yading@10: /*
yading@10:  * Lagarith lossless decoder
yading@10:  * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * Lagarith lossless decoder
yading@10:  * @author Nathan Caldwell
yading@10:  */
yading@10: 
yading@10: #include "avcodec.h"
yading@10: #include "get_bits.h"
yading@10: #include "mathops.h"
yading@10: #include "dsputil.h"
yading@10: #include "lagarithrac.h"
yading@10: #include "thread.h"
yading@10: 
yading@10: enum LagarithFrameType {
yading@10:     FRAME_RAW           = 1,    /**< uncompressed */
yading@10:     FRAME_U_RGB24       = 2,    /**< unaligned RGB24 */
yading@10:     FRAME_ARITH_YUY2    = 3,    /**< arithmetic coded YUY2 */
yading@10:     FRAME_ARITH_RGB24   = 4,    /**< arithmetic coded RGB24 */
yading@10:     FRAME_SOLID_GRAY    = 5,    /**< solid grayscale color frame */
yading@10:     FRAME_SOLID_COLOR   = 6,    /**< solid non-grayscale color frame */
yading@10:     FRAME_OLD_ARITH_RGB = 7,    /**< obsolete arithmetic coded RGB (no longer encoded by upstream since version 1.1.0) */
yading@10:     FRAME_ARITH_RGBA    = 8,    /**< arithmetic coded RGBA */
yading@10:     FRAME_SOLID_RGBA    = 9,    /**< solid RGBA color frame */
yading@10:     FRAME_ARITH_YV12    = 10,   /**< arithmetic coded YV12 */
yading@10:     FRAME_REDUCED_RES   = 11,   /**< reduced resolution YV12 frame */
yading@10: };
yading@10: 
yading@10: typedef struct LagarithContext {
yading@10:     AVCodecContext *avctx;
yading@10:     DSPContext dsp;
yading@10:     int zeros;                  /**< number of consecutive zero bytes encountered */
yading@10:     int zeros_rem;              /**< number of zero bytes remaining to output */
yading@10:     uint8_t *rgb_planes;
yading@10:     int rgb_stride;
yading@10: } LagarithContext;
yading@10: 
yading@10: /**
yading@10:  * Compute the 52bit mantissa of 1/(double)denom.
yading@10:  * This crazy format uses floats in an entropy coder and we have to match x86
yading@10:  * rounding exactly, thus ordinary floats aren't portable enough.
yading@10:  * @param denom denominator
yading@10:  * @return 52bit mantissa
yading@10:  * @see softfloat_mul
yading@10:  */
yading@10: static uint64_t softfloat_reciprocal(uint32_t denom)
yading@10: {
yading@10:     int shift = av_log2(denom - 1) + 1;
yading@10:     uint64_t ret = (1ULL << 52) / denom;
yading@10:     uint64_t err = (1ULL << 52) - ret * denom;
yading@10:     ret <<= shift;
yading@10:     err <<= shift;
yading@10:     err +=  denom / 2;
yading@10:     return ret + err / denom;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
yading@10:  * Used in combination with softfloat_reciprocal computes x/(double)denom.
