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
|
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2009 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: sse2trans.h 15675 2009-02-06 09:43:27Z tterribe $
********************************************************************/
#if !defined(_x86_vc_x86zigzag_H)
# define _x86_vc_x86zigzag_H (1)
# include "x86enc.h"
/*Converts DCT coefficients from transposed order into zig-zag scan order and
stores them in Y.
This relies on two macros to load the contents of each row:
OC_ZZ_LOAD_ROW_LO(row,reg) and OC_ZZ_LOAD_ROW_HI(row,reg), which load the
first four and second four entries of each row into the specified register,
respectively.
OC_ZZ_LOAD_ROW_LO must be called before OC_ZZ_LOAD_ROW_HI for the same row
(because when the rows are already in SSE2 registers, loading the high half
destructively modifies the register).
The index of each output element in the original 64-element array should wind
up in the following 8x8 matrix (the letters indicate the order we compute
each 4-tuple below):
A 0 8 1 2 9 16 24 17 B
C 10 3 4 11 18 25 32 40 E
F 33 26 19 12 5 6 13 20 D
G 27 34 41 48 56 49 42 35 I
L 28 21 14 7 15 22 29 36 M
H 43 50 57 58 51 44 37 30 O
N 23 31 38 45 52 59 60 53 J
P 46 39 47 54 61 62 55 63 K
The order of the coefficients within each tuple is reversed in the comments
below to reflect the usual MSB to LSB notation.*/
#define OC_TRANSPOSE_ZIG_ZAG_MMXEXT \
OC_ZZ_LOAD_ROW_LO(0,mm0) /*mm0=03 02 01 00*/ \
OC_ZZ_LOAD_ROW_LO(1,mm1) /*mm1=11 10 09 08*/ \
OC_ZZ_LOAD_ROW_LO(2,mm2) /*mm2=19 18 17 16*/ \
OC_ZZ_LOAD_ROW_LO(3,mm3) /*mm3=27 26 25 24*/ \
OC_ZZ_LOAD_ROW_HI(0,mm4) /*mm4=07 06 05 04*/ \
OC_ZZ_LOAD_ROW_HI(1,mm5) /*mm5=15 14 13 12*/ \
OC_ZZ_LOAD_ROW_HI(2,mm6) /*mm6=23 22 21 20*/ \
__asm movq mm7,mm0 /*mm7=03 02 01 00*/ \
__asm punpckhdq mm0,mm1 /*mm0=11 10 03 02*/ \
__asm pshufw mm4,mm4,0x39 /*mm4=04 07 06 05*/ \
__asm punpcklwd mm1,mm0 /*mm1=03 09 02 08*/ \
__asm pshufw mm5,mm5,0x39 /*mm5=12 15 14 13*/ \
__asm punpcklwd mm7,mm1 /*mm7=02 01 08 00 *A*/ \
__asm movq [Y+0x00],mm7 \
__asm punpckhwd mm1,mm4 /*mm1=04 03 07 09*/ \
__asm movq mm7,mm2 /*mm7=19 18 17 16*/ \
__asm punpckhdq mm0,mm1 /*mm0=04 03 11 10*/ \
__asm punpckhwd mm7,mm5 /*mm7=12 19 15 18*/ \
__asm punpcklwd mm1,mm3 /*mm1=25 07 24 09*/ \
__asm punpcklwd mm5,mm6 /*mm5=21 14 20 13*/ \
__asm punpcklwd mm1,mm2 /*mm1=17 24 16 09 *B*/ \
OC_ZZ_LOAD_ROW_LO(4,mm2) /*mm2=35 34 33 32*/ \
__asm movq [Y+0x08],mm1 \
OC_ZZ_LOAD_ROW_LO(5,mm1) /*mm1=43 42 41 40*/ \
