summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMichael Giacomelli <giac2000@hotmail.com>2009-05-10 23:18:04 +0000
committerMichael Giacomelli <giac2000@hotmail.com>2009-05-10 23:18:04 +0000
commit65649de515257493dd1b8db599201097208677bb (patch)
tree23dcc4b3ae9c156568d29f90dc5fd0f2f951b21d
parentfc28cb4ed5adf4a0bc548af38ca6de95bbf027e5 (diff)
downloadrockbox-65649de515257493dd1b8db599201097208677bb.tar.gz
rockbox-65649de515257493dd1b8db599201097208677bb.zip
Patch by Mohamed Tarek from FS #10182. Remove floating point code (FFT, MDCT, etc) from libcook.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@20902 a1c6a512-1295-4272-9138-f99709370657
Diffstat
-rw-r--r--apps/codecs/libcook/Makefile.test2
-rw-r--r--apps/codecs/libcook/cook.c267
-rw-r--r--apps/codecs/libcook/cook.h17
-rw-r--r--apps/codecs/libcook/dsputil.h908
-rw-r--r--apps/codecs/libcook/fft.c374
-rw-r--r--apps/codecs/libcook/main.c2
-rw-r--r--apps/codecs/libcook/mdct.c229
7 files changed, 2 insertions, 1797 deletions
diff --git a/apps/codecs/libcook/Makefile.test b/apps/codecs/libcook/Makefile.test
index 5b60ca3ef5..d992eceeb8 100644
--- a/apps/codecs/libcook/Makefile.test
+++ b/apps/codecs/libcook/Makefile.test
@@ -1,5 +1,5 @@
1CFLAGS = -Wall -O3 1CFLAGS = -Wall -O3
2OBJS = main.o bitstream.o cook.o fft.o libavutil/log.o mdct.o libavutil/mem.o libavutil/lfg.o libavutil/md5.o rm2wav.o 2OBJS = main.o bitstream.o cook.o libavutil/log.o libavutil/mem.o libavutil/lfg.o libavutil/md5.o rm2wav.o
3cooktest: $(OBJS) 3cooktest: $(OBJS)
4 gcc -o cooktest $(OBJS) -lm 4 gcc -o cooktest $(OBJS) -lm
5 5
diff --git a/apps/codecs/libcook/cook.c b/apps/codecs/libcook/cook.c
index 8bb3b5a113..bd72179cbd 100644
--- a/apps/codecs/libcook/cook.c
+++ b/apps/codecs/libcook/cook.c
@@ -71,23 +71,11 @@ const uint8_t ff_log2_tab[256]={
71#define MAX_SUBPACKETS 5 71#define MAX_SUBPACKETS 5
72//#define COOKDEBUG 72//#define COOKDEBUG
73#define DEBUGF(message,args ...) av_log(NULL,AV_LOG_ERROR,message,## args) 73#define DEBUGF(message,args ...) av_log(NULL,AV_LOG_ERROR,message,## args)
74
75static float pow2tab[127];
76static float rootpow2tab[127];
77#include "cook_fixpoint.h" 74#include "cook_fixpoint.h"
78 75
79/* debug functions */ 76/* debug functions */
80 77
81#ifdef COOKDEBUG 78#ifdef COOKDEBUG
82static void dump_float_table(float* table, int size, int delimiter) {
83 int i=0;
84 av_log(NULL,AV_LOG_ERROR,"\n[%d]: ",i);
85 for (i=0 ; i<size ; i++) {
86 av_log(NULL, AV_LOG_ERROR, "%5.1f, ", table[i]);
87 if ((i+1)%delimiter == 0) av_log(NULL,AV_LOG_ERROR,"\n[%d]: ",i+1);
88 }
89}
90
91static void dump_int_table(int* table, int size, int delimiter) { 79static void dump_int_table(int* table, int size, int delimiter) {
92 int i=0; 80 int i=0;
93 av_log(NULL,AV_LOG_ERROR,"\n[%d]: ",i); 81 av_log(NULL,AV_LOG_ERROR,"\n[%d]: ",i);
@@ -109,27 +97,6 @@ static void dump_short_table(short* table, int size, int delimiter) {
109#endif 97#endif
110 98
111/*************** init functions ***************/ 99/*************** init functions ***************/
112
113/* table generator */
114static av_cold void init_pow2table(void){
115 int i;
116 for (i=-63 ; i<64 ; i++){
117 pow2tab[63+i]= pow(2, i);
118 rootpow2tab[63+i]=sqrt(pow(2, i));
119 }
120}
121
122/* table generator */
123static av_cold void init_gain_table(COOKContext *q) {
124 int i;
125 q->gain_size_factor = q->samples_per_channel/8;
126 for (i=0 ; i<23 ; i++) {
127 q->gain_table[i] = pow(pow2tab[i+52] ,
128 (1.0/(double)q->gain_size_factor));
129 }
130}
131
132
133static av_cold int init_cook_vlc_tables(COOKContext *q) { 100static av_cold int init_cook_vlc_tables(COOKContext *q) {
134 int i, result; 101 int i, result;
135 102
@@ -156,42 +123,6 @@ static av_cold int init_cook_vlc_tables(COOKContext *q) {
156 av_log(NULL,AV_LOG_ERROR,"VLC tables initialized. Result = %d\n",result); 123 av_log(NULL,AV_LOG_ERROR,"VLC tables initialized. Result = %d\n",result);
157 return result; 124 return result;
158} 125}
159
160static av_cold int init_cook_mlt(COOKContext *q) {
161 int j;
162 int mlt_size = q->samples_per_channel;
163
164 if ((q->mlt_window = av_malloc(sizeof(float)*mlt_size)) == 0)
165 return -1;
166
167 /* Initialize the MLT window: simple sine window. */
168 ff_sine_window_init(q->mlt_window, mlt_size);
169 for(j=0 ; j<mlt_size ; j++)
170 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
171
172 /* Initialize the MDCT. */
173 if (ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size)+1, 1)) {
174 av_free(q->mlt_window);
175 return -1;
176 }
177 av_log(NULL,AV_LOG_ERROR,"MDCT initialized, order = %d. mlt_window = %d\n",
178 av_log2(mlt_size)+1,sizeof(q->mlt_window)*mlt_size);
179
180 return 0;
181}
182
183static const float *maybe_reformat_buffer32 (COOKContext *q, const float *ptr, int n)
184{
185 if (1)
186 return ptr;
187}
188
189static av_cold void init_cplscales_table (COOKContext *q) {
190 int i;
191 for (i=0;i<5;i++)
192 q->cplscales[i] = maybe_reformat_buffer32 (q, q->cplscales[i], (1<<(i+2))-1);
193}
194
195/*************** init functions end ***********/ 126/*************** init functions end ***********/
196 127
197/** 128/**
@@ -249,12 +180,8 @@ av_cold int cook_decode_close(COOKContext *q)
249 av_log(NULL,AV_LOG_ERROR, "Deallocating memory.\n"); 180 av_log(NULL,AV_LOG_ERROR, "Deallocating memory.\n");
250 181
251 /* Free allocated memory buffers. */ 182 /* Free allocated memory buffers. */
252 av_free(q->mlt_window);
253 av_free(q->decoded_bytes_buffer); 183 av_free(q->decoded_bytes_buffer);
254 184
255 /* Free the transform. */
256 ff_mdct_end(&q->mdct_ctx);
257
258 /* Free the VLC tables. */ 185 /* Free the VLC tables. */
259 for (i=0 ; i<13 ; i++) { 186 for (i=0 ; i<13 ; i++) {
260 free_vlc(&q->envelope_quant_index[i]); 187 free_vlc(&q->envelope_quant_index[i]);
@@ -444,37 +371,6 @@ static inline void expand_category(COOKContext *q, int* category,
444} 371}
445 372
446/** 373/**
447 * The real requantization of the mltcoefs
448 *
449 * @param q pointer to the COOKContext
450 * @param index index
451 * @param quant_index quantisation index
452 * @param subband_coef_index array of indexes to quant_centroid_tab
453 * @param subband_coef_sign signs of coefficients
454 * @param mlt_p pointer into the mlt buffer
455 */
456
457#if 0
458static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
459 int* subband_coef_index, int* subband_coef_sign,
460 float* mlt_p){
461 int i;
462 float f1;
463
464 for(i=0 ; i<SUBBAND_SIZE ; i++) {
465 if (subband_coef_index[i]) {
466 f1 = quant_centroid_tab[index][subband_coef_index[i]];
467 if (subband_coef_sign[i]) f1 = -f1;
468 } else {
469 /* noise coding if subband_coef_index[i] == 0 */
470 f1 = dither_tab[index];
471 if (av_lfg_get(&q->random_state) < 0x80000000) f1 = -f1;
472 }
473 mlt_p[i] = f1 * rootpow2tab[quant_index+63];
474 }
475}
476#endif
477/**
478 * Unpack the subband_coef_index and subband_coef_sign vectors. 374 * Unpack the subband_coef_index and subband_coef_sign vectors.