yading@10:  * @param x 32bit integer factor
yading@10:  * @param mantissa mantissa of f with exponent 0
yading@10:  * @return 32bit integer value (x*f)
yading@10:  * @see softfloat_reciprocal
yading@10:  */
yading@10: static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
yading@10: {
yading@10:     uint64_t l = x * (mantissa & 0xffffffff);
yading@10:     uint64_t h = x * (mantissa >> 32);
yading@10:     h += l >> 32;
yading@10:     l &= 0xffffffff;
yading@10:     l += 1 << av_log2(h >> 21);
yading@10:     h += l >> 32;
yading@10:     return h >> 20;
yading@10: }
yading@10: 
yading@10: static uint8_t lag_calc_zero_run(int8_t x)
yading@10: {
yading@10:     return (x << 1) ^ (x >> 7);
yading@10: }
yading@10: 
yading@10: static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
yading@10: {
yading@10:     static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
yading@10:     int i;
yading@10:     int bit     = 0;
yading@10:     int bits    = 0;
yading@10:     int prevbit = 0;
yading@10:     unsigned val;
yading@10: 
yading@10:     for (i = 0; i < 7; i++) {
yading@10:         if (prevbit && bit)
yading@10:             break;
yading@10:         prevbit = bit;
yading@10:         bit = get_bits1(gb);
yading@10:         if (bit && !prevbit)
yading@10:             bits += series[i];
yading@10:     }
yading@10:     bits--;
yading@10:     if (bits < 0 || bits > 31) {
yading@10:         *value = 0;
yading@10:         return -1;
yading@10:     } else if (bits == 0) {
yading@10:         *value = 0;
yading@10:         return 0;
yading@10:     }
yading@10: 
yading@10:     val  = get_bits_long(gb, bits);
yading@10:     val |= 1 << bits;
yading@10: 
yading@10:     *value = val - 1;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
yading@10: {
yading@10:     int i, j, scale_factor;
yading@10:     unsigned prob, cumulative_target;
yading@10:     unsigned cumul_prob = 0;
yading@10:     unsigned scaled_cumul_prob = 0;
yading@10: 
yading@10:     rac->prob[0] = 0;
yading@10:     rac->prob[257] = UINT_MAX;
yading@10:     /* Read probabilities from bitstream */
yading@10:     for (i = 1; i < 257; i++) {
yading@10:         if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
yading@10:             av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
yading@10:             return -1;
yading@10:         }
yading@10:         if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
yading@10:             av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
yading@10:             return -1;
yading@10:         }
yading@10:         cumul_prob += rac->prob[i];
yading@10:         if (!rac->prob[i]) {
yading@10:             if (lag_decode_prob(gb, &prob)) {
yading@10:                 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
yading@10:                 return -1;
yading@10:             }
yading@10:             if (prob > 256 - i)
yading@10:                 prob = 256 - i;
yading@10:             for (j = 0; j < prob; j++)
yading@10:                 rac->prob[++i] = 0;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     if (!cumul_prob) {
yading@10:         av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     /* Scale probabilities so cumulative probability is an even power of 2. */
yading@10:     scale_factor = av_log2(cumul_prob);
yading@10: 
yading@10:     if (cumul_prob & (cumul_prob - 1)) {
yading@10:         uint64_t mul = softfloat_reciprocal(cumul_prob);
yading@10:         for (i = 1; i < 257; i++) {
yading@10:             rac->prob[i] = softfloat_mul(rac->prob[i], mul);
yading@10:             scaled_cumul_prob += rac->prob[i];
yading@10:         }
yading@10: 
yading@10:         scale_factor++;
yading@10:         cumulative_target = 1 << scale_factor;
yading@10: 
yading@10:         if (scaled_cumul_prob > cumulative_target) {
yading@10:             av_log(rac->avctx, AV_LOG_ERROR,
yading@10:                    "Scaled probabilities are larger than target!\n");
yading@10:             return -1;
yading@10:         }
yading@10: 
yading@10:         scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
yading@10: 
yading@10:         for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
yading@10:             if (rac->prob[i]) {
yading@10:                 rac->prob[i]++;
yading@10:                 scaled_cumul_prob--;
yading@10:             }
yading@10:             /* Comment from reference source:
yading@10:              * if (b & 0x80 == 0) {     // order of operations is 'wrong'; it has been left this way
yading@10:              *                          // since the compression change is negligible and fixing it
yading@10:              *                          // breaks backwards compatibility
yading@10:              *      b =- (signed int)b;
yading@10:              *      b &= 0xFF;
yading@10:              * } else {
yading@10:              *      b++;
yading@10:              *      b &= 0x7f;
yading@10:              * }
yading@10:              */
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     rac->scale = scale_factor;
yading@10: 
yading@10:     /* Fill probability array with cumulative probability for each symbol. */
yading@10:     for (i = 1; i < 257; i++)
yading@10:         rac->prob[i] += rac->prob[i - 1];
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
yading@10:                                       uint8_t *diff, int w, int *left,
yading@10:                                       int *left_top)
yading@10: {
yading@10:     /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
yading@10:      * However the &0xFF on the gradient predictor yealds incorrect output
yading@10:      * for lagarith.
