Skip to content

F
ffmpeg.wasm-core
Commit 2f349de2 authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files
Options

(commit by michael) 

mmx & mmx2 quantizer
c dct permutation bugfix
dont copy input on intra only encodings if it can be avoided
dont draw edges on intra only stuff

Originally committed as revision 281 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent 580b82fa
Hide whitespace changes
Inline Side-by-side
Showing with 309 additions and 112 deletions
libavcodec/dsputil.c
View file @ 2f349de2
......@@ -49,6 +49,12 @@ UINT8 zigzag_direct[64] = {
53, 60, 61, 54, 47, 55, 62, 63
};
/* not permutated inverse zigzag_direct + 1 for MMX quantizer */
UINT16 __align8 inv_zigzag_direct16[64];
/* not permutated zigzag_direct for MMX quantizer */
UINT8 zigzag_direct_noperm[64];
UINT8 ff_alternate_horizontal_scan[64] = {
0, 1, 2, 3, 8, 9, 16, 17,
10, 11, 4, 5, 6, 7, 15, 14,
......@@ -83,6 +89,42 @@ static UINT8 simple_mmx_permutation[64]={
0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
};
/* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */
UINT32 inverse[256]={
0, 4294967295U,2147483648U,1431655766, 1073741824, 858993460, 715827883, 613566757,
536870912, 477218589, 429496730, 390451573, 357913942, 330382100, 306783379, 286331154,
268435456, 252645136, 238609295, 226050911, 214748365, 204522253, 195225787, 186737709,
178956971, 171798692, 165191050, 159072863, 153391690, 148102321, 143165577, 138547333,
134217728, 130150525, 126322568, 122713352, 119304648, 116080198, 113025456, 110127367,
107374183, 104755300, 102261127, 99882961, 97612894, 95443718, 93368855, 91382283,
89478486, 87652394, 85899346, 84215046, 82595525, 81037119, 79536432, 78090315,
76695845, 75350304, 74051161, 72796056, 71582789, 70409300, 69273667, 68174085,
67108864, 66076420, 65075263, 64103990, 63161284, 62245903, 61356676, 60492498,
59652324, 58835169, 58040099, 57266231, 56512728, 55778797, 55063684, 54366675,
53687092, 53024288, 52377650, 51746594, 51130564, 50529028, 49941481, 49367441,
48806447, 48258060, 47721859, 47197443, 46684428, 46182445, 45691142, 45210183,
44739243, 44278014, 43826197, 43383509, 42949673, 42524429, 42107523, 41698712,
41297763, 40904451, 40518560, 40139882, 39768216, 39403370, 39045158, 38693400,
38347923, 38008561, 37675152, 37347542, 37025581, 36709123, 36398028, 36092163,
35791395, 35495598, 35204650, 34918434, 34636834, 34359739, 34087043, 33818641,
33554432, 33294321, 33038210, 32786010, 32537632, 32292988, 32051995, 31814573,
31580642, 31350127, 31122952, 30899046, 30678338, 30460761, 30246249, 30034737,
29826162, 29620465, 29417585, 29217465, 29020050, 28825284, 28633116, 28443493,
28256364, 28071682, 27889399, 27709467, 27531842, 27356480, 27183338, 27012373,
26843546, 26676816, 26512144, 26349493, 26188825, 26030105, 25873297, 25718368,
25565282, 25414008, 25264514, 25116768, 24970741, 24826401, 24683721, 24542671,
24403224, 24265352, 24129030, 23994231, 23860930, 23729102, 23598722, 23469767,
23342214, 23216040, 23091223, 22967740, 22845571, 22724695, 22605092, 22486740,
22369622, 22253717, 22139007, 22025474, 21913099, 21801865, 21691755, 21582751,
21474837, 21367997, 21262215, 21157475, 21053762, 20951060, 20849356, 20748635,
20648882, 20550083, 20452226, 20355296, 20259280, 20164166, 20069941, 19976593,
19884108, 19792477, 19701685, 19611723, 19522579, 19434242, 19346700, 19259944,
19173962, 19088744, 19004281, 18920561, 18837576, 18755316, 18673771, 18592933,
18512791, 18433337, 18354562, 18276457, 18199014, 18122225, 18046082, 17970575,
17895698, 17821442, 17747799, 17674763, 17602325, 17530479, 17459217, 17388532,
17318417, 17248865, 17179870, 17111424, 17043522, 16976156, 16909321, 16843010,
};
/* used to skip zeros at the end */
UINT8 zigzag_end[64];
......@@ -515,6 +557,9 @@ void dsputil_init(void)
else
for(i=0; i<64; i++) permutation[i]=i;
for(i=0; i<64; i++) inv_zigzag_direct16[zigzag_direct[i]]= i+1;
for(i=0; i<64; i++) zigzag_direct_noperm[i]= zigzag_direct[i];
if (use_permuted_idct) {
/* permute for IDCT */
for(i=0;i<64;i++) {
......