479 * 375 *
480 * @param q pointer to the COOKContext 376 * @param q pointer to the COOKContext
@@ -585,102 +481,6 @@ static void mono_decode(COOKContext *q, REAL_T* mlt_buffer) {
585 decode_vectors(q, category, quant_index_table, mlt_buffer); 481 decode_vectors(q, category, quant_index_table, mlt_buffer);
586} 482}
587 483
588
589/**
590 * the actual requantization of the timedomain samples
591 *
592 * @param q pointer to the COOKContext
593 * @param buffer pointer to the timedomain buffer
594 * @param gain_index index for the block multiplier
595 * @param gain_index_next index for the next block multiplier
596 */
597
598#if 0
599static void interpolate_float(COOKContext *q, float* buffer,
600 int gain_index, int gain_index_next){
601 int i;
602 float fc1, fc2;
603 fc1 = pow2tab[gain_index+63];
604
605 if(gain_index == gain_index_next){ //static gain
606 for(i=0 ; i<q->gain_size_factor ; i++){
607 buffer[i]*=fc1;
608 }
609 return;
610 } else { //smooth gain
611 fc2 = q->gain_table[11 + (gain_index_next-gain_index)];
612 for(i=0 ; i<q->gain_size_factor ; i++){
613 buffer[i]*=fc1;
614 fc1*=fc2;
615 }
616 return;
617 }
618}
619#endif
620
621/**
622 * Apply transform window, overlap buffers.
623 *
624 * @param q pointer to the COOKContext
625 * @param inbuffer pointer to the mltcoefficients
626 * @param gains_ptr current and previous gains
627 * @param previous_buffer pointer to the previous buffer to be used for overlapping
628 */
629
630static void imlt_window_float (COOKContext *q, float *buffer1,
631 cook_gains *gains_ptr, float *previous_buffer)
632{
633 const float fc = pow2tab[gains_ptr->previous[0] + 63];
634 int i;
635 /* The weird thing here, is that the two halves of the time domain
636 * buffer are swapped. Also, the newest data, that we save away for
637 * next frame, has the wrong sign. Hence the subtraction below.
638 * Almost sounds like a complex conjugate/reverse data/FFT effect.
639 */
640
641 /* Apply window and overlap */
642 for(i = 0; i < q->samples_per_channel; i++){
643 buffer1[i] = buffer1[i] * fc * q->mlt_window[i] -
644 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
645 }
646}
647
648/**
649 * The modulated lapped transform, this takes transform coefficients
650 * and transforms them into timedomain samples.
651 * Apply transform window, overlap buffers, apply gain profile
652 * and buffer management.
653 *
654 * @param q pointer to the COOKContext
655 * @param inbuffer pointer to the mltcoefficients
656 * @param gains_ptr current and previous gains
657 * @param previous_buffer pointer to the previous buffer to be used for overlapping
658 */
659#if 0
660static void imlt_gain(COOKContext *q, REAL_T *inbuffer,
661 cook_gains *gains_ptr, REAL_T* previous_buffer)
662{
663 REAL_T *buffer0 = q->mono_mdct_output;
664 REAL_T *buffer1 = q->mono_mdct_output + q->samples_per_channel;
665 int i;
666
667 /* Inverse modified discrete cosine transform */
668 ff_imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
669
670 q->imlt_window (q, buffer1, gains_ptr, previous_buffer);
671
672 /* Apply gain profile */
673 for (i = 0; i < 8; i++) {
674 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
675 q->interpolate(q, &buffer1[q->gain_size_factor * i],
676 gains_ptr->now[i], gains_ptr->now[i + 1]);
677 }
678
679 /* Save away the current to be previous block. */
680 memcpy(previous_buffer, buffer0, sizeof(float)*q->samples_per_channel);
681}
682
683#endif
684/** 484/**
685 * function for getting the jointstereo coupling information 485 * function for getting the jointstereo coupling information
686 * 486 *
@@ -711,31 +511,6 @@ static void decouple_info(COOKContext *q, int* decouple_tab){
711 return; 511 return;
712} 512}
713 513
714/*
715 * function decouples a pair of signals from a single signal via multiplication.
716 *
717 * @param q pointer to the COOKContext
718 * @param subband index of the current subband
719 * @param f1 multiplier for channel 1 extraction
720 * @param f2 multiplier for channel 2 extraction
721 * @param decode_buffer input buffer
722 * @param mlt_buffer1 pointer to left channel mlt coefficients
723 * @param mlt_buffer2 pointer to right channel mlt coefficients
724 */
725static void decouple_float (COOKContext *q,
726 int subband,
727 REAL_T f1, REAL_T f2,
728 REAL_T *decode_buffer,
729 REAL_T *mlt_buffer1, REAL_T *mlt_buffer2)
730{
731 int j, tmp_idx;
732 for (j=0 ; j<SUBBAND_SIZE ; j++) {
733 tmp_idx = ((q->js_subband_start + subband)*SUBBAND_SIZE)+j;
734 mlt_buffer1[SUBBAND_SIZE*subband + j] = f1 * decode_buffer[tmp_idx];
735 mlt_buffer2[SUBBAND_SIZE*subband + j] = f2 * decode_buffer[tmp_idx];
736 }
737}
738
739/** 514/**
740 * function for decoding joint stereo data 515 * function for decoding joint stereo data
741 * 516 *
@@ -809,26 +584,6 @@ decode_bytes_and_gain(COOKContext *q, const uint8_t *inbuffer,
809 FFSWAP(int *, gains_ptr->now, gains_ptr->previous); 584 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
810} 585}
811 586
812 /**
813 * Saturate the output signal to signed 16bit integers.
814 *
815 * @param q pointer to the COOKContext
816 * @param chan channel to saturate
817 * @param out pointer to the output vector
818 */
819static void
820saturate_output_float (COOKContext *q, int chan, int16_t *out)
821{
822 int j;
823 float *output = (float*)q->mono_mdct_output + q->samples_per_channel;
824 /* Clip and convert floats to 16 bits.