yading@10:      */
yading@10:     int i;
yading@10:     uint8_t l, lt;
yading@10: 
yading@10:     l  = *left;
yading@10:     lt = *left_top;
yading@10: 
yading@10:     for (i = 0; i < w; i++) {
yading@10:         l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
yading@10:         lt = src1[i];
yading@10:         dst[i] = l;
yading@10:     }
yading@10: 
yading@10:     *left     = l;
yading@10:     *left_top = lt;
yading@10: }
yading@10: 
yading@10: static void lag_pred_line(LagarithContext *l, uint8_t *buf,
yading@10:                           int width, int stride, int line)
yading@10: {
yading@10:     int L, TL;
yading@10: 
yading@10:     if (!line) {
yading@10:         /* Left prediction only for first line */
yading@10:         L = l->dsp.add_hfyu_left_prediction(buf, buf,
yading@10:                                             width, 0);
yading@10:     } else {
yading@10:         /* Left pixel is actually prev_row[width] */
yading@10:         L = buf[width - stride - 1];
yading@10: 
yading@10:         if (line == 1) {
yading@10:             /* Second line, left predict first pixel, the rest of the line is median predicted
yading@10:              * NOTE: In the case of RGB this pixel is top predicted */
yading@10:             TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
yading@10:         } else {
yading@10:             /* Top left is 2 rows back, last pixel */
yading@10:             TL = buf[width - (2 * stride) - 1];
yading@10:         }
yading@10: 
yading@10:         add_lag_median_prediction(buf, buf - stride, buf,
yading@10:                                   width, &L, &TL);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
yading@10:                                int width, int stride, int line,
yading@10:                                int is_luma)
yading@10: {
yading@10:     int L, TL;
yading@10: 
yading@10:     if (!line) {
yading@10:         L= buf[0];
yading@10:         if (is_luma)
yading@10:             buf[0] = 0;
yading@10:         l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
yading@10:         if (is_luma)
yading@10:             buf[0] = L;
yading@10:         return;
yading@10:     }
yading@10:     if (line == 1) {
yading@10:         const int HEAD = is_luma ? 4 : 2;
yading@10:         int i;
yading@10: 
yading@10:         L  = buf[width - stride - 1];
yading@10:         TL = buf[HEAD  - stride - 1];
yading@10:         for (i = 0; i < HEAD; i++) {
yading@10:             L += buf[i];
yading@10:             buf[i] = L;
yading@10:         }
yading@10:         for (; i<width; i++) {
yading@10:             L     = mid_pred(L&0xFF, buf[i-stride], (L + buf[i-stride] - TL)&0xFF) + buf[i];
yading@10:             TL    = buf[i-stride];
yading@10:             buf[i]= L;
yading@10:         }
yading@10:     } else {
yading@10:         TL = buf[width - (2 * stride) - 1];
yading@10:         L  = buf[width - stride - 1];
yading@10:         l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
yading@10:                                         &L, &TL);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int lag_decode_line(LagarithContext *l, lag_rac *rac,
yading@10:                            uint8_t *dst, int width, int stride,
yading@10:                            int esc_count)
yading@10: {
yading@10:     int i = 0;
yading@10:     int ret = 0;
yading@10: 
yading@10:     if (!esc_count)
yading@10:         esc_count = -1;
yading@10: 
