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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
|
/*
MMX ditherer for 32 bit displays
Copyright (C) 2000 Martin Vogt
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU Library General Public License as published by
the Free Software Foundation.
For more information look at the file COPYRIGHT in this package
*/
#include "ditherMMX.h"
#include <iostream>
using namespace std;
#ifndef INTEL
void dither32_mmx(unsigned char* lum,
unsigned char* cr,
unsigned char* cb,
unsigned char* out,
int rows,
int cols,
int mod) {
printf("urgs! dither32_mmx \n");
printf("never should happen!\n");
exit(0);
}
#else
static unsigned long MMX32_80w[] = {0x00800080, 0x00800080};
static unsigned long MMX32_10w[] = {0x00100010, 0x00100010};
static unsigned long MMX32_00FFw[] = {0x00ff00ff, 0x00ff00ff};
static unsigned long MMX32_FF00w[] = {0xff00ff00, 0xff00ff00};
static unsigned short MMX32_Ycoeff[] = {0x4a, 0x4a, 0x4a, 0x4a};
static unsigned short MMX32_Vredcoeff[] = {0x59, 0x59, 0x59, 0x59};
static unsigned short MMX32_Ubluecoeff[] = {0x72, 0x72, 0x72, 0x72};
static unsigned short MMX32_Ugrncoeff[] = {0xffea,0xffea,0xffea,0xffea};
static unsigned short MMX32_Vgrncoeff[] = {0xffd2,0xffd2,0xffd2,0xffd2};
void dummy_dithermmx32() {
cout << "MMX32_10w:"<<MMX32_10w<<endl;
cout << "MMX32_80w:"<<MMX32_80w<<endl;
cout << "MMX32_Ubluecoeff:"<<MMX32_Ubluecoeff<<endl;
cout << "MMX32_Vredcoeff:"<<MMX32_Vredcoeff<<endl;
cout << "MMX32_Ugrncoeff:"<<MMX32_Ugrncoeff<<endl;
cout << "MMX32_Vgrncoeff:"<<MMX32_Vgrncoeff<<endl;
cout << "MMX32_Ycoeff:"<<MMX32_Ycoeff<<endl;
cout << "MMX32_00FFw:"<<MMX32_00FFw<<endl;
cout << "MMX32_FF00w:"<<MMX32_FF00w<<endl;
}
/**
This MMX assembler is my first assembler/MMX program ever.
Thus it maybe buggy.
Send patches to:
mvogt@rhrk.uni-kl.de
After it worked fine I have "obfuscated" the code a bit to have
more parallism in the MMX units. This means I moved
initilisation around and delayed other instruction.
Performance measurement did not show that this brought any advantage
but in theory it _should_ be faster this way.
The overall performanve gain to the C based dither was 30%-40%.
The MMX routine calculates 256bit=8RGB values in each cycle
(4 for row1 & 4 for row2)
The red/green/blue.. coefficents are taken from the mpeg_play
player. They look nice, but I dont know if you can have
better values, to avoid integer rounding errors.
IMPORTANT:
==========
It is a requirement that the cr/cb/lum are 8 byte aligned and
the out are 16byte aligned or you will/may get segfaults
*/
void dither32_mmx(unsigned char* lum,
unsigned char* cr,
unsigned char* cb,
unsigned char* out,
int rows,
int cols,
int mod) {
unsigned int *row1;
unsigned int *row2;
row1 = (unsigned int *)out; // 32 bit target
unsigned char* end = lum +cols*rows; // Pointer to the end
int x=cols;
row2=row1+cols+mod; // start of second row
mod=4*cols+8*mod; // increment for row1 in byte
// buffer for asm function
int buf[6];
buf[0]=(int)(lum+cols); // lum2 pointer
buf[1]=(int)end;
buf[2]=x;
buf[3]=mod;
buf[4]=0; //tmp0;
buf[5]=cols;
__asm__ __volatile__ (
".align 32\n"
"1:\n"
// create Cr (result in mm1)
"movd (%0), %%mm1\n" // 0 0 0 0 v3 v2 v1 v0
"pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00
"movd (%2), %%mm2\n" // 0 0 0 0 l3 l2 l1 l0
"punpcklbw %%mm7,%%mm1\n" // 0 v3 0 v2 00 v1 00 v0
"punpckldq %%mm1,%%mm1\n" // 00 v1 00 v0 00 v1 00 v0
"psubw MMX32_80w,%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
// create Cr_g (result in mm0)
"movq %%mm1,%%mm0\n" // r1 r1 r0 r0 r1 r1 r0 r0
"pmullw MMX32_Vgrncoeff,%%mm0\n" // red*-46dec=0.7136*64
"pmullw MMX32_Vredcoeff,%%mm1\n" // red*89dec=1.4013*64
"psraw $6, %%mm0\n" // red=red/64
"psraw $6, %%mm1\n" // red=red/64
// create L1 L2 (result in mm2,mm4)
// L2=lum2
"movl %2,16%5\n" // store register in tmp0
"movl %5,%2\n" // lum2->register
"movd (%2),%%mm3\n" // 0 0 0 0 L3 L2 L1 L0
"movl 16%5,%2\n" // tmp0->register
"punpckldq %%mm3,%%mm2\n" // L3 L2 L1 L0 l3 l2 l1 l0
"movq %%mm2,%%mm4\n" // L3 L2 L1 L0 l3 l2 l1 l0
"pand MMX32_FF00w, %%mm2\n" // L3 0 L1 0 l3 0 l1 0
"pand MMX32_00FFw, %%mm4\n" // 0 L2 0 L0 0 l2 0 l0
"psrlw $8,%%mm2\n" // 0 L3 0 L1 0 l3 0 l1
// create R (result in mm6)
"movq %%mm2,%%mm5\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm6\n" // 0 L2 0 L0 0 l2 0 l0
"paddsw %%mm1, %%mm5\n" // lum1+red:x R3 x R1 x r3 x r1
"paddsw %%mm1, %%mm6\n" // lum1+red:x R2 x R0 x r2 x r0
"packuswb %%mm5,%%mm5\n" // R3 R1 r3 r1 R3 R1 r3 r1
"packuswb %%mm6,%%mm6\n" // R2 R0 r2 r0 R2 R0 r2 r0
"pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00
"punpcklbw %%mm5,%%mm6\n" // R3 R2 R1 R0 r3 r2 r1 r0
// create Cb (result in mm1)
"movd (%1), %%mm1\n" // 0 0 0 0 u3 u2 u1 u0
"punpcklbw %%mm7,%%mm1\n" // 0 u3 0 u2 00 u1 00 u0
"punpckldq %%mm1,%%mm1\n" // 00 u1 00 u0 00 u1 00 u0
"psubw MMX32_80w,%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
// create Cb_g (result in mm5)
"movq %%mm1,%%mm5\n" // u1 u1 u0 u0 u1 u1 u0 u0
"pmullw MMX32_Ugrncoeff,%%mm5\n" // blue*-109dec=1.7129*64
"pmullw MMX32_Ubluecoeff,%%mm1\n" // blue*114dec=1.78125*64
"psraw $6, %%mm5\n" // blue=red/64
"psraw $6, %%mm1\n" // blue=blue/64
// create G (result in mm7)
"movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm7\n" // 0 L2 0 L0 0 l2 0 l1
"paddsw %%mm5, %%mm3\n" // lum1+Cb_g:x G3t x G1t x g3t x g1t
"paddsw %%mm5, %%mm7\n" // lum1+Cb_g:x G2t x G0t x g2t x g0t
"paddsw %%mm0, %%mm3\n" // lum1+Cr_g:x G3 x G1 x g3 x g1
"paddsw %%mm0, %%mm7\n" // lum1+blue:x G2 x G0 x g2 x g0
"packuswb %%mm3,%%mm3\n" // G3 G1 g3 g1 G3 G1 g3 g1
"packuswb %%mm7,%%mm7\n" // G2 G0 g2 g0 G2 G0 g2 g0
"punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
// create B (result in mm5)
"movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm5\n" // 0 L2 0 L0 0 l2 0 l1
"paddsw %%mm1, %%mm3\n" // lum1+blue:x B3 x B1 x b3 x b1
"paddsw %%mm1, %%mm5\n" // lum1+blue:x B2 x B0 x b2 x b0
"packuswb %%mm3,%%mm3\n" // B3 B1 b3 b1 B3 B1 b3 b1
"packuswb %%mm5,%%mm5\n" // B2 B0 b2 b0 B2 B0 b2 b0
"punpcklbw %%mm3,%%mm5\n" // B3 B2 B1 B0 b3 b2 b1 b0
// fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"pxor %%mm4,%%mm4\n" // 0 0 0 0 0 0 0 0
"movq %%mm6,%%mm1\n" // R3 R2 R1 R0 r3 r2 r1 r0
"movq %%mm5,%%mm3\n" // B3 B2 B1 B0 b3 b2 b1 b0
// process lower lum
"punpcklbw %%mm4,%%mm1\n" // 0 r3 0 r2 0 r1 0 r0
"punpcklbw %%mm4,%%mm3\n" // 0 b3 0 b2 0 b1 0 b0
"movq %%mm1,%%mm2\n" // 0 r3 0 r2 0 r1 0 r0
"movq %%mm3,%%mm0\n" // 0 b3 0 b2 0 b1 0 b0
"punpcklwd %%mm1,%%mm3\n" // 0 r1 0 b1 0 r0 0 b0
"punpckhwd %%mm2,%%mm0\n" // 0 r3 0 b3 0 r2 0 b2
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"movq %%mm7,%%mm1\n" // G3 G2 G1 G0 g3 g2 g1 g0
"punpcklbw %%mm1,%%mm2\n" // g3 0 g2 0 g1 0 g0 0
"punpcklwd %%mm4,%%mm2\n" // 0 0 g1 0 0 0 g0 0
"por %%mm3, %%mm2\n" // 0 r1 g1 b1 0 r0 g0 b0
"movq %%mm2,(%3)\n" // wrote out ! row1
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"punpcklbw %%mm1,%%mm4\n" // g3 0 g2 0 g1 0 g0 0
"punpckhwd %%mm2,%%mm4\n" // 0 0 g3 0 0 0 g2 0
"por %%mm0, %%mm4\n" // 0 r3 g3 b3 0 r2 g2 b2
"movq %%mm4,8(%3)\n" // wrote out ! row1
// fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
// this can be done "destructive"
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"punpckhbw %%mm2,%%mm6\n" // 0 R3 0 R2 0 R1 0 R0
"punpckhbw %%mm1,%%mm5\n" // G3 B3 G2 B2 G1 B1 G0 B0
"movq %%mm5,%%mm1\n" // G3 B3 G2 B2 G1 B1 G0 B0
"punpcklwd %%mm6,%%mm1\n" // 0 R1 G1 B1 0 R0 G0 B0
"movq %%mm1,(%4)\n" // wrote out ! row2
"punpckhwd %%mm6,%%mm5\n" // 0 R3 G3 B3 0 R2 G2 B2
"movq %%mm5,8(%4)\n" // wrote out ! row2
"addl $4,%2\n" // lum+4
"addl $4,%5\n" // lum2+4
"leal 16(%3),%3\n" // row1+16
"leal 16(%4),%4\n" // row2+16
"addl $2, %0\n" // cr+2
"addl $2, %1\n" // cb+2
"subl $4,8%5\n" // x+4 x is buf[2]
"cmpl $0,8%5\n"
"jne 1b\n"
"addl 20%5, %2\n" // lum += cols
"movl %2,16%5\n" // store register in tmp0
"movl 20%5,%2\n" // cols->register
"addl %2, %5\n" // lum2 += cols
"addl 12%5, %3\n" // row1+= mod is buf[0]
"addl 12%5, %4\n" // row2+= mod is buf[0]
"movl %2, 8%5\n" // x=cols
"movl 16%5,%2\n" // store tmp0 in register
"cmpl 4%5, %2\n" // buf[1] is end
"jl 1b\n"
"emms\n"
:
: "r" (cr), "r"(cb),"r"(lum),
"r"(row1),"r"(row2),"m"(buf[0])
);
}
#endif
|