__asm pshufw mm0,mm0,0x78 /*mm0=11 04 03 10 *C*/ \
__asm movq [Y+0x10],mm0 \
__asm punpckhdq mm6,mm4 /*mm6=?? 07 23 22*/ \
__asm punpckldq mm4,mm5 /*mm4=20 13 06 05 *D*/ \
__asm movq [Y+0x28],mm4 \
__asm psrlq mm3,16 /*mm3=.. 27 26 25*/ \
__asm pshufw mm0,mm2,0x0E /*mm0=?? ?? 35 34*/ \
__asm movq mm4,mm7 /*mm4=12 19 15 18*/ \
__asm punpcklwd mm2,mm3 /*mm2=26 33 25 32*/ \
__asm punpcklwd mm4,mm1 /*mm4=41 15 40 18*/ \
__asm punpckhwd mm3,mm1 /*mm3=43 .. 42 27*/ \
__asm punpckldq mm4,mm2 /*mm4=25 32 40 18*/ \
__asm punpcklwd mm3,mm0 /*mm3=35 42 34 27*/ \
OC_ZZ_LOAD_ROW_LO(6,mm0) /*mm0=51 50 49 48*/ \
__asm pshufw mm4,mm4,0x6C /*mm4=40 32 25 18 *E*/ \
__asm movq [Y+0x18],mm4 \
OC_ZZ_LOAD_ROW_LO(7,mm4) /*mm4=59 58 57 56*/ \
__asm punpckhdq mm2,mm7 /*mm2=12 19 26 33 *F*/ \
__asm movq [Y+0x20],mm2 \
__asm pshufw mm1,mm1,0xD0 /*mm1=43 41 ?? ??*/ \
__asm pshufw mm0,mm0,0x87 /*mm0=50 48 49 51*/ \
__asm movq mm2,mm3 /*mm2=35 42 34 27*/ \
__asm punpckhwd mm1,mm0 /*mm1=50 43 48 41*/ \
__asm pshufw mm4,mm4,0x93 /*mm4=58 57 56 59*/ \
__asm punpckldq mm3,mm1 /*mm3=48 41 34 27 *G*/ \
__asm movq [Y+0x30],mm3 \
__asm punpckhdq mm1,mm4 /*mm1=58 57 50 43 *H*/ \
__asm movq [Y+0x50],mm1 \
OC_ZZ_LOAD_ROW_HI(7,mm1) /*mm1=63 62 61 60*/ \
__asm punpcklwd mm4,mm0 /*mm4=49 56 51 59*/ \
OC_ZZ_LOAD_ROW_HI(6,mm0) /*mm0=55 54 53 52*/ \
__asm psllq mm6,16 /*mm6=07 23 22 ..*/ \
__asm movq mm3,mm4 /*mm3=49 56 51 59*/ \
__asm punpckhdq mm4,mm2 /*mm4=35 42 49 56 *I*/ \
OC_ZZ_LOAD_ROW_HI(3,mm2) /*mm2=31 30 29 28*/ \
__asm movq [Y+0x38],mm4 \
__asm punpcklwd mm3,mm1 /*mm3=61 51 60 59*/ \
__asm punpcklwd mm7,mm6 /*mm7=22 15 .. ??*/ \
__asm movq mm4,mm3 /*mm4=61 51 60 59*/ \
__asm punpcklwd mm3,mm0 /*mm3=53 60 52 59*/ \
__asm punpckhwd mm4,mm0 /*mm4=55 61 54 51*/ \
OC_ZZ_LOAD_ROW_HI(4,mm0) /*mm0=39 38 37 36*/ \
__asm pshufw mm3,mm3,0xE1 /*mm3=53 60 59 52 *J*/ \
__asm movq [Y+0x68],mm3 \
__asm movq mm3,mm4 /*mm3=?? ?? 54 51*/ \
__asm pshufw mm2,mm2,0x39 /*mm2=28 31 30 29*/ \
__asm punpckhwd mm4,mm1 /*mm4=63 55 62 61 *K*/ \
OC_ZZ_LOAD_ROW_HI(5,mm1) /*mm1=47 46 45 44*/ \
__asm movq [Y+0x78],mm4 \
__asm punpckhwd mm6,mm2 /*mm6=28 07 31 23*/ \
__asm punpcklwd mm2,mm0 /*mm2=37 30 36 29*/ \
__asm punpckhdq mm5,mm6 /*mm5=28 07 21 14*/ \
__asm pshufw mm2,mm2,0x4B /*mm2=36 29 30 37*/ \
__asm pshufw mm5,mm5,0x87 /*mm5=07 14 21 28 *L*/ \
__asm movq [Y+0x40],mm5 \
__asm punpckhdq mm7,mm2 /*mm7=36 29 22 15 *M*/ \
__asm movq [Y+0x48],mm7 \
__asm pshufw mm1,mm1,0x9C /*mm1=46 45 47 44*/ \
__asm punpckhwd mm0,mm1 /*mm0=46 39 45 38*/ \
__asm punpcklwd mm3,mm1 /*mm3=47 54 44 51*/ \
__asm punpckldq mm6,mm0 /*mm6=45 38 31 23 *N*/ \
__asm movq [Y+0x60],mm6 \
__asm punpckhdq mm0,mm3 /*mm0=47 54 46 39*/ \
__asm punpckldq mm3,mm2 /*mm3=30 37 44 51 *O*/ \
__asm movq [Y+0x58],mm3 \
__asm pshufw mm0,mm0,0xB1 /*mm0=54 47 39 46 *P*/ \
__asm movq [Y+0x70],mm0 \
/*Converts DCT coefficients in %[dct] from natural order into zig-zag scan
order and stores them in %[qdct].
