view libvo/jpeg_enc.c @ 27645:83d915449a10

Remove IWMMXT optimizations through libavcodec from libmpeg2. According to Siarhei Siamashka libavcodec is faster on ARM so it is better to use it directly instead of creating this hackish mix of two libraries. Plus, these local changes would never be acceptable upstream, so no good reason for keeping it in our local patchset remains.

/*
 * straightforward (to be) optimized JPEG encoder for the YUV422 format
 * based on MJPEG code from FFmpeg
 *
 * For an excellent introduction to the JPEG format, see:
 * http://www.ece.purdue.edu/~bouman/grad-labs/lab8/pdf/lab.pdf
 *
 * Copyright (c) 2002, Rik Snel
 * parts from FFmpeg Copyright (c) 2000-2002 Fabrice Bellard
 *
 * This file is part of MPlayer.
 *
 * MPlayer 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.
 *
 * MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
 */



#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
#include "config.h"
#include "mp_msg.h"
/* We need this #define because we need ../libavcodec/common.h to #define 
 * be2me_32, otherwise the linker will complain that it doesn't exist */
#define HAVE_AV_CONFIG_H
#include "libavcodec/avcodec.h"
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
#include "libavcodec/mjpegenc.h"

#include "jpeg_enc.h"

extern int avcodec_initialized;


/* Begin excessive code duplication ************************************/
/* Code coming from mpegvideo.c and mjpeg.c in ../libavcodec ***********/

static const unsigned short aanscales[64] = {
    /* precomputed values scaled up by 14 bits */
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
    21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
    19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
    8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
    4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
};

static void convert_matrix(MpegEncContext *s, int (*qmat)[64], 
		uint16_t (*qmat16)[2][64], const uint16_t *quant_matrix,
		int bias, int qmin, int qmax)
{
    int qscale;

    for(qscale=qmin; qscale<=qmax; qscale++){
        int i;
	if (s->dsp.fdct == ff_jpeg_fdct_islow) {
		for (i = 0; i < 64; i++) {
			const int j = s->dsp.idct_permutation[i];
			/* 16    <= qscale * quant_matrix[i] <= 7905 
			 * 19952 <= aanscales[i] *  \
			 * 	        qscale * quant_matrix[i]     <= 205026 
			 * (1<<36)/19952 >= (1<<36)/(aanscales[i] * \
			 * 	qscale * quant_matrix[i]) >= (1<<36)/249205025
			 * 3444240       >= (1<<36)/(aanscales[i] *
			 *      qscale * quant_matrix[i]) >= 275              */
			qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT-3))/
					(qscale * quant_matrix[j]));
		}
	} else if (s->dsp.fdct == fdct_ifast) {
            for(i=0;i<64;i++) {
                const int j = s->dsp.idct_permutation[i];
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
                /* 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[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / 
                                (aanscales[i] * qscale * quant_matrix[j]));
            }
        } else {
            for(i=0;i<64;i++) {
		const int j = s->dsp.idct_permutation[i];
                /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                   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  [qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]));
                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);

                if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
                qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
            }
        }
    }
}

static inline void encode_dc(MpegEncContext *s, int val, 
                             uint8_t *huff_size, uint16_t *huff_code)
{
    int mant, nbits;

    if (val == 0) {
        put_bits(&s->pb, huff_size[0], huff_code[0]);
    } else {
        mant = val;
        if (val < 0) {
            val = -val;
            mant--;
        }
        
        /* compute the log (XXX: optimize) */
        nbits = 0;
        while (val != 0) {
            val = val >> 1;
            nbits++;
        }
            
        put_bits(&s->pb, huff_size[nbits], huff_code[nbits]);
        
        put_bits(&s->pb, nbits, mant & ((1 << nbits) - 1));
    }
}

static void encode_block(MpegEncContext *s, DCTELEM *block, int n)
{
    int mant, nbits, code, i, j;
    int component, dc, run, last_index, val;
    MJpegContext *m = s->mjpeg_ctx;
    uint8_t *huff_size_ac;
    uint16_t *huff_code_ac;
    