825 */
826 for (j = 0; j < q->samples_per_channel; j++) {
827 out[chan + q->nb_channels * j] =
828 av_clip_int16(lrintf(output[j]));
829 }
830}
831
832/** 587/**
833 * Final part of subpacket decoding: 588 * Final part of subpacket decoding:
834 * Apply modulated lapped transform, gain compensation, 589 * Apply modulated lapped transform, gain compensation,
@@ -965,18 +720,6 @@ static void dump_cook_context(COOKContext *q)
965} 720}
966#endif 721#endif
967 722
968#if 0
969static av_cold int cook_count_channels(unsigned int mask){
970 int i;
971 int channels = 0;
972 for(i = 0;i<32;i++){
973 if(mask & (1<<i))
974 ++channels;
975 }
976 return channels;
977}
978#endif
979
980/** 723/**
981 * Cook initialization 724 * Cook initialization
982 */ 725 */
@@ -1057,10 +800,6 @@ av_cold int cook_decode_init(RMContext *rmctx, COOKContext *q)
1057 q->numvector_size = (1 << q->log2_numvector_size); 800 q->numvector_size = (1 << q->log2_numvector_size);
1058 801
1059 /* Generate tables */ 802 /* Generate tables */
1060 init_pow2table();
1061 init_gain_table(q);
1062 init_cplscales_table(q);
1063
1064 if (init_cook_vlc_tables(q) != 0) 803 if (init_cook_vlc_tables(q) != 0)
1065 return -1; 804 return -1;
1066 805
@@ -1092,17 +831,11 @@ av_cold int cook_decode_init(RMContext *rmctx, COOKContext *q)
1092 q->gains2.now = q->gain_3; 831 q->gains2.now = q->gain_3;
1093 q->gains2.previous = q->gain_4; 832 q->gains2.previous = q->gain_4;
1094 833
1095 /* Initialize transform. */
1096 if ( init_cook_mlt(q) != 0 )
1097 return -1;
1098 834
1099 /* Initialize COOK signal arithmetic handling */ 835 /* Initialize COOK signal arithmetic handling */
1100 if (1) { 836 if (1) {
1101 q->scalar_dequant = scalar_dequant_math; 837 q->scalar_dequant = scalar_dequant_math;
1102 q->decouple = decouple_float;
1103 q->imlt_window = imlt_window_float;
1104 q->interpolate = interpolate_math; 838 q->interpolate = interpolate_math;
1105 q->saturate_output = saturate_output_float;
1106 } 839 }
1107 840
1108 /* Try to catch some obviously faulty streams, othervise it might be exploitable */ 841 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
diff --git a/apps/codecs/libcook/cook.h b/apps/codecs/libcook/cook.h
index c4c06cd4f1..e34883255b 100644
--- a/apps/codecs/libcook/cook.h
+++ b/apps/codecs/libcook/cook.h
@@ -25,7 +25,6 @@
25#include <stdint.h> 25#include <stdint.h>
26#include "libavutil/lfg.h" 26#include "libavutil/lfg.h"
27#include "bitstream.h" 27#include "bitstream.h"
28#include "dsputil.h"
29#include "bytestream.h" 28#include "bytestream.h"
30#include "rm2wav.h" 29#include "rm2wav.h"
31#include "cookdata_fixpoint.h" 30#include "cookdata_fixpoint.h"
@@ -44,20 +43,9 @@ typedef struct cook {
44 int* subband_coef_index, int* subband_coef_sign, 43 int* subband_coef_index, int* subband_coef_sign,
45 REAL_T* mlt_p); 44 REAL_T* mlt_p);
46 45
47 void (* decouple) (struct cook *q,
48 int subband,
49 REAL_T f1, REAL_T f2,
50 REAL_T *decode_buffer,
51 REAL_T *mlt_buffer1, REAL_T *mlt_buffer2);
52
53 void (* imlt_window) (struct cook *q, float *buffer1,
54 cook_gains *gains_ptr, float *previous_buffer);
55
56 void (* interpolate) (struct cook *q, REAL_T* buffer, 46 void (* interpolate) (struct cook *q, REAL_T* buffer,
57 int gain_index, int gain_index_next); 47 int gain_index, int gain_index_next);
58 48
59 void (* saturate_output) (struct cook *q, int chan, int16_t *out);
60
61 GetBitContext gb; 49 GetBitContext gb;
62 int frame_number; 50 int frame_number;
63 int block_align; 51 int block_align;
@@ -79,9 +67,6 @@ typedef struct cook {
79 int cookversion; 67 int cookversion;
80 /* states */ 68 /* states */
81 AVLFG random_state; 69 AVLFG random_state;
82 /* transform data */
83 MDCTContext mdct_ctx;
84 float* mlt_window;
85 70
86 /* gain buffers */ 71 /* gain buffers */
87 cook_gains gains1; 72 cook_gains gains1;
@@ -110,8 +95,6 @@ typedef struct cook {
110 REAL_T decode_buffer_1[1024]; 95 REAL_T decode_buffer_1[1024];
111 REAL_T decode_buffer_2[1024]; 96 REAL_T decode_buffer_2[1024];
112 REAL_T decode_buffer_0[1060]; /* static allocation for joint decode */ 97 REAL_T decode_buffer_0[1060]; /* static allocation for joint decode */
113
114 const float *cplscales[5];
115} COOKContext; 98} COOKContext;
116 99
117av_cold int cook_decode_init(RMContext *rmctx, COOKContext *q); 100av_cold int cook_decode_init(RMContext *rmctx, COOKContext *q);
diff --git a/apps/codecs/libcook/dsputil.h b/apps/codecs/libcook/dsputil.h
index 4573c17a62..e69de29bb2 100644
--- a/apps/codecs/libcook/dsputil.h
+++ b/apps/codecs/libcook/dsputil.h
@@ -1,908 +0,0 @@
1/*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
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/**
24 * @file libavcodec/dsputil.h
25 * DSP utils.
26 * note, many functions in here may use MMX which trashes the FPU state, it is
27 * absolutely necessary to call emms_c() between dsp & float/double code
28 */
29
30#ifndef AVCODEC_DSPUTIL_H
31#define AVCODEC_DSPUTIL_H
32
33#include "libavutil/intreadwrite.h"
34#include "avcodec.h"
35
36
37//#define DEBUG
38/* dct code */
39#if 0 /*MT : DELETE THIS LINE.*/
40typedef short DCTELEM;
41typedef int DWTELEM;
42typedef short IDWTELEM;
43
44void fdct_ifast (DCTELEM *data);
45void fdct_ifast248 (DCTELEM *data);
46void ff_jpeg_fdct_islow (DCTELEM *data);
47void ff_fdct248_islow (DCTELEM *data);
48
49void j_rev_dct (DCTELEM *data);
50void j_rev_dct4 (DCTELEM *data);
51void j_rev_dct2 (DCTELEM *data);
52void j_rev_dct1 (DCTELEM *data);
53void ff_wmv2_idct_c(DCTELEM *data);
54
55void ff_fdct_mmx(DCTELEM *block);
56void ff_fdct_mmx2(DCTELEM *block);
57void ff_fdct_sse2(DCTELEM *block);
58
59void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
60void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
61void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
62void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
63void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
64void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
65void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
66void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
67void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
68void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
69
70void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
71 const float *src2, int src3, int blocksize, int step);
72void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
73 const float *win, float add_bias, int len);
74void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
75void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
76
77/* encoding scans */
78extern const uint8_t ff_alternate_horizontal_scan[64];
79extern const uint8_t ff_alternate_vertical_scan[64];
80extern const uint8_t ff_zigzag_direct[64];
81extern const uint8_t ff_zigzag248_direct[64];
82
83/* pixel operations */
84#define MAX_NEG_CROP 1024
85
86/* temporary */
87extern uint32_t ff_squareTbl[512];
88extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
89
90/* VP3 DSP functions */
91void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
92void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
93void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
94
95void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
96void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
97
98/* VP6 DSP functions */
99void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride,
100 const int16_t *h_weights, const int16_t *v_weights);
101
102/* 1/2^n downscaling functions from imgconvert.c */
103void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
104void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
105void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
106void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
107
108void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
109 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
110
111/* minimum alignment rules ;)
112If you notice errors in the align stuff, need more alignment for some ASM code
113for some CPU or need to use a function with less aligned data then send a mail
114to the ffmpeg-devel mailing list, ...
115
116!warning These alignments might not match reality, (missing attribute((align))
117stuff somewhere possible).
118I (Michael) did not check them, these are just the alignments which I think
119could be reached easily ...
120
121!future video codecs might need functions with less strict alignment
122*/
123
124/*
125void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
126void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
127void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
128void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
129void clear_blocks_c(DCTELEM *blocks);
130*/
131
132/* add and put pixel (decoding) */
133// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
134//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
135typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
136typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
137typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
138typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
139typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
140typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
141
142#define DEF_OLD_QPEL(name)\
143void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
144void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
145void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
146
147DEF_OLD_QPEL(qpel16_mc11_old_c)
148DEF_OLD_QPEL(qpel16_mc31_old_c)
149DEF_OLD_QPEL(qpel16_mc12_old_c)
150DEF_OLD_QPEL(qpel16_mc32_old_c)
151DEF_OLD_QPEL(qpel16_mc13_old_c)
152DEF_OLD_QPEL(qpel16_mc33_old_c)
153DEF_OLD_QPEL(qpel8_mc11_old_c)
154DEF_OLD_QPEL(qpel8_mc31_old_c)
155DEF_OLD_QPEL(qpel8_mc12_old_c)
156DEF_OLD_QPEL(qpel8_mc32_old_c)
157DEF_OLD_QPEL(qpel8_mc13_old_c)
158DEF_OLD_QPEL(qpel8_mc33_old_c)
159
160#define CALL_2X_PIXELS(a, b, n)\
161static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
162 b(block , pixels , line_size, h);\
163 b(block+n, pixels+n, line_size, h);\
164}
165
166/* motion estimation */
167// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
168// although currently h<4 is not used as functions with width <8 are neither used nor implemented
169typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
170
171
172// for snow slices
173typedef struct slice_buffer_s slice_buffer;
174
175/**
176 * Scantable.
177 */
178typedef struct ScanTable{
179 const uint8_t *scantable;
180 uint8_t permutated[64];
181 uint8_t raster_end[64];
182#if ARCH_PPC
183 /** Used by dct_quantize_altivec to find last-non-zero */
184 DECLARE_ALIGNED(16, uint8_t, inverse[64]);
185#endif
186} ScanTable;
187
188void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
189
190void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
191 int block_w, int block_h,
192 int src_x, int src_y, int w, int h);
193
194/**
195 * DSPContext.
196 */
197typedef struct DSPContext {
198 /* pixel ops : interface with DCT */
199 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
200 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
201 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
202 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
203 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
204 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
205 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
206 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
207 /**
208 * translational global motion compensation.
209 */
210 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
211 /**
212 * global motion compensation.
213 */
214 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
215 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
216 void (*clear_block)(DCTELEM *block/*align 16*/);
217 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
218 int (*pix_sum)(uint8_t * pix, int line_size);
219 int (*pix_norm1)(uint8_t * pix, int line_size);
220// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
221
222 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
223 me_cmp_func sse[6];
224 me_cmp_func hadamard8_diff[6];
225 me_cmp_func dct_sad[6];
226 me_cmp_func quant_psnr[6];
227 me_cmp_func bit[6];
228 me_cmp_func rd[6];
229 me_cmp_func vsad[6];
230 me_cmp_func vsse[6];
231 me_cmp_func nsse[6];
232 me_cmp_func w53[6];
233 me_cmp_func w97[6];
234 me_cmp_func dct_max[6];
235 me_cmp_func dct264_sad[6];
236
237 me_cmp_func me_pre_cmp[6];
238 me_cmp_func me_cmp[6];
239 me_cmp_func me_sub_cmp[6];
240 me_cmp_func mb_cmp[6];
241 me_cmp_func ildct_cmp[6]; //only width 16 used
242 me_cmp_func frame_skip_cmp[6]; //only width 8 used
243
244 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
245 int size);
246
247 /**
248 * Halfpel motion compensation with rounding (a+b+1)>>1.
249 * this is an array[4][4] of motion compensation functions for 4
250 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
251 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
252 * @param block destination where the result is stored
253 * @param pixels source
254 * @param line_size number of bytes in a horizontal line of block
255 * @param h height
256 */
257 op_pixels_func put_pixels_tab[4][4];
258
259 /**
260 * Halfpel motion compensation with rounding (a+b+1)>>1.
261 * This is an array[4][4] of motion compensation functions for 4
262 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
263 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
264 * @param block destination into which the result is averaged (a+b+1)>>1
265 * @param pixels source
266 * @param line_size number of bytes in a horizontal line of block
267 * @param h height
268 */
269 op_pixels_func avg_pixels_tab[4][4];
270
271 /**
272 * Halfpel motion compensation with no rounding (a+b)>>1.
273 * this is an array[2][4] of motion compensation functions for 2
274 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
275 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
276 * @param block destination where the result is stored
277 * @param pixels source
278 * @param line_size number of bytes in a horizontal line of block
279 * @param h height
280 */
281 op_pixels_func put_no_rnd_pixels_tab[4][4];
282
283 /**
284 * Halfpel motion compensation with no rounding (a+b)>>1.
285 * this is an array[2][4] of motion compensation functions for 2
286 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
287 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
288 * @param block destination into which the result is averaged (a+b)>>1
289 * @param pixels source
290 * @param line_size number of bytes in a horizontal line of block
291 * @param h height
292 */
293 op_pixels_func avg_no_rnd_pixels_tab[4][4];
294
295 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
296
297 /**
298 * Thirdpel motion compensation with rounding (a+b+1)>>1.
299 * this is an array[12] of motion compensation functions for the 9 thirdpe
300 * positions<br>
301 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
302 * @param block destination where the result is stored
303 * @param pixels source
304 * @param line_size number of bytes in a horizontal line of block
305 * @param h height
306 */
307 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
308 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
309
310 qpel_mc_func put_qpel_pixels_tab[2][16];
311 qpel_mc_func avg_qpel_pixels_tab[2][16];
312 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
313 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
314 qpel_mc_func put_mspel_pixels_tab[8];
315
316 /**
317 * h264 Chroma MC
318 */
319 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
320 /* This is really one func used in VC-1 decoding */
321 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
322 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
323
324 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
325 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
326
327 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
328 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
329
330 h264_weight_func weight_h264_pixels_tab[10];
331 h264_biweight_func biweight_h264_pixels_tab[10];
332
333 /* AVS specific */
334 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
335 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
336 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
337 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
338 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
339 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
340 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
341
342 me_cmp_func pix_abs[2][4];
343
344 /* huffyuv specific */
345 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
346 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
347 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
348 /**
349 * subtract huffyuv's variant of median prediction
350 * note, this might read from src1[-1], src2[-1]
351 */
352 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
353 void (*add_hfyu_median_prediction)(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top);
354 /* this might write to dst[w] */
355 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
356 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
357
358 void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
359 void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
360 /* v/h_loop_filter_luma_intra: align 16 */
361 void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
362 void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
363 void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
364 void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
365 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
366 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
367 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
368 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
369 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
370
371 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
372 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
373
374 void (*h261_loop_filter)(uint8_t *src, int stride);
375
376 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
377 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
378
379 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
380 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
381
382 void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
383 const int16_t *h_weights,const int16_t *v_weights);
384
385 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
386 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
387 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
388 /* no alignment needed */
389 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
390 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
391 void (*vector_fmul)(float *dst, const float *src, int len);
392 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
393 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
394 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
395 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
396 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
397 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
398 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
399
400 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
401 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
402 void (*float_to_int16)(int16_t *dst, const float *src, long len);
403 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
404
405 /* (I)DCT */
406 void (*fdct)(DCTELEM *block/* align 16*/);
407 void (*fdct248)(DCTELEM *block/* align 16*/);
408
409 /* IDCT really*/
410 void (*idct)(DCTELEM *block/* align 16*/);
411
412 /**
413 * block -> idct -> clip to unsigned 8 bit -> dest.
414 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
415 * @param line_size size in bytes of a horizontal line of dest
416 */
417 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
418
419 /**
420 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
421 * @param line_size size in bytes of a horizontal line of dest
422 */
423 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
424
425 /**
426 * idct input permutation.
427 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
428 * IDCT)
429 * this permutation must be performed before the idct_put/add, note, normally this can be merged
430 * with the zigzag/alternate scan<br>
431 * an example to avoid confusion:
432 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
433 * - (x -> referece dct -> reference idct -> x)
434 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
435 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
436 */
437 uint8_t idct_permutation[64];
438 int idct_permutation_type;
439#define FF_NO_IDCT_PERM 1
440#define FF_LIBMPEG2_IDCT_PERM 2
441#define FF_SIMPLE_IDCT_PERM 3
442#define FF_TRANSPOSE_IDCT_PERM 4
443#define FF_PARTTRANS_IDCT_PERM 5
444#define FF_SSE2_IDCT_PERM 6
445
446 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
447 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
448#define BASIS_SHIFT 16
449#define RECON_SHIFT 6
450
451 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
452#define EDGE_WIDTH 16
453
454 /* h264 functions */
455 /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
456 NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them
457 The reason for above, is that no 2 out of one list may use a different permutation.
458 */
459 void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
460 void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
461 void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
462 void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
463 void (*h264_dct)(DCTELEM block[4][4]);
464 void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
465 void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
466 void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
467 void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
468
469 /* snow wavelet */
470 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
471 void (*horizontal_compose97i)(IDWTELEM *b, int width);
472 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
473
474 void (*prefetch)(void *mem, int stride, int h);
475
476 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
477
478 /* vc1 functions */
479 void (*vc1_inv_trans_8x8)(DCTELEM *b);
480 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
481 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
482 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
483 void (*vc1_v_overlap)(uint8_t* src, int stride);
484 void (*vc1_h_overlap)(uint8_t* src, int stride);
485 /* put 8x8 block with bicubic interpolation and quarterpel precision
486 * last argument is actually round value instead of height
487 */
488 op_pixels_func put_vc1_mspel_pixels_tab[16];
489
490 /* intrax8 functions */
491 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
492 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
493 int * range, int * sum, int edges);
494
495 /* ape functions */
496 /**
497 * Add contents of the second vector to the first one.
498 * @param len length of vectors, should be multiple of 16
499 */
500 void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
501 /**
502 * Add contents of the second vector to the first one.
503 * @param len length of vectors, should be multiple of 16
504 */
505 void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
506 /**
507 * Calculate scalar product of two vectors.
508 * @param len length of vectors, should be multiple of 16
509 * @param shift number of bits to discard from product
510 */
511 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
512
513 /* rv30 functions */
514 qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
515 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
516
517 /* rv40 functions */
518 qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
519 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
520 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
521 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
522} DSPContext;
523
524void dsputil_static_init(void);
525void dsputil_init(DSPContext* p, AVCodecContext *avctx);
526
527int ff_check_alignment(void);
528
529/**
530 * permute block according to permuatation.
531 * @param last last non zero element in scantable order
532 */
533void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
534
535void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
536
537#define BYTE_VEC32(c) ((c)*0x01010101UL)
538
539static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
540{
541 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
542}
543
544static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
545{
546 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
547}
548
549static inline int get_penalty_factor(int lambda, int lambda2, int type){
550 switch(type&0xFF){
551 default:
552 case FF_CMP_SAD:
553 return lambda>>FF_LAMBDA_SHIFT;
554 case FF_CMP_DCT:
555 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
556 case FF_CMP_W53:
557 return (4*lambda)>>(FF_LAMBDA_SHIFT);
558 case FF_CMP_W97:
559 return (2*lambda)>>(FF_LAMBDA_SHIFT);
560 case FF_CMP_SATD:
561 case FF_CMP_DCT264:
562 return (2*lambda)>>FF_LAMBDA_SHIFT;
563 case FF_CMP_RD:
564 case FF_CMP_PSNR:
565 case FF_CMP_SSE:
566 case FF_CMP_NSSE:
567 return lambda2>>FF_LAMBDA_SHIFT;
568 case FF_CMP_BIT:
569 return 1;
570 }
571}
572
573/**
574 * Empty mmx state.
575 * this must be called between any dsp function and float/double code.
576 * for example sin(); dsp->idct_put(); emms_c(); cos()
577 */
578#define emms_c()
579
580/* should be defined by architectures supporting
581 one or more MultiMedia extension */
582int mm_support(void);
583
584void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
585void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
586void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
587void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
588void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
589void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
590void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
591void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
592void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
593
594#endif /*MT : DELETE THIS LINE ONLY. */
595#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
596
597#if 0 /*MT : DELETE THIS LINE ONLY. */
598#if HAVE_MMX
599
600#undef emms_c
601
602extern int mm_flags;
603
604void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
605void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
606void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
607
608static inline void emms(void)
609{
610 __asm__ volatile ("emms;":::"memory");
611}
612
613
614#define emms_c() \
615{\
616 if (mm_flags & FF_MM_MMX)\
617 emms();\
618}
619
620void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
621
622#elif ARCH_ARM
623
624extern int mm_flags;
625
626#if HAVE_NEON
627# define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
628# define STRIDE_ALIGN 16
629#endif
630
631#elif ARCH_PPC
632
633extern int mm_flags;
634
635#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
636#define STRIDE_ALIGN 16
637
638#elif HAVE_MMI
639
640#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
641#define STRIDE_ALIGN 16
642
643#else
644
645#define mm_flags 0
646#define mm_support() 0
647
648#endif
649
650#endif /* MT : DELETE THIS LINE ONLY */
651#ifndef DECLARE_ALIGNED_8
652# define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
653#endif
654
655#if 0 /* MT : DELETE THIS LINE ONLY */
656#ifndef STRIDE_ALIGN
657# define STRIDE_ALIGN 8
658#endif
659
660/* PSNR */
661void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
662 int orig_linesize[3], int coded_linesize,
663 AVCodecContext *avctx);
664
665#endif /*MT : DELETE THIS LINE.*/
666/* FFT computation */
667
668/* NOTE: soon integer code will be added, so you must use the
669 FFTSample type */
670typedef float FFTSample;
671
672struct MDCTContext;
673
674typedef struct FFTComplex {
675 FFTSample re, im;
676} FFTComplex;
677
678typedef struct FFTContext {
679 int nbits;
680 int inverse;
681 uint16_t *revtab;
682 FFTComplex *exptab;
683 FFTComplex *exptab1; /* only used by SSE code */
684 FFTComplex *tmp_buf;
685 void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
686 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
687 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
688 void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
689} FFTContext;
690
691extern FFTSample* ff_cos_tabs[13];
692
693/**
694 * Sets up a complex FFT.
695 * @param nbits log2 of the length of the input array
696 * @param inverse if 0 perform the forward transform, if 1 perform the inverse
697 */
698int ff_fft_init(FFTContext *s, int nbits, int inverse);
699void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
700void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
701void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
702void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
703void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
704void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
705void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
706
707/**
708 * Do the permutation needed BEFORE calling ff_fft_calc().
709 */
710static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
711{
712 s->fft_permute(s, z);
713}
714/**
715 * Do a complex FFT with the parameters defined in ff_fft_init(). The
716 * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
717 */
718static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
719{
720 s->fft_calc(s, z);
721}
722void ff_fft_end(FFTContext *s);
723
724#endif /*MT : DELETE THIS LINE.*/
725/* MDCT computation */
726
727typedef struct MDCTContext {
728 int n; /* size of MDCT (i.e. number of input data * 2) */
729 int nbits; /* n = 2^nbits */
730 /* pre/post rotation tables */
731 FFTSample *tcos;
732 FFTSample *tsin;
733 FFTContext fft;
734} MDCTContext;
735
736static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
737{
738 s->fft.imdct_calc(s, output, input);
739}
740static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
741{
742 s->fft.imdct_half(s, output, input);
743}
744
745#if 0 /* MT : DELETE THIS LINE. */
746/**
747 * Generate a Kaiser-Bessel Derived Window.
748 * @param window pointer to half window
749 * @param alpha determines window shape
750 * @param n size of half window
751 */
752void ff_kbd_window_init(float *window, float alpha, int n);
753#endif /* MT : DELETE THIS LINE.*/
754
755/**
756 * Generate a sine window.
757 * @param window pointer to half window
758 * @param n size of half window
759 */
760void ff_sine_window_init(float *window, int n);
761extern float ff_sine_128 [ 128];
762extern float ff_sine_256 [ 256];
763extern float ff_sine_512 [ 512];
764extern float ff_sine_1024[1024];
765extern float ff_sine_2048[2048];
766extern float ff_sine_4096[4096];
767extern float *ff_sine_windows[6];
768
769int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
770void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
771void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
772void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
773void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
774void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
775void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
776void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
777void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
778void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
779void ff_mdct_end(MDCTContext *s);
780
781#if 0 /* MT : DELETE THIS LINE.*/
782/* Real Discrete Fourier Transform */
783
784enum RDFTransformType {
785 RDFT,
786 IRDFT,
787 RIDFT,
788 IRIDFT,
789};
790
791typedef struct {
792 int nbits;
793 int inverse;
794 int sign_convention;
795
796 /* pre/post rotation tables */
797 FFTSample *tcos;
798 FFTSample *tsin;
799 FFTContext fft;
800} RDFTContext;
801
802/**
803 * Sets up a real FFT.
804 * @param nbits log2 of the length of the input array
805 * @param trans the type of transform
806 */
807int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
808void ff_rdft_calc(RDFTContext *s, FFTSample *data);
809void ff_rdft_end(RDFTContext *s);
810
811#define WRAPPER8_16(name8, name16)\
812static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
813 return name8(s, dst , src , stride, h)\
814 +name8(s, dst+8 , src+8 , stride, h);\
815}
816
817#define WRAPPER8_16_SQ(name8, name16)\
818static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
819 int score=0;\
820 score +=name8(s, dst , src , stride, 8);\
821 score +=name8(s, dst+8 , src+8 , stride, 8);\
822 if(h==16){\
823 dst += 8*stride;\
824 src += 8*stride;\
825 score +=name8(s, dst , src , stride, 8);\
826 score +=name8(s, dst+8 , src+8 , stride, 8);\
827 }\
828 return score;\
829}
830
831
832static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
833{
834 int i;
835 for(i=0; i<h; i++)
836 {
837 AV_WN16(dst , AV_RN16(src ));
838 dst+=dstStride;
839 src+=srcStride;
840 }
841}
842
843static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
844{
845 int i;
846 for(i=0; i<h; i++)
847 {
848 AV_WN32(dst , AV_RN32(src ));
849 dst+=dstStride;
850 src+=srcStride;
851 }
852}
853
854static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
855{
856 int i;
857 for(i=0; i<h; i++)
858 {
859 AV_WN32(dst , AV_RN32(src ));
860 AV_WN32(dst+4 , AV_RN32(src+4 ));
861 dst+=dstStride;
862 src+=srcStride;
863 }
864}
865
866static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
867{
868 int i;
869 for(i=0; i<h; i++)
870 {
871 AV_WN32(dst , AV_RN32(src ));
872 AV_WN32(dst+4 , AV_RN32(src+4 ));
873 dst[8]= src[8];
874 dst+=dstStride;
875 src+=srcStride;
876 }
877}
878
879static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
880{
881 int i;
882 for(i=0; i<h; i++)
883 {
884 AV_WN32(dst , AV_RN32(src ));
885 AV_WN32(dst+4 , AV_RN32(src+4 ));
886 AV_WN32(dst+8 , AV_RN32(src+8 ));
887 AV_WN32(dst+12, AV_RN32(src+12));
888 dst+=dstStride;
889 src+=srcStride;
890 }
891}
892
893static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
894{
895 int i;
896 for(i=0; i<h; i++)
897 {
898 AV_WN32(dst , AV_RN32(src ));
899 AV_WN32(dst+4 , AV_RN32(src+4 ));
900 AV_WN32(dst+8 , AV_RN32(src+8 ));
901 AV_WN32(dst+12, AV_RN32(src+12));
902 dst[16]= src[16];
903 dst+=dstStride;
904 src+=srcStride;
905 }
906}
907
908#endif /* AVCODEC_DSPUTIL_H */
diff --git a/apps/codecs/libcook/fft.c b/apps/codecs/libcook/fft.c
index a3f1151472..e69de29bb2 100644
--- a/apps/codecs/libcook/fft.c
+++ b/apps/codecs/libcook/fft.c
@@ -1,374 +0,0 @@
1/*
2 * FFT/IFFT transforms
3 * Copyright (c) 2008 Loren Merritt
4 * Copyright (c) 2002 Fabrice Bellard
5 * Partly based on libdjbfft by D. J. Bernstein
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24/**
25 * @file libavcodec/fft.c
26 * FFT/IFFT transforms.
27 */
28
29#include "dsputil.h"
30
31/* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */
32DECLARE_ALIGNED_16(FFTSample, ff_cos_16[8]);
33DECLARE_ALIGNED_16(FFTSample, ff_cos_32[16]);
34DECLARE_ALIGNED_16(FFTSample, ff_cos_64[32]);
35DECLARE_ALIGNED_16(FFTSample, ff_cos_128[64]);
36DECLARE_ALIGNED_16(FFTSample, ff_cos_256[128]);
37DECLARE_ALIGNED_16(FFTSample, ff_cos_512[256]);
38DECLARE_ALIGNED_16(FFTSample, ff_cos_1024[512]);
39DECLARE_ALIGNED_16(FFTSample, ff_cos_2048[1024]);
40DECLARE_ALIGNED_16(FFTSample, ff_cos_4096[2048]);
41DECLARE_ALIGNED_16(FFTSample, ff_cos_8192[4096]);
42DECLARE_ALIGNED_16(FFTSample, ff_cos_16384[8192]);
43DECLARE_ALIGNED_16(FFTSample, ff_cos_32768[16384]);
44DECLARE_ALIGNED_16(FFTSample, ff_cos_65536[32768]);
45FFTSample *ff_cos_tabs[] = {
46 ff_cos_16, ff_cos_32, ff_cos_64, ff_cos_128, ff_cos_256, ff_cos_512, ff_cos_1024,
47 ff_cos_2048, ff_cos_4096, ff_cos_8192, ff_cos_16384, ff_cos_32768, ff_cos_65536,
48};
49
50static int split_radix_permutation(int i, int n, int inverse)
51{
52 int m;
53 if(n <= 2) return i&1;
54 m = n >> 1;
55 if(!(i&m)) return split_radix_permutation(i, m, inverse)*2;
56 m >>= 1;
57 if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;
58 else return split_radix_permutation(i, m, inverse)*4 - 1;
59}
60
61av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
62{
63 int i, j, m, n;
64 float alpha, c1, s1, s2;
65 int split_radix = 1;
66 int av_unused has_vectors;
67
68 if (nbits < 2 || nbits > 16)
69 goto fail;
70 s->nbits = nbits;
71 n = 1 << nbits;
72
73 s->tmp_buf = NULL;
74 s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
75 if (!s->exptab)
76 goto fail;
77 s->revtab = av_malloc(n * sizeof(uint16_t));
78 if (!s->revtab)
79 goto fail;
80 s->inverse = inverse;
81
82 s2 = inverse ? 1.0 : -1.0;
83
84 s->fft_permute = ff_fft_permute_c;
85 s->fft_calc = ff_fft_calc_c;
86 s->imdct_calc = ff_imdct_calc_c;
87 s->imdct_half = ff_imdct_half_c;
88 s->exptab1 = NULL;
89
90#if HAVE_MMX && HAVE_YASM
91 has_vectors = mm_support();
92 if (has_vectors & FF_MM_SSE && HAVE_SSE) {
93 /* SSE for P3/P4/K8 */
94 s->imdct_calc = ff_imdct_calc_sse;
95 s->imdct_half = ff_imdct_half_sse;
96 s->fft_permute = ff_fft_permute_sse;
97 s->fft_calc = ff_fft_calc_sse;
98 } else if (has_vectors & FF_MM_3DNOWEXT && HAVE_AMD3DNOWEXT) {
99 /* 3DNowEx for K7 */
100 s->imdct_calc = ff_imdct_calc_3dn2;
101 s->imdct_half = ff_imdct_half_3dn2;
102 s->fft_calc = ff_fft_calc_3dn2;
103 } else if (has_vectors & FF_MM_3DNOW && HAVE_AMD3DNOW) {
104 /* 3DNow! for K6-2/3 */
105 s->imdct_calc = ff_imdct_calc_3dn;
106 s->imdct_half = ff_imdct_half_3dn;
107 s->fft_calc = ff_fft_calc_3dn;
108 }
109#elif HAVE_ALTIVEC
110 has_vectors = mm_support();
111 if (has_vectors & FF_MM_ALTIVEC) {
112 s->fft_calc = ff_fft_calc_altivec;
113 split_radix = 0;
114 }
115#endif
116
117 if (split_radix) {
118 for(j=4; j<=nbits; j++) {
119 int m = 1<<j;
120 double freq = 2*M_PI/m;
121 FFTSample *tab = ff_cos_tabs[j-4];
122 for(i=0; i<=m/4; i++)
123 tab[i] = cos(i*freq);
124 for(i=1; i<m/4; i++)
125 tab[m/2-i] = tab[i];
126 }
127 for(i=0; i<n; i++)
128 s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = i;
129 s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
130 } else {
131 int np, nblocks, np2, l;
132 FFTComplex *q;
133
134 for(i=0; i<(n/2); i++) {
135 alpha = 2 * M_PI * (float)i / (float)n;
136 c1 = cos(alpha);
137 s1 = sin(alpha) * s2;
138 s->exptab[i].re = c1;
139 s->exptab[i].im = s1;
140 }
141
142 np = 1 << nbits;
143 nblocks = np >> 3;
144 np2 = np >> 1;
145 s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
146 if (!s->exptab1)
147 goto fail;
148 q = s->exptab1;
149 do {
150 for(l = 0; l < np2; l += 2 * nblocks) {
151 *q++ = s->exptab[l];
152 *q++ = s->exptab[l + nblocks];
153
154 q->re = -s->exptab[l].im;
155 q->im = s->exptab[l].re;
156 q++;
157 q->re = -s->exptab[l + nblocks].im;
158 q->im = s->exptab[l + nblocks].re;
159 q++;
160 }
161 nblocks = nblocks >> 1;
162 } while (nblocks != 0);
163 av_freep(&s->exptab);
164
165 /* compute bit reverse table */
166 for(i=0;i<n;i++) {
167 m=0;
168 for(j=0;j<nbits;j++) {
169 m |= ((i >> j) & 1) << (nbits-j-1);
170 }
171 s->revtab[i]=m;
172 }
173 }
174
175 return 0;
176 fail:
177 av_freep(&s->revtab);
178 av_freep(&s->exptab);
179 av_freep(&s->exptab1);
180 av_freep(&s->tmp_buf);
181 return -1;
182}
183
184void ff_fft_permute_c(FFTContext *s, FFTComplex *z)
185{
186 int j, k, np;
187 FFTComplex tmp;
188 const uint16_t *revtab = s->revtab;
189 np = 1 << s->nbits;
190
191 if (s->tmp_buf) {
192 /* TODO: handle split-radix permute in a more optimal way, probably in-place */
193 for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
194 memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
195 return;
196 }
197
198 /* reverse */
199 for(j=0;j<np;j++) {
200 k = revtab[j];
201 if (k < j) {
202 tmp = z[k];
203 z[k] = z[j];
204 z[j] = tmp;
205 }
206 }
207}
208
209av_cold void ff_fft_end(FFTContext *s)
210{
211 av_freep(&s->revtab);
212 av_freep(&s->exptab);
213 av_freep(&s->exptab1);
214 av_freep(&s->tmp_buf);
215}
216
217#define sqrthalf (float)M_SQRT1_2
218
219#define BF(x,y,a,b) {\
220 x = a - b;\
221 y = a + b;\
222}
223
224#define BUTTERFLIES(a0,a1,a2,a3) {\
225 BF(t3, t5, t5, t1);\
226 BF(a2.re, a0.re, a0.re, t5);\
227 BF(a3.im, a1.im, a1.im, t3);\
228 BF(t4, t6, t2, t6);\
229 BF(a3.re, a1.re, a1.re, t4);\
230 BF(a2.im, a0.im, a0.im, t6);\
231}
232
233// force loading all the inputs before storing any.
234// this is slightly slower for small data, but avoids store->load aliasing
235// for addresses separated by large powers of 2.
236#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
237 FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
238 BF(t3, t5, t5, t1);\
239 BF(a2.re, a0.re, r0, t5);\
240 BF(a3.im, a1.im, i1, t3);\
241 BF(t4, t6, t2, t6);\
242 BF(a3.re, a1.re, r1, t4);\
243 BF(a2.im, a0.im, i0, t6);\
244}
245
246#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
247 t1 = a2.re * wre + a2.im * wim;\
248 t2 = a2.im * wre - a2.re * wim;\
249 t5 = a3.re * wre - a3.im * wim;\
250 t6 = a3.im * wre + a3.re * wim;\
251 BUTTERFLIES(a0,a1,a2,a3)\
252}
253
254#define TRANSFORM_ZERO(a0,a1,a2,a3) {\
255 t1 = a2.re;\
256 t2 = a2.im;\
257 t5 = a3.re;\
258 t6 = a3.im;\
259 BUTTERFLIES(a0,a1,a2,a3)\
260}
261
262/* z[0...8n-1], w[1...2n-1] */
263#define PASS(name)\
264static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
265{\
266 FFTSample t1, t2, t3, t4, t5, t6;\
267 int o1 = 2*n;\
268 int o2 = 4*n;\
269 int o3 = 6*n;\
270 const FFTSample *wim = wre+o1;\
271 n--;\
272\
273 TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
274 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
275 do {\
276 z += 2;\
277 wre += 2;\
278 wim -= 2;\
279 TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
280 TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
281 } while(--n);\
282}
283
284PASS(pass)
285#undef BUTTERFLIES
286#define BUTTERFLIES BUTTERFLIES_BIG
287PASS(pass_big)
288
289#define DECL_FFT(n,n2,n4)\
290static void fft##n(FFTComplex *z)\
291{\
292 fft##n2(z);\
293 fft##n4(z+n4*2);\
294 fft##n4(z+n4*3);\
295 pass(z,ff_cos_##n,n4/2);\
296}
297
298static void fft4(FFTComplex *z)
299{
300 FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
301
302 BF(t3, t1, z[0].re, z[1].re);
303 BF(t8, t6, z[3].re, z[2].re);
304 BF(z[2].re, z[0].re, t1, t6);
305 BF(t4, t2, z[0].im, z[1].im);
306 BF(t7, t5, z[2].im, z[3].im);
307 BF(z[3].im, z[1].im, t4, t8);
308 BF(z[3].re, z[1].re, t3, t7);
309 BF(z[2].im, z[0].im, t2, t5);
310}
311
312static void fft8(FFTComplex *z)
313{
314 FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
315
316 fft4(z);
317
318 BF(t1, z[5].re, z[4].re, -z[5].re);
319 BF(t2, z[5].im, z[4].im, -z[5].im);
320 BF(t3, z[7].re, z[6].re, -z[7].re);
321 BF(t4, z[7].im, z[6].im, -z[7].im);
322 BF(t8, t1, t3, t1);
323 BF(t7, t2, t2, t4);
324 BF(z[4].re, z[0].re, z[0].re, t1);
325 BF(z[4].im, z[0].im, z[0].im, t2);
326 BF(z[6].re, z[2].re, z[2].re, t7);
327 BF(z[6].im, z[2].im, z[2].im, t8);
328
329 TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
330}
331
332#if !CONFIG_SMALL
333static void fft16(FFTComplex *z)
334{
335 FFTSample t1, t2, t3, t4, t5, t6;
336
337 fft8(z);
338 fft4(z+8);
339 fft4(z+12);
340
341 TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
342 TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);
343 TRANSFORM(z[1],z[5],z[9],z[13],ff_cos_16[1],ff_cos_16[3]);
344 TRANSFORM(z[3],z[7],z[11],z[15],ff_cos_16[3],ff_cos_16[1]);
345}
346#else
347DECL_FFT(16,8,4)
348#endif
349DECL_FFT(32,16,8)
350DECL_FFT(64,32,16)
351DECL_FFT(128,64,32)
352DECL_FFT(256,128,64)
353DECL_FFT(512,256,128)
354#if !CONFIG_SMALL
355#define pass pass_big
356#endif
357DECL_FFT(1024,512,256)
358DECL_FFT(2048,1024,512)
359DECL_FFT(4096,2048,1024)
360DECL_FFT(8192,4096,2048)
361DECL_FFT(16384,8192,4096)
362DECL_FFT(32768,16384,8192)
363DECL_FFT(65536,32768,16384)
364
365static void (*fft_dispatch[])(FFTComplex*) = {
366 fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,
367 fft2048, fft4096, fft8192, fft16384, fft32768, fft65536,
368};
369
370void ff_fft_calc_c(FFTContext *s, FFTComplex *z)
371{
372 fft_dispatch[s->nbits-2](z);
373}
374
diff --git a/apps/codecs/libcook/main.c b/apps/codecs/libcook/main.c
index 2f85295fb5..40e2fd8a4f 100644
--- a/apps/codecs/libcook/main.c
+++ b/apps/codecs/libcook/main.c
@@ -5,7 +5,7 @@
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < < 5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \ 6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/ 7 * \/ \/ \/ \/ \/
8 * $Id$ 8 * $Id$ main.c 20898 2009-05-09 23:24:02Z dave $
9 * 9 *
10 * Copyright (C) 2009 Mohamed Tarek 10 * Copyright (C) 2009 Mohamed Tarek
11 * 11 *
diff --git a/apps/codecs/libcook/mdct.c b/apps/codecs/libcook/mdct.c
index cb3388f6ff..e69de29bb2 100644
--- a/apps/codecs/libcook/mdct.c
+++ b/apps/codecs/libcook/mdct.c
@@ -1,229 +0,0 @@
1/*
2 * MDCT/IMDCT transforms
3 * Copyright (c) 2002 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21#include "dsputil.h"
22
23/**
24 * @file libavcodec/mdct.c
25 * MDCT/IMDCT transforms.
