Mercurial > libavcodec.hg
annotate faandct.c @ 11560:8a4984c5cacc libavcodec
Define AVMediaType enum, and use it instead of enum CodecType, which
is deprecated and will be dropped at the next major bump.
author | stefano |
---|---|
date | Tue, 30 Mar 2010 23:30:55 +0000 |
parents | e9d9d946f213 |
children | 7dd2a45249a9 |
rev | line source |
---|---|
1557 | 1 /* |
2 * Floating point AAN DCT | |
5214 | 3 * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c) |
4 * | |
7821 | 5 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> |
6 * Copyright (c) 2003 Roman Shaposhnik | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
7 * |
7821 | 8 * Permission to use, copy, modify, and/or distribute this software for any |
9 * purpose with or without fee is hereby granted, provided that the above | |
10 * copyright notice and this permission notice appear in all copies. | |
1557 | 11 * |
7821 | 12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF | |
14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR | |
15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES | |
16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN | |
17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF | |
18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. | |
1557 | 19 */ |
20 | |
21 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
7821
diff
changeset
|
22 * @file libavcodec/faandct.c |
2967 | 23 * @brief |
1557 | 24 * Floating point AAN DCT |
25 * @author Michael Niedermayer <michaelni@gmx.at> | |
26 */ | |
27 | |
28 #include "dsputil.h" | |
29 #include "faandct.h" | |
30 | |
31 #define FLOAT float | |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1561
diff
changeset
|
32 #ifdef FAAN_POSTSCALE |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1561
diff
changeset
|
33 # define SCALE(x) postscale[x] |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1561
diff
changeset
|
34 #else |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1561
diff
changeset
|
35 # define SCALE(x) 1 |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1561
diff
changeset
|
36 #endif |
1557 | 37 |
38 //numbers generated by simple c code (not as accurate as they could be) | |
39 /* | |
40 for(i=0; i<8; i++){ | |
41 printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2))); | |
42 } | |
43 */ | |
44 #define B0 1.00000000000000000000 | |
45 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1 | |
46 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1 | |
47 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1 | |
48 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1 | |
49 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1 | |
50 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1 | |
51 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1 | |
52 | |
53 | |
54 #define A1 0.70710678118654752438 // cos(pi*4/16) | |
55 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2) | |
56 #define A5 0.38268343236508977170 // cos(pi*6/16) | |
57 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2) | |
58 | |
7129 | 59 static const FLOAT postscale[64]={ |
1557 | 60 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7, |
61 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7, | |
62 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7, | |
63 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7, | |
64 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7, | |
65 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7, | |
66 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7, | |
67 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7, | |
68 }; | |
69 | |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
3947
diff
changeset
|
70 static av_always_inline void row_fdct(FLOAT temp[64], DCTELEM * data) |
1557 | 71 { |
72 FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; | |
73 FLOAT tmp10, tmp11, tmp12, tmp13; | |
6471 | 74 FLOAT z2, z4, z11, z13; |
75 FLOAT av_unused z5; | |
1557 | 76 int i; |
77 | |
78 for (i=0; i<8*8; i+=8) { | |
79 tmp0= data[0 + i] + data[7 + i]; | |
80 tmp7= data[0 + i] - data[7 + i]; | |
81 tmp1= data[1 + i] + data[6 + i]; | |
82 tmp6= data[1 + i] - data[6 + i]; | |
83 tmp2= data[2 + i] + data[5 + i]; | |
84 tmp5= data[2 + i] - data[5 + i]; | |
85 tmp3= data[3 + i] + data[4 + i]; | |
86 tmp4= data[3 + i] - data[4 + i]; | |
2967 | 87 |
1557 | 88 tmp10= tmp0 + tmp3; |
89 tmp13= tmp0 - tmp3; | |
90 tmp11= tmp1 + tmp2; | |
91 tmp12= tmp1 - tmp2; | |
2967 | 92 |
1557 | 93 temp[0 + i]= tmp10 + tmp11; |
94 temp[4 + i]= tmp10 - tmp11; | |
2967 | 95 |
6408
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
96 tmp12 += tmp13; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
97 tmp12 *= A1; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
98 temp[2 + i]= tmp13 + tmp12; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
99 temp[6 + i]= tmp13 - tmp12; |
2967 | 100 |
6409
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
101 tmp4 += tmp5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
102 tmp5 += tmp6; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
103 tmp6 += tmp7; |
1557 | 104 |
6411
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
105 #if 0 |
6409
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
106 z5= (tmp4 - tmp6) * A5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
107 z2= tmp4*A2 + z5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
108 z4= tmp6*A4 + z5; |
6411
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
109 #else |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
110 z2= tmp4*(A2+A5) - tmp6*A5; |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
111 z4= tmp6*(A4-A5) + tmp4*A5; |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
112 #endif |
6410 | 113 tmp5*=A1; |
1557 | 114 |
6410 | 115 z11= tmp7 + tmp5; |
116 z13= tmp7 - tmp5; | |
1557 | 117 |
118 temp[5 + i]= z13 + z2; | |
119 temp[3 + i]= z13 - z2; | |
120 temp[1 + i]= z11 + z4; | |
121 temp[7 + i]= z11 - z4; | |
2967 | 122 } |
1589 | 123 } |
124 | |
125 void ff_faandct(DCTELEM * data) | |
126 { | |
127 FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; | |
128 FLOAT tmp10, tmp11, tmp12, tmp13; | |
6471 | 129 FLOAT z2, z4, z11, z13; |
130 FLOAT av_unused z5; | |
1589 | 131 FLOAT temp[64]; |
132 int i; | |
133 | |
134 emms_c(); | |
135 | |
136 row_fdct(temp, data); | |
1557 | 137 |
138 for (i=0; i<8; i++) { | |
139 tmp0= temp[8*0 + i] + temp[8*7 + i]; | |