libavcodec/i386/mpegvideo_mmx.c
View file @ 2f349de2
......@@ -22,9 +22,16 @@
#include "../dsputil.h"
#include "../mpegvideo.h"
#include "../avcodec.h"
#include "../mangle.h"
extern UINT8 zigzag_end[64];
extern void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w);
extern int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale);
extern UINT8 zigzag_direct_noperm[64];
extern UINT16 inv_zigzag_direct16[64];
extern UINT32 inverse[256];
#if 0
......@@ -252,7 +259,7 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s,
}
} else {
i = 0;
unquant_even:
// unquant_even:
quant_matrix = s->non_intra_matrix;
/* Align on 4 elements boundary */
while(i&7)
......@@ -411,6 +418,20 @@ static void draw_edges_mmx(UINT8 *buf, int wrap, int width, int height, int w)
}
}
static volatile int esp_temp;
void unused_var_warning_killer(){
esp_temp++;
}
#undef HAVE_MMX2
#define RENAME(a) a ## _MMX
#include "mpegvideo_mmx_template.c"
#define HAVE_MMX2
#undef RENAME
#define RENAME(a) a ## _MMX2
#include "mpegvideo_mmx_template.c"
void MPV_common_init_mmx(MpegEncContext *s)
{
......@@ -421,5 +442,11 @@ void MPV_common_init_mmx(MpegEncContext *s)
s->dct_unquantize = dct_unquantize_mpeg1_mmx;
draw_edges = draw_edges_mmx;
if(mm_flags & MM_MMXEXT){
dct_quantize= dct_quantize_MMX2;
}else{
dct_quantize= dct_quantize_MMX;
}
}
}
libavcodec/i386/mpegvideo_mmx_template.c 0 → 100644
View file @ 2f349de2
/*
Copyright (C) 2002 Michael Niedermayer <michaelni@gmx.at>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#undef SPREADW
#undef PMAXW
#ifdef HAVE_MMX2
#define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t"
#define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t"
#else
#define SPREADW(a) \
"punpcklwd " #a ", " #a " \n\t"\
"punpcklwd " #a ", " #a " \n\t"
#define PMAXW(a,b) \
"psubusw " #a ", " #b " \n\t"\
"paddw " #a ", " #b " \n\t"
#endif
static int RENAME(dct_quantize)(MpegEncContext *s,
DCTELEM *block, int n,
int qscale)
{
int i, level, last_non_zero_p1, q;
const UINT16 *qmat;
static __align8 INT16 temp_block[64];
int minLevel, maxLevel;
if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){
/* mpeg4 */
minLevel= -2048;
maxLevel= 2047;
}else if(s->out_format==FMT_MPEG1){
/* mpeg1 */
minLevel= -255;
maxLevel= 255;
}else{
/* h263 / msmpeg4 */
minLevel= -128;
maxLevel= 127;
}
av_fdct (block);
if (s->mb_intra) {
int dummy;
if (n < 4)
q = s->y_dc_scale;
else
q = s->c_dc_scale;
/* note: block[0] is assumed to be positive */
#if 1
asm volatile (
"xorl %%edx, %%edx \n\t"
"mul %%ebx \n\t"
: "=d" (temp_block[0]), "=a"(dummy)
: "a" (block[0] + (q >> 1)), "b" (inverse[q])
);
#else
asm volatile (
"xorl %%edx, %%edx \n\t"
"divw %%bx \n\t"
"movzwl %%ax, %%eax \n\t"
: "=a" (temp_block[0])
: "a" (block[0] + (q >> 1)), "b" (q)
: "%edx"
);
#endif
// temp_block[0] = (block[0] + (q >> 1)) / q;
i = 1;
last_non_zero_p1 = 1;
if (s->out_format == FMT_H263) {
qmat = s->q_non_intra_matrix16;
} else {
qmat = s->q_intra_matrix16;
}
for(i=1;i<4;i++) {
level = block[i] * qmat[i];
level = level / (1 << (QMAT_SHIFT_MMX - 3));
