t prB = (pr0 + pr4 + pr5 + pr6 + 2) >> 2;
uint32_t pr = (Text.APMPostA.p(prA, bpos) + Text.APMPostB.p(prB, bpos) + 1) >> 1;
return pr;
break;
}
case BlockType::IMAGE24:
case BlockType::IMAGE32: {
uint32_t pr0 = Image.Color.APMs[0].p(pr_orig, static_cast<uint32_t>(c0) << 3 | (bpos==0 ? 0 : (misses3&3))); //11
uint32_t pr1 = Image.Color.APMs[1].p(pr_orig, finalize64(hash(c0, shared->State.Image.pixels.W, shared->State.Image.pixels.WW), 16)); //16
uint32_t pr2 = Image.Color.APMs[2].p(pr_orig, finalize64(hash(c0, shared->State.Image.pixels.N, shared->State.Image.pixels.NN), 16)); //16
uint32_t pr3 = Image.Color.APMs[3].p(pr_orig, (c0 << 8) | shared->State.Image.ctx); //16
uint32_t prA = (pr_orig + pr1 + pr2 + pr3 + 2) >> 2;
uint32_t pr4 = Image.Color.APM1s[0].p(pr0, finalize64(hash(c0, shared->State.Image.pixels.W, (c4 & 0xFF) - shared->State.Image.pixels.Wp1, shared->State.Image.plane), 16)); //16
uint32_t pr5 = Image.Color.APM1s[1].p(pr0, finalize64(hash(c0, shared->State.Image.pixels.N, (c4 & 0xFF) - shared->State.Image.pixels.Np1, shared->State.Image.plane), 16)); //16
uint32_t prB = (pr0 * 2 + pr4 * 3 + pr5 * 3 + 4) >> 3;
uint32_t pr = (Image.Color.APMPostA.p(prA, bpos) + Image.Color.APMPostB.p(prB, bpos) + 1) >> 1;
return pr;
break;
}
case BlockType::IMAGE8GRAY: {
uint32_t pr0 = Image.Gray.APMs[0].p(pr_orig, static_cast<uint32_t>(c0) << 3 | (bpos == 0 ? 0 : (misses3 & 3))); //11
uint32_t pr1 = Image.Gray.APMs[1].p(pr0, (c0 << 8) | shared->State.Image.ctx); //16
uint32_t pr2 = Image.Gray.APMs[2].p(pr_orig, bpos | (shared->State.Image.ctx & 0xF8) | (shared->State.Match.expectedByte << 8)); //16
int prA = (2 * pr_orig + pr1 + pr2 + 2) >> 2;
uint32_t pr = (Image.Gray.APMPostA.p(pr0, bpos) + Image.Gray.APMPostB.p(prA, bpos) + 1) >> 1;
return pr;
break;
}
case BlockType::IMAGE8: {
uint32_t pr0 = Image.Palette.APMs[0].p(pr_orig, static_cast<uint32_t>(c0) << 3 | (bpos == 0 ? 0 : (misses3 & 3))); //11
uint32_t pr1 = Image.Palette.APMs[1].p(pr_orig, finalize64(hash(c0 | shared->State.Image.pixels.W << 8 | shared->State.Image.pixels.N << 16), 16)); //16
uint32_t pr2 = Image.Palette.APMs[2].p(pr_orig, finalize64(hash(c0 | shared->State.Image.pixels.N << 8 | shared->State.Image.pixels.NN << 16), 16)); //16
uint32_t pr3 = Image.Palette.APMs[3].p(pr_orig, finalize64(hash(c0 | shared->State.Image.pixels.W << 8 | shared->State.Image.pixels.WW << 16), 16)); //16
uint32_t prA = (pr_orig + pr1 + pr2 + pr3 + 2) >> 2;
uint32_t pr4 = Image.Palette.APM1s[0].p(prA, finalize64(hash(c0 | shared->State.Match.expectedByte << 8 | shared->State.Image.pixels.N << 16), 16)); //16
uint32_t pr5 = Image.Palette.APM1s[1].p(pr0, finalize64(hash(c0 | shared->State.Image.pixels.W << 8 | shared->State.Image.pixels.N << 16), 16)); //16
uint32_t prB = (pr0 * 2 + pr4 + pr5 + 2) >> 2;