140 tmp7= temp[8*0 + i] - temp[8*7 + i]; | |
141 tmp1= temp[8*1 + i] + temp[8*6 + i]; | |
142 tmp6= temp[8*1 + i] - temp[8*6 + i]; | |
143 tmp2= temp[8*2 + i] + temp[8*5 + i]; | |
144 tmp5= temp[8*2 + i] - temp[8*5 + i]; | |
145 tmp3= temp[8*3 + i] + temp[8*4 + i]; | |
146 tmp4= temp[8*3 + i] - temp[8*4 + i]; | |
2967 | 147 |
1557 | 148 tmp10= tmp0 + tmp3; |
149 tmp13= tmp0 - tmp3; | |
150 tmp11= tmp1 + tmp2; | |
151 tmp12= tmp1 - tmp2; | |
2967 | 152 |
1563
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
153 data[8*0 + i]= lrintf(SCALE(8*0 + i) * (tmp10 + tmp11)); |
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
154 data[8*4 + i]= lrintf(SCALE(8*4 + i) * (tmp10 - tmp11)); |
2967 | 155 |
6408
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
156 tmp12 += tmp13; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
157 tmp12 *= A1; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
158 data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + tmp12)); |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
159 data[8*6 + i]= lrintf(SCALE(8*6 + i) * (tmp13 - tmp12)); |
2967 | 160 |
6409
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
161 tmp4 += tmp5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
162 tmp5 += tmp6; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
163 tmp6 += tmp7; |
1557 | 164 |
6411
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
165 #if 0 |
6409
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
166 z5= (tmp4 - tmp6) * A5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
167 z2= tmp4*A2 + z5; |
813ff53ed933
Avoid a=b+c but rather use a+=b (gcc is too stupid to do this itself),
michael
parents:
6408
diff
changeset
|
168 z4= tmp6*A4 + z5; |
6411
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
169 #else |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
170 z2= tmp4*(A2+A5) - tmp6*A5; |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
171 z4= tmp6*(A4-A5) + tmp4*A5; |
5d50e8a7eb09
Alternative and faster (on gcc x86) plane rotation.
michael
parents:
6410
diff
changeset
|
172 #endif |
6410 | 173 tmp5*=A1; |
1557 | 174 |
6410 | 175 z11= tmp7 + tmp5; |
176 z13= tmp7 - tmp5; | |
1557 | 177 |
1563
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
178 data[8*5 + i]= lrintf(SCALE(8*5 + i) * (z13 + z2)); |
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
179 data[8*3 + i]= lrintf(SCALE(8*3 + i) * (z13 - z2)); |
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
180 data[8*1 + i]= lrintf(SCALE(8*1 + i) * (z11 + z4)); |
820e06c6ca9b
4.9-RC FreeBSD doesnt like lrint() so change to lrintf(), not that bsd supports that but we emulate it ...
michael
parents:
1562
diff
changeset
|
181 data[8*7 + i]= lrintf(SCALE(8*7 + i) * (z11 - z4)); |
1557 | 182 } |
183 } | |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
184 |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
185 void ff_faandct248(DCTELEM * data) |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
186 { |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
187 FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
188 FLOAT tmp10, tmp11, tmp12, tmp13; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
189 FLOAT temp[64]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
190 int i; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
191 |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
192 emms_c(); |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
193 |
1589 | 194 row_fdct(temp, data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
195 |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
196 for (i=0; i<8; i++) { |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
197 tmp0 = temp[8*0 + i] + temp[8*1 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
198 tmp1 = temp[8*2 + i] + temp[8*3 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
199 tmp2 = temp[8*4 + i] + temp[8*5 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
200 tmp3 = temp[8*6 + i] + temp[8*7 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
201 tmp4 = temp[8*0 + i] - temp[8*1 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
202 tmp5 = temp[8*2 + i] - temp[8*3 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
203 tmp6 = temp[8*4 + i] - temp[8*5 + i]; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
204 tmp7 = temp[8*6 + i] - temp[8*7 + i]; |
2967 | 205 |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
206 tmp10 = tmp0 + tmp3; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
207 tmp11 = tmp1 + tmp2; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
208 tmp12 = tmp1 - tmp2; |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
209 tmp13 = tmp0 - tmp3; |
2967 | 210 |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
211 data[8*0 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11)); |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
212 data[8*4 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11)); |
2967 | 213 |
6408
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
214 tmp12 += tmp13; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
215 tmp12 *= A1; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
216 data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12)); |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
217 data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12)); |
2967 | 218 |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
219 tmp10 = tmp4 + tmp7; |
2979 | 220 tmp11 = tmp5 + tmp6; |
221 tmp12 = tmp5 - tmp6; | |
222 tmp13 = tmp4 - tmp7; | |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
223 |
2979 | 224 data[8*1 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11)); |
225 data[8*5 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11)); | |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
226 |
6408
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
227 tmp12 += tmp13; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
228 tmp12 *= A1; |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
229 data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12)); |
7af599600f2d
One variable less. No real change to the actual code.
michael
parents:
6398
diff
changeset
|
230 data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12)); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
231 } |
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1563
diff
changeset
|
232 } |