/* XXX: currently, this code is not optimal. the range should be:
mpeg1: -255..255
mpeg2: -2048..2047
h263: -128..127
mpeg4: -2048..2047
*/
if (level > maxLevel)
level = maxLevel;
else if (level < minLevel)
level = minLevel;
temp_block[i] = level;
if(level)
if(last_non_zero_p1 < inv_zigzag_direct16[i]) last_non_zero_p1= inv_zigzag_direct16[i];
block[i]=0;
}
} else {
i = 0;
last_non_zero_p1 = 0;
qmat = s->q_non_intra_matrix16;
}
asm volatile( /* XXX: small rounding bug, but it shouldnt matter */
"movd %3, %%mm3 \n\t"
SPREADW(%%mm3)
"movd %4, %%mm4 \n\t"
SPREADW(%%mm4)
"movd %5, %%mm5 \n\t"
SPREADW(%%mm5)
"pxor %%mm7, %%mm7 \n\t"
"movd %%eax, %%mm2 \n\t"
SPREADW(%%mm2)
"movl %6, %%eax \n\t"
".balign 16 \n\t"
"1: \n\t"
"movq (%1, %%eax), %%mm0 \n\t"
"movq (%2, %%eax), %%mm1 \n\t"
"movq %%mm0, %%mm6 \n\t"
"psraw $15, %%mm6 \n\t"
"pmulhw %%mm0, %%mm1 \n\t"
"psubsw %%mm6, %%mm1 \n\t"
#ifdef HAVE_MMX2
"pminsw %%mm3, %%mm1 \n\t"
"pmaxsw %%mm4, %%mm1 \n\t"
#else
"paddsw %%mm3, %%mm1 \n\t"
"psubusw %%mm4, %%mm1 \n\t"
"paddsw %%mm5, %%mm1 \n\t"
#endif
"movq %%mm1, (%8, %%eax) \n\t"
"pcmpeqw %%mm7, %%mm1 \n\t"
"movq (%7, %%eax), %%mm0 \n\t"
"movq %%mm7, (%1, %%eax) \n\t"
"pandn %%mm0, %%mm1 \n\t"
PMAXW(%%mm1, %%mm2)
"addl $8, %%eax \n\t"
" js 1b \n\t"
"movq %%mm2, %%mm0 \n\t"
"psrlq $32, %%mm2 \n\t"
PMAXW(%%mm0, %%mm2)
"movq %%mm2, %%mm0 \n\t"
"psrlq $16, %%mm2 \n\t"
PMAXW(%%mm0, %%mm2)
"movd %%mm2, %%eax \n\t"
"movzbl %%al, %%eax \n\t"
: "+a" (last_non_zero_p1)
: "r" (block+64), "r" (qmat+64),
#ifdef HAVE_MMX2
"m" (maxLevel), "m" (minLevel), "m" (0 /* dummy */), "g" (2*i - 128),
#else
"m" (0x7FFF - maxLevel), "m" (0x7FFF -maxLevel + minLevel), "m" (minLevel), "g" (2*i - 128),
#endif
"r" (inv_zigzag_direct16+64), "r" (temp_block+64)
);
// last_non_zero_p1=64;
/* permute for IDCT */
asm volatile(
"movl %0, %%eax \n\t"
"pushl %%ebp \n\t"
"movl %%esp, " MANGLE(esp_temp) "\n\t"
"1: \n\t"
"movzbl (%1, %%eax), %%ebx \n\t"
"movzbl 1(%1, %%eax), %%ebp \n\t"
"movw (%2, %%ebx, 2), %%cx \n\t"
"movw (%2, %%ebp, 2), %%sp \n\t"
"movzbl " MANGLE(permutation) "(%%ebx), %%ebx\n\t"
"movzbl " MANGLE(permutation) "(%%ebp), %%ebp\n\t"
"movw %%cx, (%3, %%ebx, 2) \n\t"
"movw %%sp, (%3, %%ebp, 2) \n\t"
"addl $2, %%eax \n\t"
" js 1b \n\t"
"movl " MANGLE(esp_temp) ", %%esp\n\t"
"popl %%ebp \n\t"
:
: "g" (-last_non_zero_p1), "d" (zigzag_direct_noperm+last_non_zero_p1), "S" (temp_block), "D" (block)
: "%eax", "%ebx", "%ecx"
);
/*
for(i=0; i<last_non_zero_p1; i++)
{
int j= zigzag_direct_noperm[i];
block[block_permute_op(j)]= temp_block[j];
}
*/
//block_permute(block);
return last_non_zero_p1 - 1;
}
libavcodec/mpegvideo.c
View file @ 2f349de2
......@@ -35,12 +35,10 @@ static void dct_unquantize_mpeg1_c(MpegEncContext *s,
DCTELEM *block, int n, int qscale);
static void dct_unquantize_h263_c(MpegEncContext *s,
DCTELEM *block, int n, int qscale);
static int dct_quantize(MpegEncContext *s, DCTELEM *block, int n, int qscale);
static int dct_quantize_mmx(MpegEncContext *s,
DCTELEM *block, int n,
int qscale);
static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale);