    /* DC coef */
    component = (n <= 3 ? 0 : n - 4 + 1);
    dc = block[0]; /* overflow is impossible */
    val = dc - s->last_dc[component];
    if (n < 4) {
        encode_dc(s, val, m->huff_size_dc_luminance, m->huff_code_dc_luminance);
        huff_size_ac = m->huff_size_ac_luminance;
        huff_code_ac = m->huff_code_ac_luminance;
    } else {
        encode_dc(s, val, m->huff_size_dc_chrominance, m->huff_code_dc_chrominance);
        huff_size_ac = m->huff_size_ac_chrominance;
        huff_code_ac = m->huff_code_ac_chrominance;
    }
    s->last_dc[component] = dc;
    
    /* AC coefs */
    
    run = 0;
    last_index = s->block_last_index[n];
    for(i=1;i<=last_index;i++) {
        j = s->intra_scantable.permutated[i];
        val = block[j];
        if (val == 0) {
            run++;
        } else {
            while (run >= 16) {
                put_bits(&s->pb, huff_size_ac[0xf0], huff_code_ac[0xf0]);
                run -= 16;
            }
            mant = val;
            if (val < 0) {
                val = -val;
                mant--;
            }
            
            /* compute the log (XXX: optimize) */
            nbits = 0;
            while (val != 0) {
                val = val >> 1;
                nbits++;
            }
            code = (run << 4) | nbits;

            put_bits(&s->pb, huff_size_ac[code], huff_code_ac[code]);
        
            put_bits(&s->pb, nbits, mant & ((1 << nbits) - 1));
            run = 0;
        }
    }

    /* output EOB only if not already 64 values */
    if (last_index < 63 || run != 0)
        put_bits(&s->pb, huff_size_ac[0], huff_code_ac[0]);
}

static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
{
    int i;
    const int maxlevel= s->max_qcoeff;
    const int minlevel= s->min_qcoeff;

    for(i=0; i<=last_index; i++){
        const int j = s->intra_scantable.permutated[i];
        int level = block[j];
       
        if     (level>maxlevel) level=maxlevel;
        else if(level<minlevel) level=minlevel;
        block[j]= level;
    }
}

/* End excessive code duplication **************************************/

/* this function is a reproduction of the one in mjpeg, it includes two
 * changes, it allows for black&white encoding (it skips the U and V
 * macroblocks and it outputs the huffman code for 'no change' (dc) and
 * 'all zero' (ac)) and it takes 4 macroblocks (422) instead of 6 (420) */
static void zr_mjpeg_encode_mb(jpeg_enc_t *j) {

	MJpegContext *m = j->s->mjpeg_ctx;

	encode_block(j->s, j->s->block[0], 0);
	encode_block(j->s, j->s->block[1], 1);
	if (j->bw) {
		/* U */
		put_bits(&j->s->pb, m->huff_size_dc_chrominance[0],
				m->huff_code_dc_chrominance[0]);
		put_bits(&j->s->pb, m->huff_size_ac_chrominance[0],
				m->huff_code_ac_chrominance[0]);
		/* V */
		put_bits(&j->s->pb, m->huff_size_dc_chrominance[0],
				m->huff_code_dc_chrominance[0]);
		put_bits(&j->s->pb, m->huff_size_ac_chrominance[0],
				m->huff_code_ac_chrominance[0]);
    	} else {
		/* we trick encode_block here so that it uses
		 * chrominance huffman tables instead of luminance ones 
		 * (see the effect of second argument of encode_block) */
		encode_block(j->s, j->s->block[2], 4); 
		encode_block(j->s, j->s->block[3], 5);
    	}
}

/* this function can take all kinds of YUV colorspaces
 * YV12, YVYU, UYVY. The necesary parameters must be set up by the caller
 * y_ps means "y pixel size", y_rs means "y row size".
 * For YUYV, for example, is u_buf = y_buf + 1, v_buf = y_buf + 3, 
 * y_ps = 2, u_ps = 4, v_ps = 4, y_rs = u_rs = v_rs.
 *
 *  The actual buffers must be passed with mjpeg_encode_frame, this is
 *  to make it possible to call encode on the buffer provided by the
 *  codec in draw_frame.
 *  
 * The data is straightened out at the moment it is put in DCT
 * blocks, there are therefore no spurious memcopies involved */
/* Notice that w must be a multiple of 16 and h must be a multiple of 8 */
/* We produce YUV422 jpegs, the colors must be subsampled horizontally,
 * if the colors are also subsampled vertically, then this function
 * performs cheap upsampling (better solution will be: a DCT that is
 * optimized in the case that every two rows are the same) */
/* cu = 0 means 'No cheap upsampling'
 * cu = 1 means 'perform cheap upsampling' */
/* The encoder doesn't know anything about interlacing, the halve height
 * needs to be passed and the double rowstride. Which field gets encoded
 * is decided by what buffers are passed to mjpeg_encode_frame */
jpeg_enc_t *jpeg_enc_init(int w, int h, int y_psize, int y_rsize, 
		int u_psize, int u_rsize, int v_psize, int v_rsize,
		int cu, int q, int b) {
	jpeg_enc_t *j;
	int i = 0;
	mp_msg(MSGT_VO, MSGL_V, "JPEnc init: %dx%d %d %d %d %d %d %d\n",
			w, h, y_psize, y_rsize, u_psize, 
			u_rsize, v_psize, v_rsize);

	j = av_malloc(sizeof(jpeg_enc_t));
	if (j == NULL) return NULL;

	j->s = av_malloc(sizeof(MpegEncContext));
	memset(j->s,0x00,sizeof(MpegEncContext));
	if (j->s == NULL) {
		av_free(j);
		return NULL;
	}

	/* info on how to access the pixels */
	j->y_ps = y_psize; 
	j->u_ps = u_psize; 
	j->v_ps = v_psize;
	j->y_rs = y_rsize; 
	j->u_rs = u_rsize; 
	j->v_rs = v_rsize;

	j->s->width = w;
	j->s->height = h;
	j->s->qscale = q;

	j->s->out_format = FMT_MJPEG;
	j->s->intra_only = 1;
	j->s->encoding = 1;
	j->s->pict_type = FF_I_TYPE;
	j->s->y_dc_scale = 8;
	j->s->c_dc_scale = 8;

	//FIXME j->s->mjpeg_write_tables = 1;
	j->s->mjpeg_vsample[0] = 1;
	j->s->mjpeg_vsample[1] = 1;
	j->s->mjpeg_vsample[2] = 1;
	j->s->mjpeg_hsample[0] = 2;
	j->s->mjpeg_hsample[1] = 1;
	j->s->mjpeg_hsample[2] = 1;

	j->cheap_upsample = cu;
	j->bw = b;

	/* if libavcodec is used by the decoder then we must not
	 * initialize again, but if it is not initialized then we must
	 * initialize it here. */
	if (!avcodec_initialized) {
		/* we need to initialize libavcodec */
		avcodec_init();
		avcodec_register_all();
		avcodec_initialized=1;
	}

	if (ff_mjpeg_encode_init(j->s) < 0) {
		av_free(j->s);
		av_free(j);
		return NULL;
	}

	/* alloc bogus avctx to keep MPV_common_init from segfaulting */
	j->s->avctx = calloc(sizeof(*j->s->avctx), 1);
	/* Set up to encode mjpeg */
	j->s->avctx->codec_id = CODEC_ID_MJPEG;

	/* make MPV_common_init allocate important buffers, like s->block */
	j->s->avctx->thread_count = 1;

	if (MPV_common_init(j->s) < 0) {
		av_free(j->s);
		av_free(j);
		return NULL;
	}

	/* correct the value for sc->mb_height */
	j->s->mb_height = j->s->height/8;
	j->s->mb_intra = 1;

	j->s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
	for (i = 1; i < 64; i++) 
		j->s->intra_matrix[i] = av_clip_uint8(
			(ff_mpeg1_default_intra_matrix[i]*j->s->qscale) >> 3);
	convert_matrix(j->s, j->s->q_intra_matrix, j->s->q_intra_matrix16, 
			j->s->intra_matrix, j->s->intra_quant_bias, 8, 8);
	return j;
}	

int jpeg_enc_frame(jpeg_enc_t *j, unsigned char *y_data, 
		unsigned char *u_data, unsigned char *v_data, char *bufr) {
	int i, k, mb_x, mb_y, overflow;
	short int *dest;
	unsigned char *source;
	/* initialize the buffer */

	init_put_bits(&j->s->pb, bufr, 1024*256);

	ff_mjpeg_encode_picture_header(j->s);

	j->s->header_bits = put_bits_count(&j->s->pb);

