Mercurial > libavcodec.hg
comparison bfin/mpegvideo_bfin.c @ 5131:757efe20a3c8 libavcodec
Blackfin dct_quantize_bfin routine
2x performance boost in performing quantization for mpeg encoding
author | mhoffman |
---|---|
date | Tue, 12 Jun 2007 19:13:50 +0000 |
parents | |
children | f7cbb7733146 |
comparison
equal
deleted
inserted
replaced
5130:bd2133a6973c | 5131:757efe20a3c8 |
---|---|
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 02110-1301 USA | |
21 */ | |
22 | |
23 #include "dsputil.h" | |
24 #include "mpegvideo.h" | |
25 #include "avcodec.h" | |
26 #include "dsputil_bfin.h" | |
27 | |
28 | |
29 extern void ff_bfin_fdct (DCTELEM *block) attribute_l1_text; | |
30 | |
31 | |
32 static int dct_quantize_bfin (MpegEncContext *s, | |
33 DCTELEM *block, int n, | |
34 int qscale, int *overflow) | |
35 { | |
36 int last_non_zero, q, start_i; | |
37 const short *qmat; | |
38 short *bias; | |
39 const uint8_t *scantable= s->intra_scantable.scantable; | |
40 short dc; | |
41 int max=0; | |
42 | |
43 PROF("fdct",0); | |
44 ff_bfin_fdct (block); | |
45 EPROF(); | |
46 | |
47 PROF("denoise",1); | |
48 if(s->dct_error_sum) | |
49 s->denoise_dct(s, block); | |
50 EPROF(); | |
51 | |
52 PROF("quant-init",2); | |
53 if (s->mb_intra) { | |
54 if (!s->h263_aic) { | |
55 if (n < 4) | |
56 q = s->y_dc_scale; | |
57 else | |
58 q = s->c_dc_scale; | |
59 q = q << 3; | |
60 } else | |
61 /* For AIC we skip quant/dequant of INTRADC */ | |
62 q = 1 << 3; | |
63 | |
64 /* note: block[0] is assumed to be positive */ | |
65 dc = block[0] = (block[0] + (q >> 1)) / q; | |
66 start_i = 1; | |
67 last_non_zero = 0; | |
68 bias = s->q_intra_matrix16[qscale][1]; | |
69 qmat = s->q_intra_matrix16[qscale][0]; | |
70 | |
71 } else { | |
72 start_i = 0; | |
73 last_non_zero = -1; | |
74 bias = s->q_inter_matrix16[qscale][1]; | |
75 qmat = s->q_inter_matrix16[qscale][0]; | |
76 | |
77 } | |
78 EPROF(); | |
79 | |
80 PROF("quantize",4); | |
81 | |
82 /* for(i=start_i; i<64; i++) { */ | |
83 /* sign = (block[i]>>15)|1; */ | |
84 /* level = ((abs(block[i])+bias[0])*qmat[i])>>16; */ | |
85 /* if (level < 0) level = 0; */ | |
86 /* max |= level; */ | |
87 /* level = level * sign; */ | |
88 /* block[i] = level; */ | |
89 /* } */ | |
90 | |
91 asm volatile | |
92 ("i2=%1;\n\t" | |
93 "r1=[%1++]; \n\t" | |
94 "r0=r1>>>15 (v); \n\t" | |
95 "lsetup (0f,1f) lc0=%3; \n\t" | |
96 "0: r0=r0|%4; \n\t" | |
97 " r1=abs r1 (v) || r2=[%2++];\n\t" | |
98 " r1=r1+|+%5; \n\t" | |
99 " r1=max(r1,%6) (v); \n\t" | |
100 " r1.h=(a1 =r1.h*r2.h), r1.l=(a0 =r1.l*r2.l) (tfu); \n\t" | |
101 " %0=%0|r1; \n\t" | |
102 " r0.h=(a1 =r1.h*r0.h), r0.l=(a0 =r1.l*r0.l) (is) || r1=[%1++];\n\t" | |
103 "1: r0=r1>>>15 (v) || [i2++]=r0;\n\t" | |
104 "r1=%0>>16; \n\t" | |
105 "%0=%0|r1; \n\t" | |
106 "%0.h=0; \n\t" | |
107 : "=&d" (max) | |
108 : "b" (block), "b" (qmat), "a" (32), "d" (0x00010001), "d" (bias[0]*0x10001), "d" (0) | |
109 : "R0","R1","R2", "I2"); | |
110 if (start_i == 1) block[0] = dc; | |
111 | |
112 EPROF(); | |
113 | |
114 | |
115 PROF("zzscan",5); | |
116 | |
117 asm volatile | |
118 ("r0=b[%1--] (x); \n\t" | |
119 "lsetup (0f,1f) lc0=%3; \n\t" /* for(i=63; i>=start_i; i--) { */ | |
120 "0: p0=r0; \n\t" /* j = scantable[i]; */ | |
121 " p0=%2+(p0<<1); \n\t" /* if (block[j]) { */ | |
122 " r0=w[p0]; \n\t" /* last_non_zero = i; */ | |
123 " cc=r0==0; \n\t" /* break; */ | |
124 " if !cc jump 2f; \n\t" /* } */ | |
125 "1: r0=b[%1--] (x); \n\t" /* } */ | |
126 " %0=%4; \n\t" | |
127 " jump 3f; \n\t" | |
128 "2: %0=lc0; \n\t" | |
129 "3:\n\t" | |
130 | |
131 : "=d" (last_non_zero) | |
132 : "a" (scantable+63), "a" (block), "a" (63), "d" (last_non_zero) | |
133 : "P0","R0"); | |
134 | |
135 EPROF(); | |
136 | |
137 *overflow= s->max_qcoeff < max; //overflow might have happened | |
138 | |
139 bfprof(); | |
140 | |
141 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */ | |
142 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM) | |
143 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero); | |
144 | |
145 return last_non_zero; | |
146 } | |
147 | |
148 void MPV_common_init_bfin (MpegEncContext *s) | |
149 { | |
150 s->dct_quantize= dct_quantize_bfin; | |
151 } | |
152 |