int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c;
void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;
#define EDGE_WIDTH 16
......@@ -74,29 +72,29 @@ int motion_estimation_method = ME_LOG;
extern UINT8 zigzag_end[64];
/* XXX: should use variable shift ? */
#define QMAT_SHIFT_MMX 19
#define QMAT_SHIFT 25
static void convert_matrix(int *qmat, const UINT16 *quant_matrix, int qscale)
static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale)
{
int i;
if (av_fdct == jpeg_fdct_ifast) {
for(i=0;i<64;i++) {
/* 16 <= qscale * quant_matrix[i] <= 7905 */
/* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
/* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
/* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
/* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
qmat[i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
(aanscales[i] * qscale * quant_matrix[i]));
qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
(aanscales[i] * qscale * quant_matrix[block_permute_op(i)]));
}
} else {
for(i=0;i<64;i++) {
/* We can safely suppose that 16 <= quant_matrix[i] <= 255
So 16 <= qscale * quant_matrix[i] <= 7905
so (1 << QMAT_SHIFT) / 16 >= qmat[i] >= (1 << QMAT_SHIFT) / 7905
So 16 <= qscale * quant_matrix[i] <= 7905
so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
*/
qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
}
}
}
......@@ -418,7 +416,7 @@ void MPV_frame_start(MpegEncContext *s)
void MPV_frame_end(MpegEncContext *s)
{
/* draw edge for correct motion prediction if outside */
if (s->pict_type != B_TYPE) {
if (s->pict_type != B_TYPE && !s->intra_only) {
if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4){
draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
......@@ -457,7 +455,7 @@ int MPV_encode_picture(AVCodecContext *avctx,
avctx->key_frame = (s->pict_type == I_TYPE);
MPV_frame_start(s);
for(i=0;i<3;i++) {
UINT8 *src = pict->data[i];
UINT8 *dest = s->current_picture[i];
......@@ -472,11 +470,15 @@ int MPV_encode_picture(AVCodecContext *avctx,
h >>= 1;
}
for(j=0;j<h;j++) {
memcpy(dest, src, w);
dest += dest_wrap;
src += src_wrap;
}
if(s->intra_only && dest_wrap==src_wrap){
s->current_picture[i] = pict->data[i];
}else {
for(j=0;j<h;j++) {
memcpy(dest, src, w);
dest += dest_wrap;
src += src_wrap;
}
}
s->new_picture[i] = s->current_picture[i];
}
......@@ -873,10 +875,10 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->intra_matrix[0] = default_intra_matrix[0];
for(i=1;i<64;i++)
s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3;
convert_matrix(s->q_intra_matrix, s->intra_matrix, 8);
convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8);
} else {
convert_matrix(s->q_intra_matrix, s->intra_matrix, s->qscale);
convert_matrix(s->q_non_intra_matrix, s->non_intra_matrix, s->qscale);
convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale);
convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale);
}
switch(s->out_format) {
......@@ -1011,14 +1013,8 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->y_dc_scale = 8;
s->c_dc_scale = 8;
}
for(i=0;i<6;i++) {
int last_index;