	j->s->last_dc[0] = 128; 
	j->s->last_dc[1] = 128; 
	j->s->last_dc[2] = 128;

	for (mb_y = 0; mb_y < j->s->mb_height; mb_y++) {
		for (mb_x = 0; mb_x < j->s->mb_width; mb_x++) {
			/* conversion 8 to 16 bit and filling of blocks
			 * must be mmx optimized */
			/* fill 2 Y macroblocks and one U and one V */
			source = mb_y * 8 * j->y_rs + 
				16 * j->y_ps * mb_x + y_data;
			dest = j->s->block[0];
			for (i = 0; i < 8; i++) {
				for (k = 0; k < 8; k++) {
					dest[k] = source[k*j->y_ps];
				}
				dest += 8;
				source += j->y_rs;
			}
			source = mb_y * 8 * j->y_rs + 
				(16*mb_x + 8)*j->y_ps + y_data;
			dest = j->s->block[1];
			for (i = 0; i < 8; i++) {
				for (k = 0; k < 8; k++) {
					dest[k] = source[k*j->y_ps];
				}
				dest += 8;
				source += j->y_rs;
			}
			if (!j->bw && j->cheap_upsample) {
				source = mb_y*4*j->u_rs + 
					8*mb_x*j->u_ps + u_data;
				dest = j->s->block[2];
				for (i = 0; i < 4; i++) {
					for (k = 0; k < 8; k++) {
						dest[k] = source[k*j->u_ps];
						dest[k+8] = source[k*j->u_ps];
					}
					dest += 16;
					source += j->u_rs;
				}
				source = mb_y*4*j->v_rs + 
					8*mb_x*j->v_ps + v_data;
				dest = j->s->block[3];
				for (i = 0; i < 4; i++) {
					for (k = 0; k < 8; k++) {
						dest[k] = source[k*j->v_ps];
						dest[k+8] = source[k*j->v_ps];
					}
					dest += 16;
					source += j->u_rs;
				}
			} else if (!j->bw && !j->cheap_upsample) {
				source = mb_y*8*j->u_rs + 
					8*mb_x*j->u_ps + u_data;
				dest = j->s->block[2];
				for (i = 0; i < 8; i++) {
					for (k = 0; k < 8; k++) 
						dest[k] = source[k*j->u_ps];
					dest += 8;
					source += j->u_rs;
				}
				source = mb_y*8*j->v_rs + 
					8*mb_x*j->v_ps + v_data;
				dest = j->s->block[3];
				for (i = 0; i < 8; i++) {
					for (k = 0; k < 8; k++) 
						dest[k] = source[k*j->v_ps];
					dest += 8;
					source += j->u_rs;
				}
			}
			emms_c(); /* is this really needed? */

			j->s->block_last_index[0] = 
				j->s->dct_quantize(j->s, j->s->block[0], 
						0, 8, &overflow);
			if (overflow) clip_coeffs(j->s, j->s->block[0], 
					j->s->block_last_index[0]);
			j->s->block_last_index[1] = 
				j->s->dct_quantize(j->s, j->s->block[1], 
						1, 8, &overflow);
			if (overflow) clip_coeffs(j->s, j->s->block[1], 
					j->s->block_last_index[1]);

			if (!j->bw) {
				j->s->block_last_index[4] =
					j->s->dct_quantize(j->s, j->s->block[2],
							4, 8, &overflow);
				if (overflow) clip_coeffs(j->s, j->s->block[2], 
						j->s->block_last_index[2]);
				j->s->block_last_index[5] =
					j->s->dct_quantize(j->s, j->s->block[3],
							5, 8, &overflow);
				if (overflow) clip_coeffs(j->s, j->s->block[3], 
						j->s->block_last_index[3]);
			}
			zr_mjpeg_encode_mb(j);
		}
	}
	emms_c();
	ff_mjpeg_encode_picture_trailer(j->s);
	flush_put_bits(&j->s->pb);	

	//FIXME
	//if (j->s->mjpeg_write_tables == 1)
	//	j->s->mjpeg_write_tables = 0;
	
	return pbBufPtr(&(j->s->pb)) - j->s->pb.buf;
}

void jpeg_enc_uninit(jpeg_enc_t *j) {
	ff_mjpeg_encode_close(j->s);
	av_free(j->s);
	av_free(j);
}