yading@10:     /* Output any zeros remaining from the previous run */
yading@10: handle_zeros:
yading@10:     if (l->zeros_rem) {
yading@10:         int count = FFMIN(l->zeros_rem, width - i);
yading@10:         memset(dst + i, 0, count);
yading@10:         i += count;
yading@10:         l->zeros_rem -= count;
yading@10:     }
yading@10: 
yading@10:     while (i < width) {
yading@10:         dst[i] = lag_get_rac(rac);
yading@10:         ret++;
yading@10: 
yading@10:         if (dst[i])
yading@10:             l->zeros = 0;
yading@10:         else
yading@10:             l->zeros++;
yading@10: 
yading@10:         i++;
yading@10:         if (l->zeros == esc_count) {
yading@10:             int index = lag_get_rac(rac);
yading@10:             ret++;
yading@10: 
yading@10:             l->zeros = 0;
yading@10: 
yading@10:             l->zeros_rem = lag_calc_zero_run(index);
yading@10:             goto handle_zeros;
yading@10:         }
yading@10:     }
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
yading@10:                                     const uint8_t *src, const uint8_t *src_end,
yading@10:                                     int width, int esc_count)
yading@10: {
yading@10:     int i = 0;
yading@10:     int count;
yading@10:     uint8_t zero_run = 0;
yading@10:     const uint8_t *src_start = src;
yading@10:     uint8_t mask1 = -(esc_count < 2);
yading@10:     uint8_t mask2 = -(esc_count < 3);
yading@10:     uint8_t *end = dst + (width - 2);
yading@10: 
yading@10: output_zeros:
yading@10:     if (l->zeros_rem) {
yading@10:         count = FFMIN(l->zeros_rem, width - i);
yading@10:         if (end - dst < count) {
yading@10:             av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
yading@10:             return AVERROR_INVALIDDATA;
yading@10:         }
yading@10: 
yading@10:         memset(dst, 0, count);
yading@10:         l->zeros_rem -= count;
yading@10:         dst += count;
yading@10:     }
yading@10: 
yading@10:     while (dst < end) {
yading@10:         i = 0;
yading@10:         while (!zero_run && dst + i < end) {
yading@10:             i++;
yading@10:             if (i+2 >= src_end - src)
yading@10:                 return AVERROR_INVALIDDATA;
yading@10:             zero_run =
yading@10:                 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
yading@10:         }
yading@10:         if (zero_run) {
yading@10:             zero_run = 0;
yading@10:             i += esc_count;
yading@10:             memcpy(dst, src, i);
yading@10:             dst += i;
yading@10:             l->zeros_rem = lag_calc_zero_run(src[i]);
yading@10: 
yading@10:             src += i + 1;
yading@10:             goto output_zeros;
yading@10:         } else {
yading@10:             memcpy(dst, src, i);
yading@10:             src += i;
yading@10:             dst += i;
yading@10:         }
yading@10:     }
yading@10:     return  src - src_start;
yading@10: }
yading@10: 
yading@10: 
yading@10: 
yading@10: static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
yading@10:                                   int width, int height, int stride,
yading@10:                                   const uint8_t *src, int src_size)
yading@10: {
yading@10:     int i = 0;
yading@10:     int read = 0;
yading@10:     uint32_t length;
yading@10:     uint32_t offset = 1;
yading@10:     int esc_count;
yading@10:     GetBitContext gb;
yading@10:     lag_rac rac;