The index of each output element in the original 64-element array should wind
up in the following 8x8 matrix (the letters indicate the order we compute
each 4-tuple below):
A 0 1 8 16 9 2 3 10 B
C 17 24 32 25 18 11 4 5 D
E 12 19 26 33 40 48 41 34 I
H 27 20 13 6 7 14 21 28 G
K 35 42 49 56 57 50 43 36 J
F 29 22 15 23 30 37 44 51 M
P 58 59 52 45 38 31 39 46 L
N 53 60 61 54 47 55 62 63 O
The order of the coefficients within each tuple is reversed in the comments
below to reflect the usual MSB to LSB notation.*/
#define OC_ZIG_ZAG_MMXEXT \
"movq 0x00(%[dct]),%%mm0\n\t" /*mm0=03 02 01 00*/ \
"movq 0x08(%[dct]),%%mm1\n\t" /*mm1=07 06 05 04*/ \
"movq 0x10(%[dct]),%%mm2\n\t" /*mm2=11 10 09 08*/ \
"movq 0x20(%[dct]),%%mm3\n\t" /*mm3=19 18 17 16*/ \
"movq 0x30(%[dct]),%%mm4\n\t" /*mm4=27 26 25 24*/ \
"movq 0x40(%[dct]),%%mm5\n\t" /*mm5=35 34 33 32*/ \
"movq %%mm2,%%mm7\n\t" /*mm7=11 10 09 08*/ \
"punpcklwd %%mm3,%%mm2\n\t" /*mm2=17 09 16 08*/ \
"movq %%mm0,%%mm6\n\t" /*mm6=03 02 01 00*/ \
"punpckldq %%mm2,%%mm0\n\t" /*mm0=16 08 01 00 *A*/ \
"movq %%mm0,0x00(%[qdct])\n\t" \
"movq 0x18(%[dct]),%%mm0\n\t" /*mm0=15 14 13 12*/ \
"punpckhdq %%mm6,%%mm6\n\t" /*mm6=03 02 03 02*/ \
"psrlq $16,%%mm7\n\t" /*mm7=.. 11 10 09*/ \
"punpckldq %%mm7,%%mm6\n\t" /*mm6=10 09 03 02*/ \
"punpckhwd %%mm7,%%mm3\n\t" /*mm3=.. 19 11 18*/ \
"pshufw $0xD2,%%mm6,%%mm6\n\t" /*mm6=10 03 02 09 *B*/ \
"movq %%mm6,0x08(%[qdct])\n\t" \
"psrlq $48,%%mm2\n\t" /*mm2=.. .. .. 17*/ \
"movq %%mm1,%%mm6\n\t" /*mm6=07 06 05 04*/ \
"punpcklwd %%mm5,%%mm2\n\t" /*mm2=33 .. 32 17*/ \
"movq %%mm3,%%mm7\n\t" /*mm7=.. 19 11 18*/ \
"punpckldq %%mm1,%%mm3\n\t" /*mm3=05 04 11 18 *C*/ \
"por %%mm2,%%mm7\n\t" /*mm7=33 19 ?? ??*/ \
"punpcklwd %%mm4,%%mm2\n\t" /*mm2=25 32 24 17 *D**/ \
"movq %%mm2,0x10(%[qdct])\n\t" \
"movq %%mm3,0x18(%[qdct])\n\t" \
"movq 0x28(%[dct]),%%mm2\n\t" /*mm2=23 22 21 20*/ \
"movq 0x38(%[dct]),%%mm1\n\t" /*mm1=31 30 29 28*/ \
"pshufw $0x9C,%%mm0,%%mm3\n\t" /*mm3=14 13 15 12*/ \
"punpckhdq %%mm7,%%mm7\n\t" /*mm7=33 19 33 19*/ \
"punpckhwd %%mm3,%%mm6\n\t" /*mm6=14 07 13 06*/ \
"punpckldq %%mm0,%%mm0\n\t" /*mm0=13 12 13 12*/ \
"punpcklwd %%mm1,%%mm3\n\t" /*mm3=29 15 28 12*/ \
"punpckhwd %%mm4,%%mm0\n\t" /*mm0=27 13 26 12*/ \
"pshufw $0xB4,%%mm3,%%mm3\n\t" /*mm3=15 29 28 12*/ \
"psrlq $48,%%mm4\n\t" /*mm4=.. .. .. 27*/ \
"punpcklwd %%mm7,%%mm0\n\t" /*mm0=33 26 19 12 *E*/ \
"punpcklwd %%mm1,%%mm4\n\t" /*mm4=29 .. 28 27*/ \
"punpckhwd %%mm2,%%mm3\n\t" /*mm3=23 15 22 29 *F*/ \
"movq %%mm0,0x20(%[qdct])\n\t" \
"movq %%mm3,0x50(%[qdct])\n\t" \
"movq 0x60(%[dct]),%%mm3\n\t" /*mm3=51 50 49 48*/ \
"movq 0x70(%[dct]),%%mm7\n\t" /*mm7=59 58 57 56*/ \
"movq 0x50(%[dct]),%%mm0\n\t" /*mm0=43 42 41 40*/ \
"punpcklwd %%mm4,%%mm2\n\t" /*mm2=28 21 27 20*/ \
"psrlq $32,%%mm5\n\t" /*mm5=.. .. 35 34*/ \
"movq %%mm2,%%mm4\n\t" /*mm4=28 21 27 20*/ \
"punpckldq %%mm6,%%mm2\n\t" /*mm2=13 06 27 20*/ \
"punpckhdq %%mm4,%%mm6\n\t" /*mm6=28 21 14 07 *G*/ \
"movq %%mm3,%%mm4\n\t" /*mm4=51 50 49 48*/ \
"pshufw $0xB1,%%mm2,%%mm2\n\t" /*mm2=06 13 20 27 *H*/ \
"movq %%mm2,0x30(%[qdct])\n\t" \
"movq %%mm6,0x38(%[qdct])\n\t" \
"movq 0x48(%[dct]),%%mm2\n\t" /*mm2=39 38 37 36*/ \
"punpcklwd %%mm5,%%mm4\n\t" /*mm4=35 49 34 48*/ \
"movq 0x58(%[dct]),%%mm5\n\t" /*mm5=47 46 45 44*/ \
"punpckldq %%mm7,%%mm6\n\t" /*mm6=57 56 14 07*/ \
"psrlq $32,%%mm3\n\t" /*mm3=.. .. 51 50*/ \
"punpckhwd %%mm0,%%mm6\n\t" /*mm6=43 57 42 56*/ \
"punpcklwd %%mm4,%%mm0\n\t" /*mm0=34 41 48 40 *I*/ \
"pshufw $0x4E,%%mm6,%%mm6\n\t" /*mm6=42 56 43 57*/ \
"movq %%mm0,0x28(%[qdct])\n\t" \
"punpcklwd %%mm2,%%mm3\n\t" /*mm3=37 51 36 50*/ \
"punpckhwd %%mm6,%%mm4\n\t" /*mm4=42 35 56 49*/ \
"punpcklwd %%mm3,%%mm6\n\t" /*mm6=36 43 50 57 *J*/ \
"pshufw $0x4E,%%mm4,%%mm4\n\t" /*mm4=56 49 42 35 *K*/ \
"movq %%mm4,0x40(%[qdct])\n\t" \
"movq %%mm6,0x48(%[qdct])\n\t" \
"movq 0x68(%[dct]),%%mm6\n\t" /*mm6=55 54 53 52*/ \
"movq 0x78(%[dct]),%%mm0\n\t" /*mm0=63 62 61 60*/ \
"psrlq $32,%%mm1\n\t" /*mm1=.. .. 31 30*/ \
"pshufw $0xD8,%%mm5,%%mm5\n\t" /*mm5=47 45 46 44*/ \
"pshufw $0x0B,%%mm3,%%mm3\n\t" /*mm3=50 50 51 37*/ \
"punpcklwd %%mm5,%%mm1\n\t" /*mm1=46 31 44 30*/ \
"pshufw $0xC9,%%mm6,%%mm6\n\t" /*mm6=55 52 54 53*/ \
"punpckhwd %%mm1,%%mm2\n\t" /*mm2=46 39 31 38 *L*/ \
"punpcklwd %%mm3,%%mm1\n\t" /*mm1=51 44 37 30 *M*/ \
"movq %%mm2,0x68(%[qdct])\n\t" \
"movq %%mm1,0x58(%[qdct])\n\t" \
"punpckhwd %%mm6,%%mm5\n\t" /*mm5=55 47 52 45*/ \
"punpckldq %%mm0,%%mm6\n\t" /*mm6=61 60 54 53*/ \
"pshufw $0x10,%%mm5,%%mm4\n\t" /*mm4=45 52 45 45*/ \
"pshufw $0x78,%%mm6,%%mm6\n\t" /*mm6=53 60 61 54 *N*/ \
"punpckhdq %%mm0,%%mm5\n\t" /*mm5=63 62 55 47 *O*/ \
"punpckhdq %%mm4,%%mm7\n\t" /*mm7=45 52 59 58 *P*/ \
"movq %%mm6,0x70(%[qdct])\n\t" \
"movq %%mm5,0x78(%[qdct])\n\t" \
"movq %%mm7,0x60(%[qdct])\n\t" \
#endif
|