uint32_t pr = (Image.Palette.APMPostA.p(prA, 0) + Image.Palette.APMPostB.p(prB, 0) + 1) >> 1;
return pr;
break;
}
//case BlockType::IMAGE4: //TODO
//case BlockType::IMAGE1: //TODO
case BlockType::AUDIO:
case BlockType::AUDIO_LE: {
uint32_t pr0 = Audio.APMs[0].p(pr_orig, shared->State.Audio << 6 | static_cast<uint32_t>(bpos) << 3 | misses3); //14
uint32_t pr = (Audio.APMPostA.p(pr_orig, bpos) + Audio.APMPostB.p(pr0, bpos) + 1) >> 1;
return pr;
break;
}
case BlockType::JPEG: {
uint32_t pr0 = Jpeg.APMs[0].p(pr_orig, shared->State.JPEG.state);
uint32_t pr = (Jpeg.APMPostA.p(pr_orig, 0) + Jpeg.APMPostB.p(pr0, 0) + 1) >> 1;
return pr;
break;
}
case BlockType::DEC_ALPHA: {
uint32_t pr0 = DEC.APMs[0].p(pr_orig, (shared->State.DEC.state * 26) + shared->State.DEC.bcount);
uint32_t pr = (DEC.APMPostA.p(pr_orig, 0) + DEC.APMPostB.p(pr0, 0) + 1) >> 1;
return pr;
break;
}
case BlockType::EXE: {
uint32_t pr0 = x86_64.APMs[0].p(pr_orig, shared->State.x86_64.state << 6 | misses3 << 3 | bpos);//14
uint32_t pr1 = x86_64.APMs[1].p(pr_orig, shared->State.x86_64.state << 8 | c0); // 16
uint32_t pr2 = x86_64.APMs[2].p((pr0 + pr1 + 1) >> 1, finalize64(hash(c4 & 0xFF, bpos, misses3 & 1, shared->State.x86_64.state >> 3), 16)); //16
uint32_t prA = (pr_orig + pr0 + pr1 + pr2 + 2) >> 2;
uint32_t pr4 = x86_64.APM1s[0].p(prA, shared->State.Match.expectedByte << 5 | shared->State.NormalModel.order << 2 | shared->State.Match.length2); //13
uint32_t pr5 = x86_64.APM1s[1].p(prA, c0 | (c4 & 0xFF) << 8); //16
uint32_t pr6 = x86_64.APM1s[1].p(pr_orig, c0 | (c4 & 0xFF) << 8); //16
uint32_t prB = (pr0 + pr4 + pr5 + pr6 + 2) >> 2;
uint32_t pr = (x86_64.APMPostA.p(prA, 0) + x86_64.APMPostB.p(prB, 0) + 1) >> 1;
return pr;
break;
}
default: {
uint32_t pr0 = Generic.APMs[0].p(pr_orig, shared->State.Match.length2 << 6 | static_cast<uint32_t>(bpos) << 3 | misses3); //8
uint32_t pr1 = Generic.APMs[1].p(pr_orig, shared->State.NormalModel.order << 5 | shared->State.Match.length2 << 3 | bpos); //8
uint32_t pr2 = Generic.APMs[2].p(pr_orig, c0 | (c4 & 0xFF) << 8); //16
uint32_t pr3 = Generic.APMs[3].p(pr_orig, c0 << 8 | (c4 & 0xF0) | (c4 & 0xF000 >> 12)); //16
uint32_t prA = (pr_orig + pr1 + pr2 + pr3 + 2) >> 2;
uint32_t pr4 = Generic.APM1s[0].p(prA, misses3<<13 | shared->State.Match.expectedByte << 5 | shared->State.NormalModel.order << 2 | shared->State.Match.length2); //16
uint32_t pr5 = Generic.APM1s[1].p(pr0, misses3<<13 | shared->State.Match.length2 << 11 | (shared->State.WordModel.order >> 2) << 8 | c0); //16
uint32_t pr6 = Generic.APM1s[2].p(pr0, (misses3&3)<<14 | (bpos>>1) << 12 | (c4 & 0xFF) << 4 | (static_cast<uint32_t>(shared->State.WordModel.order) >> 1)); //16
uint32_t prB = (pr0 + pr4 + pr5 + pr6 + 2) >> 2;
uint32_t pr = (Generic.APMPostA.p(prA, shared->State.NormalModel.order) + Generic.APMPostB.p(prB, shared->State.Match.mode3) + 1) >> 1;
return pr;
}
}
}