yading@10:     const uint8_t *src_end = src + src_size;
yading@10: 
yading@10:     rac.avctx = l->avctx;
yading@10:     l->zeros = 0;
yading@10: 
yading@10:     if(src_size < 2)
yading@10:         return AVERROR_INVALIDDATA;
yading@10: 
yading@10:     esc_count = src[0];
yading@10:     if (esc_count < 4) {
yading@10:         length = width * height;
yading@10:         if(src_size < 5)
yading@10:             return AVERROR_INVALIDDATA;
yading@10:         if (esc_count && AV_RL32(src + 1) < length) {
yading@10:             length = AV_RL32(src + 1);
yading@10:             offset += 4;
yading@10:         }
yading@10: 
yading@10:         init_get_bits(&gb, src + offset, src_size * 8);
yading@10: 
yading@10:         if (lag_read_prob_header(&rac, &gb) < 0)
yading@10:             return -1;
yading@10: 
yading@10:         ff_lag_rac_init(&rac, &gb, length - stride);
yading@10: 
yading@10:         for (i = 0; i < height; i++)
yading@10:             read += lag_decode_line(l, &rac, dst + (i * stride), width,
yading@10:                                     stride, esc_count);
yading@10: 
yading@10:         if (read > length)
yading@10:             av_log(l->avctx, AV_LOG_WARNING,
yading@10:                    "Output more bytes than length (%d of %d)\n", read,
yading@10:                    length);
yading@10:     } else if (esc_count < 8) {
yading@10:         esc_count -= 4;
yading@10:         if (esc_count > 0) {
yading@10:             /* Zero run coding only, no range coding. */
yading@10:             for (i = 0; i < height; i++) {
yading@10:                 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
yading@10:                                                    src_end, width, esc_count);
yading@10:                 if (res < 0)
yading@10:                     return res;
yading@10:                 src += res;
yading@10:             }
yading@10:         } else {
yading@10:             if (src_size < width * height)
yading@10:                 return AVERROR_INVALIDDATA; // buffer not big enough
yading@10:             /* Plane is stored uncompressed */
yading@10:             for (i = 0; i < height; i++) {
yading@10:                 memcpy(dst + (i * stride), src, width);
yading@10:                 src += width;
yading@10:             }
yading@10:         }
yading@10:     } else if (esc_count == 0xff) {
yading@10:         /* Plane is a solid run of given value */
yading@10:         for (i = 0; i < height; i++)
yading@10:             memset(dst + i * stride, src[1], width);
yading@10:         /* Do not apply prediction.
yading@10:            Note: memset to 0 above, setting first value to src[1]
yading@10:            and applying prediction gives the same result. */
yading@10:         return 0;
yading@10:     } else {
yading@10:         av_log(l->avctx, AV_LOG_ERROR,
yading@10:                "Invalid zero run escape code! (%#x)\n", esc_count);
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
yading@10:         for (i = 0; i < height; i++) {
yading@10:             lag_pred_line(l, dst, width, stride, i);
yading@10:             dst += stride;
yading@10:         }
yading@10:     } else {
yading@10:         for (i = 0; i < height; i++) {
yading@10:             lag_pred_line_yuy2(l, dst, width, stride, i,
yading@10:                                width == l->avctx->width);
yading@10:             dst += stride;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Decode a frame.
