ffmpeg / libavcodec / bfin / mpegvideo_bfin.c @ 08ec1397
History  View  Annotate  Download (5.36 KB)
1 
/*


2 
* BlackFin MPEGVIDEO OPTIMIZATIONS

3 
*

4 
* Copyright (C) 2007 Marc Hoffman <mmh@pleasantst.com>

5 
*

6 
* This file is part of FFmpeg.

7 
*

8 
* FFmpeg is free software; you can redistribute it and/or

9 
* modify it under the terms of the GNU Lesser General Public

10 
* License as published by the Free Software Foundation; either

11 
* version 2.1 of the License, or (at your option) any later version.

12 
*

13 
* FFmpeg is distributed in the hope that it will be useful,

14 
* but WITHOUT ANY WARRANTY; without even the implied warranty of

15 
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU

16 
* Lesser General Public License for more details.

17 
*

18 
* You should have received a copy of the GNU Lesser General Public

19 
* License along with FFmpeg; if not, write to the Free Software

20 
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 021101301 USA

21 
*/

22  
23 
#include "libavcodec/avcodec.h" 
24 
#include "libavcodec/dsputil.h" 
25 
#include "libavcodec/mpegvideo.h" 
26 
#include "dsputil_bfin.h" 
27  
28 
static int dct_quantize_bfin (MpegEncContext *s, 
29 
DCTELEM *block, int n,

30 
int qscale, int *overflow) 
31 
{ 
32 
int last_non_zero, q, start_i;

33 
const short *qmat; 
34 
short *bias;

35 
const uint8_t *scantable= s>intra_scantable.scantable;

36 
short dc;

37 
int max=0; 
38  
39 
PROF("fdct",0); 
40 
s>dsp.fdct(block); 
41 
EPROF(); 
42  
43 
PROF("denoise",1); 
44 
if(s>dct_error_sum)

45 
s>denoise_dct(s, block); 
46 
EPROF(); 
47  
48 
PROF("quantinit",2); 
49 
if (s>mb_intra) {

50 
if (!s>h263_aic) {

51 
if (n < 4) 
52 
q = s>y_dc_scale; 
53 
else

54 
q = s>c_dc_scale; 
55 
q = q << 3;

56 
} else

57 
/* For AIC we skip quant/dequant of INTRADC */

58 
q = 1 << 3; 
59  
60 
/* note: block[0] is assumed to be positive */

61 
dc = block[0] = (block[0] + (q >> 1)) / q; 
62 
start_i = 1;

63 
last_non_zero = 0;

64 
bias = s>q_intra_matrix16[qscale][1];

65 
qmat = s>q_intra_matrix16[qscale][0];

66  
67 
} else {

68 
start_i = 0;

69 
last_non_zero = 1;

70 
bias = s>q_inter_matrix16[qscale][1];

71 
qmat = s>q_inter_matrix16[qscale][0];

72  
73 
} 
74 
EPROF(); 
75  
76 
PROF("quantize",4); 
77  
78 
/* for(i=start_i; i<64; i++) { */

79 
/* sign = (block[i]>>15)1; */

80 
/* level = ((abs(block[i])+bias[0])*qmat[i])>>16; */

81 
/* if (level < 0) level = 0; */

82 
/* max = level; */

83 
/* level = level * sign; */

84 
/* block[i] = level; */

85 
/* } */

86  
87 
__asm__ volatile

88 
("i2=%1;\n\t"

89 
"r1=[%1++]; \n\t"

90 
"r0=r1>>>15 (v); \n\t"

91 
"lsetup (0f,1f) lc0=%3; \n\t"

92 
"0: r0=r0%4; \n\t"

93 
" r1=abs r1 (v)  r2=[%2++];\n\t"

94 
" r1=r1++%5; \n\t"

95 
" r1=max(r1,%6) (v); \n\t"

96 
" r1.h=(a1 =r1.h*r2.h), r1.l=(a0 =r1.l*r2.l) (tfu); \n\t"

97 
" %0=%0r1; \n\t"

98 
" r0.h=(a1 =r1.h*r0.h), r0.l=(a0 =r1.l*r0.l) (is)  r1=[%1++];\n\t"

99 
"1: r0=r1>>>15 (v)  [i2++]=r0;\n\t"

100 
"r1=%0>>16; \n\t"

101 
"%0=%0r1; \n\t"

102 
"%0.h=0; \n\t"

103 
: "=&d" (max)

104 
: "b" (block), "b" (qmat), "a" (32), "d" (0x00010001), "d" (bias[0]*0x10001), "d" (0) 
105 
: "R0","R1","R2", "I2"); 
106 
if (start_i == 1) block[0] = dc; 
107  
108 
EPROF(); 
109  
110  
111 
PROF("zzscan",5); 
112  
113 
__asm__ volatile

114 
("r0=b[%1] (x); \n\t"

115 
"lsetup (0f,1f) lc0=%3; \n\t" /* for(i=63; i>=start_i; i) { */ 
116 
"0: p0=r0; \n\t" /* j = scantable[i]; */ 
117 
" p0=%2+(p0<<1); \n\t" /* if (block[j]) { */ 
118 
" r0=w[p0]; \n\t" /* last_non_zero = i; */ 
119 
" cc=r0==0; \n\t" /* break; */ 
120 
" if !cc jump 2f; \n\t" /* } */ 
121 
"1: r0=b[%1] (x); \n\t" /* } */ 
122 
" %0=%4; \n\t"

123 
" jump 3f; \n\t"

124 
"2: %0=lc0; \n\t"

125 
"3:\n\t"

126  
127 
: "=d" (last_non_zero)

128 
: "a" (scantable+63), "a" (block), "a" (63), "d" (last_non_zero) 
129 
: "P0","R0"); 
130  
131 
EPROF(); 
132  
133 
*overflow= s>max_qcoeff < max; //overflow might have happened

134  
135 
bfprof(); 
136  
137 
/* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */

138 
if (s>dsp.idct_permutation_type != FF_NO_IDCT_PERM)

139 
ff_block_permute(block, s>dsp.idct_permutation, scantable, last_non_zero); 
140  
141 
return last_non_zero;

142 
} 
143  
144 
void MPV_common_init_bfin (MpegEncContext *s)

145 
{ 
146 
s>dct_quantize= dct_quantize_bfin; 
147 
} 
148 