26 */
27
28// Generate a Kaiser-Bessel Derived Window.
29#define BESSEL_I0_ITER 50 // default: 50 iterations of Bessel I0 approximation
30av_cold void ff_kbd_window_init(float *window, float alpha, int n)
31{
32 int i, j;
33 double sum = 0.0, bessel, tmp;
34 double local_window[n];
35 double alpha2 = (alpha * M_PI / n) * (alpha * M_PI / n);
36
37 for (i = 0; i < n; i++) {
38 tmp = i * (n - i) * alpha2;
39 bessel = 1.0;
40 for (j = BESSEL_I0_ITER; j > 0; j--)
41 bessel = bessel * tmp / (j * j) + 1;
42 sum += bessel;
43 local_window[i] = sum;
44 }
45
46 sum++;
47 for (i = 0; i < n; i++)
48 window[i] = sqrt(local_window[i] / sum);
49}
50
51DECLARE_ALIGNED(16, float, ff_sine_128 [ 128]);
52DECLARE_ALIGNED(16, float, ff_sine_256 [ 256]);
53DECLARE_ALIGNED(16, float, ff_sine_512 [ 512]);
54DECLARE_ALIGNED(16, float, ff_sine_1024[1024]);
55DECLARE_ALIGNED(16, float, ff_sine_2048[2048]);
56DECLARE_ALIGNED(16, float, ff_sine_4096[4096]);
57float *ff_sine_windows[6] = {
58 ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048, ff_sine_4096
59};
60
61// Generate a sine window.
62av_cold void ff_sine_window_init(float *window, int n) {
63 int i;
64 for(i = 0; i < n; i++)
65 window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n)));
66}
67
68/**
69 * init MDCT or IMDCT computation.
70 */
71av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
72{
73 int n, n4, i;
74 double alpha;
75
76 memset(s, 0, sizeof(*s));
77 n = 1 << nbits;
78 s->nbits = nbits;
79 s->n = n;
80 n4 = n >> 2;
81 s->tcos = av_malloc(n4 * sizeof(FFTSample));
82 if (!s->tcos)
83 goto fail;
84 s->tsin = av_malloc(n4 * sizeof(FFTSample));
85 if (!s->tsin)
86 goto fail;
87
88 for(i=0;i<n4;i++) {
89 alpha = 2 * M_PI * (i + 1.0 / 8.0) / n;
90 s->tcos[i] = -cos(alpha);
91 s->tsin[i] = -sin(alpha);
92 }
93 if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
94 goto fail;
95 return 0;
96 fail:
97 av_freep(&s->tcos);
98 av_freep(&s->tsin);
99 return -1;
100}
101
102/* complex multiplication: p = a * b */
103#define CMUL(pre, pim, are, aim, bre, bim) \
104{\
105 FFTSample _are = (are);\
106 FFTSample _aim = (aim);\
107 FFTSample _bre = (bre);\
108 FFTSample _bim = (bim);\
109 (pre) = _are * _bre - _aim * _bim;\
110 (pim) = _are * _bim + _aim * _bre;\
111}
112
113/**
114 * Compute the middle half of the inverse MDCT of size N = 2^nbits,
115 * thus excluding the parts that can be derived by symmetry
116 * @param output N/2 samples
117 * @param input N/2 samples
118 */
119void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
120{
121 int k, n8, n4, n2, n, j;
122 const uint16_t *revtab = s->fft.revtab;
123 const FFTSample *tcos = s->tcos;
124 const FFTSample *tsin = s->tsin;
125 const FFTSample *in1, *in2;
126 FFTComplex *z = (FFTComplex *)output;
127
128 n = 1 << s->nbits;
129 n2 = n >> 1;
130 n4 = n >> 2;
131 n8 = n >> 3;
132
133 /* pre rotation */
134 in1 = input;
135 in2 = input + n2 - 1;
136 for(k = 0; k < n4; k++) {
137 j=revtab[k];
138 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
139 in1 += 2;
140 in2 -= 2;
141 }
142 ff_fft_calc(&s->fft, z);
143
144 /* post rotation + reordering */
145 output += n4;
146 for(k = 0; k < n8; k++) {
147 FFTSample r0, i0, r1, i1;
148 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
149 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]);
150 z[n8-k-1].re = r0;
151 z[n8-k-1].im = i0;
152 z[n8+k ].re = r1;
153 z[n8+k ].im = i1;
154 }
155}
156
157/**
158 * Compute inverse MDCT of size N = 2^nbits
159 * @param output N samples
160 * @param input N/2 samples
161 */
162void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
163{
164 int k;
165 int n = 1 << s->nbits;
166 int n2 = n >> 1;
167 int n4 = n >> 2;
168
169 ff_imdct_half_c(s, output+n4, input);
170
171 for(k = 0; k < n4; k++) {
172 output[k] = -output[n2-k-1];
173 output[n-k-1] = output[n2+k];
174 }
175}
176
177/**
178 * Compute MDCT of size N = 2^nbits
179 * @param input N samples
180 * @param out N/2 samples
181 */
182void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input)
183{
184 int i, j, n, n8, n4, n2, n3;
185 FFTSample re, im;
186 const uint16_t *revtab = s->fft.revtab;
187 const FFTSample *tcos = s->tcos;
188 const FFTSample *tsin = s->tsin;
189 FFTComplex *x = (FFTComplex *)out;
190
191 n = 1 << s->nbits;
192 n2 = n >> 1;
193 n4 = n >> 2;
194 n8 = n >> 3;
195 n3 = 3 * n4;
196
197 /* pre rotation */
198 for(i=0;i<n8;i++) {
199 re = -input[2*i+3*n4] - input[n3-1-2*i];
200 im = -input[n4+2*i] + input[n4-1-2*i];
201 j = revtab[i];
202 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
203
204 re = input[2*i] - input[n2-1-2*i];
205 im = -(input[n2+2*i] + input[n-1-2*i]);
206 j = revtab[n8 + i];
207 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
208 }
209
210 ff_fft_calc(&s->fft, x);
211
212 /* post rotation */
213 for(i=0;i<n8;i++) {
214 FFTSample r0, i0, r1, i1;
215 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
216 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]);
217 x[n8-i-1].re = r0;
218 x[n8-i-1].im = i0;
219 x[n8+i ].re = r1;
220 x[n8+i ].im = i1;
221 }
222}
223
224av_cold void ff_mdct_end(MDCTContext *s)
225{
226 av_freep(&s->tcos);
227 av_freep(&s->tsin);
228 ff_fft_end(&s->fft);
229}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-05 15:07:18 +0000