yading@10:  * @param avctx codec context
yading@10:  * @param data output AVFrame
yading@10:  * @param data_size size of output data or 0 if no picture is returned
yading@10:  * @param avpkt input packet
yading@10:  * @return number of consumed bytes on success or negative if decode fails
yading@10:  */
yading@10: static int lag_decode_frame(AVCodecContext *avctx,
yading@10:                             void *data, int *got_frame, AVPacket *avpkt)
yading@10: {
yading@10:     const uint8_t *buf = avpkt->data;
yading@10:     unsigned int buf_size = avpkt->size;
yading@10:     LagarithContext *l = avctx->priv_data;
yading@10:     ThreadFrame frame = { .f = data };
yading@10:     AVFrame *const p  = data;
yading@10:     uint8_t frametype = 0;
yading@10:     uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
yading@10:     uint32_t offs[4];
yading@10:     uint8_t *srcs[4], *dst;
yading@10:     int i, j, planes = 3;
yading@10:     int ret;
yading@10: 
yading@10:     p->key_frame = 1;
yading@10: 
yading@10:     frametype = buf[0];
yading@10: 
yading@10:     offset_gu = AV_RL32(buf + 1);
yading@10:     offset_bv = AV_RL32(buf + 5);
yading@10: 
yading@10:     switch (frametype) {
yading@10:     case FRAME_SOLID_RGBA:
yading@10:         avctx->pix_fmt = AV_PIX_FMT_RGB32;
yading@10:     case FRAME_SOLID_GRAY:
yading@10:         if (frametype == FRAME_SOLID_GRAY)
yading@10:             if (avctx->bits_per_coded_sample == 24) {
yading@10:                 avctx->pix_fmt = AV_PIX_FMT_RGB24;
yading@10:             } else {
yading@10:                 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
yading@10:                 planes = 4;
yading@10:             }
yading@10: 
yading@10:         if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:         dst = p->data[0];
yading@10:         if (frametype == FRAME_SOLID_RGBA) {
yading@10:         for (j = 0; j < avctx->height; j++) {
yading@10:             for (i = 0; i < avctx->width; i++)
yading@10:                 AV_WN32(dst + i * 4, offset_gu);
yading@10:             dst += p->linesize[0];
yading@10:         }
yading@10:         } else {
yading@10:             for (j = 0; j < avctx->height; j++) {
yading@10:                 memset(dst, buf[1], avctx->width * planes);
yading@10:                 dst += p->linesize[0];
yading@10:             }
yading@10:         }
yading@10:         break;
yading@10:     case FRAME_ARITH_RGBA:
yading@10:         avctx->pix_fmt = AV_PIX_FMT_RGB32;
yading@10:         planes = 4;
yading@10:         offset_ry += 4;
yading@10:         offs[3] = AV_RL32(buf + 9);
yading@10:     case FRAME_ARITH_RGB24:
yading@10:     case FRAME_U_RGB24:
yading@10:         if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
yading@10:             avctx->pix_fmt = AV_PIX_FMT_RGB24;
yading@10: 
yading@10:         if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:         offs[0] = offset_bv;
yading@10:         offs[1] = offset_gu;
yading@10:         offs[2] = offset_ry;
yading@10: 
yading@10:         if (!l->rgb_planes) {
yading@10:             l->rgb_stride = FFALIGN(avctx->width, 16);
yading@10:             l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * 4 + 16);
yading@10:             if (!l->rgb_planes) {
yading@10:                 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
yading@10:                 return AVERROR(ENOMEM);
yading@10:             }
yading@10:         }
yading@10:         for (i = 0; i < planes; i++)
yading@10:             srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
yading@10:         for (i = 0; i < planes; i++)
yading@10:             if (buf_size <= offs[i]) {
yading@10:                 av_log(avctx, AV_LOG_ERROR,
yading@10:                         "Invalid frame offsets\n");
yading@10:                 return AVERROR_INVALIDDATA;
yading@10:             }
yading@10: 
yading@10:         for (i = 0; i < planes; i++)
yading@10:             lag_decode_arith_plane(l, srcs[i],
yading@10:                                    avctx->width, avctx->height,
yading@10:                                    -l->rgb_stride, buf + offs[i],
yading@10:                                    buf_size - offs[i]);
yading@10:         dst = p->data[0];
yading@10:         for (i = 0; i < planes; i++)
yading@10:             srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
yading@10:         for (j = 0; j < avctx->height; j++) {
yading@10:             for (i = 0; i < avctx->width; i++) {
yading@10:                 uint8_t r, g, b, a;
yading@10:                 r = srcs[0][i];
yading@10:                 g = srcs[1][i];
yading@10:                 b = srcs[2][i];
yading@10:                 r += g;
yading@10:                 b += g;
yading@10:                 if (frametype == FRAME_ARITH_RGBA) {
yading@10:                     a = srcs[3][i];
yading@10:                     AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
yading@10:                 } else {
yading@10:                     dst[i * 3 + 0] = r;
yading@10:                     dst[i * 3 + 1] = g;
yading@10:                     dst[i * 3 + 2] = b;
yading@10:                 }
yading@10:             }
yading@10:             dst += p->linesize[0];
yading@10:             for (i = 0; i < planes; i++)
yading@10:                 srcs[i] += l->rgb_stride;
yading@10:         }
yading@10:         break;
yading@10:     case FRAME_ARITH_YUY2:
yading@10:         avctx->pix_fmt = AV_PIX_FMT_YUV422P;
yading@10: 
yading@10:         if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:         if (offset_ry >= buf_size ||
yading@10:             offset_gu >= buf_size ||
yading@10:             offset_bv >= buf_size) {
yading@10:             av_log(avctx, AV_LOG_ERROR,
yading@10:                    "Invalid frame offsets\n");
yading@10:             return AVERROR_INVALIDDATA;
yading@10:         }
yading@10: 
yading@10:         lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
yading@10:                                p->linesize[0], buf + offset_ry,
yading@10:                                buf_size - offset_ry);
yading@10:         lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
yading@10:                                avctx->height, p->linesize[1],
yading@10:                                buf + offset_gu, buf_size - offset_gu);
yading@10:         lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
yading@10:                                avctx->height, p->linesize[2],
yading@10:                                buf + offset_bv, buf_size - offset_bv);
yading@10:         break;
yading@10:     case FRAME_ARITH_YV12:
yading@10:         avctx->pix_fmt = AV_PIX_FMT_YUV420P;
yading@10: 
yading@10:         if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
yading@10:             return ret;
yading@10:         if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
yading@10:             return AVERROR_INVALIDDATA;
yading@10:         }
yading@10: 
yading@10:         if (offset_ry >= buf_size ||
yading@10:             offset_gu >= buf_size ||
yading@10:             offset_bv >= buf_size) {
yading@10:             av_log(avctx, AV_LOG_ERROR,
yading@10:                    "Invalid frame offsets\n");
yading@10:             return AVERROR_INVALIDDATA;
yading@10:         }
yading@10: 
yading@10:         lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
yading@10:                                p->linesize[0], buf + offset_ry,
yading@10:                                buf_size - offset_ry);
yading@10:         lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
yading@10:                                avctx->height / 2, p->linesize[2],
yading@10:                                buf + offset_gu, buf_size - offset_gu);
yading@10:         lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
yading@10:                                avctx->height / 2, p->linesize[1],
yading@10:                                buf + offset_bv, buf_size - offset_bv);
yading@10:         break;
yading@10:     default:
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Unsupported Lagarith frame type: %#x\n", frametype);
yading@10:         return AVERROR_PATCHWELCOME;
yading@10:     }
yading@10: 
yading@10:     *got_frame = 1;
yading@10: 
yading@10:     return buf_size;
yading@10: }
yading@10: 
yading@10: static av_cold int lag_decode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     LagarithContext *l = avctx->priv_data;
yading@10:     l->avctx = avctx;
yading@10: 
yading@10:     ff_dsputil_init(&l->dsp, avctx);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold int lag_decode_end(AVCodecContext *avctx)
yading@10: {
yading@10:     LagarithContext *l = avctx->priv_data;
yading@10: 
yading@10:     av_freep(&l->rgb_planes);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: AVCodec ff_lagarith_decoder = {
yading@10:     .name           = "lagarith",
yading@10:     .type           = AVMEDIA_TYPE_VIDEO,
yading@10:     .id             = AV_CODEC_ID_LAGARITH,
yading@10:     .priv_data_size = sizeof(LagarithContext),
yading@10:     .init           = lag_decode_init,
yading@10:     .close          = lag_decode_end,
yading@10:     .decode         = lag_decode_frame,
yading@10:     .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
yading@10: };