if (av_fdct == jpeg_fdct_ifast)
last_index = dct_quantize(s, s->block[i], i, s->qscale);
else
last_index = dct_quantize_mmx(s, s->block[i], i, s->qscale);
s->block_last_index[i] = last_index;
s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale);
}
/* huffman encode */
......@@ -1060,7 +1056,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
// fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
}
static int dct_quantize(MpegEncContext *s,
static int dct_quantize_c(MpegEncContext *s,
DCTELEM *block, int n,
int qscale)
{
......@@ -1157,85 +1153,7 @@ static int dct_quantize(MpegEncContext *s,
level = maxLevel;
else if (level < minLevel)
level = minLevel;
block[j] = level;
last_non_zero = i;
} else {
block[j] = 0;
}
}
return last_non_zero;
}
static int dct_quantize_mmx(MpegEncContext *s,
DCTELEM *block, int n,
int qscale)
{
int i, j, level, last_non_zero, q;
const int *qmat;
int minLevel, maxLevel;
if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){
/* mpeg4 */
minLevel= -2048;
maxLevel= 2047;
}else if(s->out_format==FMT_MPEG1){
/* mpeg1 */
minLevel= -255;
maxLevel= 255;
}else{
/* h263 / msmpeg4 */
minLevel= -128;
maxLevel= 127;
}
av_fdct (block);
/* we need this permutation so that we correct the IDCT
permutation. will be moved into DCT code */
block_permute(block);
if (s->mb_intra) {
if (n < 4)
q = s->y_dc_scale;
else
q = s->c_dc_scale;
/* note: block[0] is assumed to be positive */
block[0] = (block[0] + (q >> 1)) / q;
i = 1;
last_non_zero = 0;
if (s->out_format == FMT_H263) {
qmat = s->q_non_intra_matrix;
} else {
qmat = s->q_intra_matrix;
}
} else {
i = 0;
last_non_zero = -1;
qmat = s->q_non_intra_matrix;
}
for(;i<64;i++) {
j = zigzag_direct[i];
level = block[j];
level = level * qmat[j];
/* XXX: slight error for the low range. Test should be equivalent to
(level <= -(1 << (QMAT_SHIFT_MMX - 3)) || level >= (1 <<
(QMAT_SHIFT_MMX - 3)))
*/
if (((level << (31 - (QMAT_SHIFT_MMX - 3))) >> (31 - (QMAT_SHIFT_MMX - 3))) !=
level) {
level = level / (1 << (QMAT_SHIFT_MMX - 3));
/* XXX: currently, this code is not optimal. the range should be:
mpeg1: -255..255
mpeg2: -2048..2047
h263: -128..127
mpeg4: -2048..2047
*/
if (level > maxLevel)
level = maxLevel;
else if (level < minLevel)
level = minLevel;
block[j] = level;
last_non_zero = i;
} else {
......
libavcodec/mpegvideo.h
View file @ 2f349de2
......@@ -30,6 +30,9 @@ enum OutputFormat {
#define MPEG_BUF_SIZE (16 * 1024)
#define QMAT_SHIFT_MMX 19
#define QMAT_SHIFT 25
typedef struct MpegEncContext {
struct AVCodecContext *avctx;
/* the following parameters must be initialized before encoding */
......@@ -120,6 +123,9 @@ typedef struct MpegEncContext {
/* precomputed matrix (combine qscale and DCT renorm) */
int q_intra_matrix[64];
int q_non_intra_matrix[64];
/* identical to the above but for MMX & these are not permutated */
UINT16 __align8 q_intra_matrix16[64] ;
UINT16 __align8 q_non_intra_matrix16[64];
int block_last_index[6]; /* last non zero coefficient in block */
void *opaque; /* private data for the user */
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment