]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/mjpegdec.c
x264: Add option to force IDR frames
[ffmpeg] / libavcodec / mjpegdec.c
index be32332ccb417d0932391804115ac06347cd31ef..d8fc9defa308b8433ce9368f7fccbb07a52382bf 100644 (file)
  * MJPEG decoder.
  */
 
-// #define DEBUG
 #include <assert.h>
 
 #include "libavutil/imgutils.h"
 #include "libavutil/opt.h"
 #include "avcodec.h"
+#include "blockdsp.h"
+#include "idctdsp.h"
 #include "internal.h"
+#include "jpegtables.h"
 #include "mjpeg.h"
 #include "mjpegdec.h"
 #include "jpeglsdec.h"
+#include "put_bits.h"
 
 
 static int build_vlc(VLC *vlc, const uint8_t *bits_table,
@@ -85,19 +88,26 @@ av_cold int ff_mjpeg_decode_init(AVCodecContext *avctx)
 {
     MJpegDecodeContext *s = avctx->priv_data;
 
-    if (!s->picture_ptr)
-        s->picture_ptr = &s->picture;
+    if (!s->picture_ptr) {
+        s->picture = av_frame_alloc();
+        if (!s->picture)
+            return AVERROR(ENOMEM);
+        s->picture_ptr = s->picture;
+    }
 
     s->avctx = avctx;
+    ff_blockdsp_init(&s->bdsp, avctx);
     ff_hpeldsp_init(&s->hdsp, avctx->flags);
-    ff_dsputil_init(&s->dsp, avctx);
-    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
+    ff_idctdsp_init(&s->idsp, avctx);
+    ff_init_scantable(s->idsp.idct_permutation, &s->scantable,
+                      ff_zigzag_direct);
     s->buffer_size   = 0;
     s->buffer        = NULL;
     s->start_code    = -1;
     s->first_picture = 1;
     s->org_height    = avctx->coded_height;
     avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
+    avctx->colorspace = AVCOL_SPC_BT470BG;
 
     build_basic_mjpeg_vlc(s);
 
@@ -212,18 +222,20 @@ int ff_mjpeg_decode_dht(MJpegDecodeContext *s)
 
 int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
 {
-    int len, nb_components, i, width, height, pix_fmt_id;
+    int h_count[MAX_COMPONENTS] = { 0 };
+    int v_count[MAX_COMPONENTS] = { 0 };
+    int len, nb_components, i, width, height, bits, pix_fmt_id, ret;
 
     /* XXX: verify len field validity */
     len     = get_bits(&s->gb, 16);
-    s->bits = get_bits(&s->gb, 8);
+    bits    = get_bits(&s->gb, 8);
 
     if (s->pegasus_rct)
-        s->bits = 9;
-    if (s->bits == 9 && !s->pegasus_rct)
+        bits = 9;
+    if (bits == 9 && !s->pegasus_rct)
         s->rct  = 1;    // FIXME ugly
 
-    if (s->bits != 8 && !s->lossless) {
+    if (bits != 8 && !s->lossless) {
         av_log(s->avctx, AV_LOG_ERROR, "only 8 bits/component accepted\n");
         return -1;
     }
@@ -243,7 +255,14 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
     if (nb_components <= 0 ||
         nb_components > MAX_COMPONENTS)
         return -1;
-    if (s->ls && !(s->bits <= 8 || nb_components == 1)) {
+    if (s->interlaced && (s->bottom_field == !s->interlace_polarity)) {
+        if (nb_components != s->nb_components) {
+            av_log(s->avctx, AV_LOG_ERROR,
+                   "nb_components changing in interlaced picture\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+    if (s->ls && !(bits <= 8 || nb_components == 1)) {
         avpriv_report_missing_feature(s->avctx,
                                       "JPEG-LS that is not <= 8 "
                                       "bits/component or 16-bit gray");
@@ -255,25 +274,25 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
     for (i = 0; i < nb_components; i++) {
         /* component id */
         s->component_id[i] = get_bits(&s->gb, 8) - 1;
-        s->h_count[i]      = get_bits(&s->gb, 4);
-        s->v_count[i]      = get_bits(&s->gb, 4);
+        h_count[i]         = get_bits(&s->gb, 4);
+        v_count[i]         = get_bits(&s->gb, 4);
         /* compute hmax and vmax (only used in interleaved case) */
-        if (s->h_count[i] > s->h_max)
-            s->h_max = s->h_count[i];
-        if (s->v_count[i] > s->v_max)
-            s->v_max = s->v_count[i];
+        if (h_count[i] > s->h_max)
+            s->h_max = h_count[i];
+        if (v_count[i] > s->v_max)
+            s->v_max = v_count[i];
         s->quant_index[i] = get_bits(&s->gb, 8);
         if (s->quant_index[i] >= 4)
             return AVERROR_INVALIDDATA;
-        if (!s->h_count[i] || !s->v_count[i]) {
+        if (!h_count[i] || !v_count[i]) {
             av_log(s->avctx, AV_LOG_ERROR,
                    "Invalid sampling factor in component %d %d:%d\n",
-                   i, s->h_count[i], s->v_count[i]);
+                   i, h_count[i], v_count[i]);
             return AVERROR_INVALIDDATA;
         }
 
         av_log(s->avctx, AV_LOG_DEBUG, "component %d %d:%d id: %d quant:%d\n",
-               i, s->h_count[i], s->v_count[i],
+               i, h_count[i], v_count[i],
                s->component_id[i], s->quant_index[i]);
     }
 
@@ -286,10 +305,14 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
         s->rgb = 1;
 
     /* if different size, realloc/alloc picture */
-    /* XXX: also check h_count and v_count */
-    if (width != s->width || height != s->height) {
+    if (width != s->width || height != s->height || bits != s->bits ||
+        memcmp(s->h_count, h_count, sizeof(h_count))                ||
+        memcmp(s->v_count, v_count, sizeof(v_count))) {
         s->width      = width;
         s->height     = height;
+        s->bits       = bits;
+        memcpy(s->h_count, h_count, sizeof(h_count));
+        memcpy(s->v_count, v_count, sizeof(v_count));
         s->interlaced = 0;
 
         /* test interlaced mode */
@@ -303,7 +326,9 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
             height *= 2;
         }
 
-        avcodec_set_dimensions(s->avctx, width, height);
+        ret = ff_set_dimensions(s->avctx, width, height);
+        if (ret < 0)
+            return ret;
 
         s->first_picture = 0;
     }
@@ -326,8 +351,10 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
     case 0x11111100:
         if (s->rgb)
             s->avctx->pix_fmt = AV_PIX_FMT_BGRA;
-        else
+        else {
             s->avctx->pix_fmt = s->cs_itu601 ? AV_PIX_FMT_YUV444P : AV_PIX_FMT_YUVJ444P;
+            s->avctx->color_range = s->cs_itu601 ? AVCOL_RANGE_MPEG : AVCOL_RANGE_JPEG;
+        }
         assert(s->nb_components == 3);
         break;
     case 0x11000000:
@@ -335,12 +362,15 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
         break;
     case 0x12111100:
         s->avctx->pix_fmt = s->cs_itu601 ? AV_PIX_FMT_YUV440P : AV_PIX_FMT_YUVJ440P;
+        s->avctx->color_range = s->cs_itu601 ? AVCOL_RANGE_MPEG : AVCOL_RANGE_JPEG;
         break;
     case 0x21111100:
         s->avctx->pix_fmt = s->cs_itu601 ? AV_PIX_FMT_YUV422P : AV_PIX_FMT_YUVJ422P;
+        s->avctx->color_range = s->cs_itu601 ? AVCOL_RANGE_MPEG : AVCOL_RANGE_JPEG;
         break;
     case 0x22111100:
         s->avctx->pix_fmt = s->cs_itu601 ? AV_PIX_FMT_YUV420P : AV_PIX_FMT_YUVJ420P;
+        s->avctx->color_range = s->cs_itu601 ? AVCOL_RANGE_MPEG : AVCOL_RANGE_JPEG;
         break;
     default:
         av_log(s->avctx, AV_LOG_ERROR, "Unhandled pixel format 0x%x\n", pix_fmt_id);
@@ -355,6 +385,12 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
             s->avctx->pix_fmt = AV_PIX_FMT_GRAY16;
     }
 
+    s->pix_desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
+    if (!s->pix_desc) {
+        av_log(s->avctx, AV_LOG_ERROR, "Could not get a pixel format descriptor.\n");
+        return AVERROR_BUG;
+    }
+
     av_frame_unref(s->picture_ptr);
     if (ff_get_buffer(s->avctx, s->picture_ptr, AV_GET_BUFFER_FLAG_REF) < 0) {
         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
@@ -367,7 +403,7 @@ int ff_mjpeg_decode_sof(MJpegDecodeContext *s)
     for (i = 0; i < 3; i++)
         s->linesize[i] = s->picture_ptr->linesize[i] << s->interlaced;
 
-    av_dlog(s->avctx, "%d %d %d %d %d %d\n",
+    ff_dlog(s->avctx, "%d %d %d %d %d %d\n",
             s->width, s->height, s->linesize[0], s->linesize[1],
             s->interlaced, s->avctx->height);
 
@@ -463,7 +499,7 @@ static int decode_dc_progressive(MJpegDecodeContext *s, int16_t *block,
                                  int16_t *quant_matrix, int Al)
 {
     int val;
-    s->dsp.clear_block(block);
+    s->bdsp.clear_block(block);
     val = mjpeg_decode_dc(s, dc_index);
     if (val == 0xffff) {
         av_log(s->avctx, AV_LOG_ERROR, "error dc\n");
@@ -812,26 +848,12 @@ static int mjpeg_decode_scan(MJpegDecodeContext *s, int nb_components, int Ah,
     if (mb_bitmask)
         init_get_bits(&mb_bitmask_gb, mb_bitmask, s->mb_width * s->mb_height);
 
-    if (s->flipped && s->avctx->flags & CODEC_FLAG_EMU_EDGE) {
-        av_log(s->avctx, AV_LOG_ERROR,
-               "Can not flip image with CODEC_FLAG_EMU_EDGE set!\n");
-        s->flipped = 0;
-    }
-
     for (i = 0; i < nb_components; i++) {
         int c   = s->comp_index[i];
         data[c] = s->picture_ptr->data[c];
         reference_data[c] = reference ? reference->data[c] : NULL;
         linesize[c] = s->linesize[c];
         s->coefs_finished[c] |= 1;
-        if (s->flipped) {
-            // picture should be flipped upside-down for this codec
-            int offset = (linesize[c] * (s->v_scount[i] *
-                         (8 * s->mb_height - ((s->height / s->v_max) & 7)) - 1));
-            data[c]           += offset;
-            reference_data[c] += offset;
-            linesize[c]       *= -1;
-        }
     }
 
     for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
@@ -869,7 +891,7 @@ static int mjpeg_decode_scan(MJpegDecodeContext *s, int nb_components, int Ah,
                                 reference_data[c] + block_offset,
                                 linesize[c], 8);
                         else {
-                            s->dsp.clear_block(s->block);
+                            s->bdsp.clear_block(s->block);
                             if (decode_block(s, s->block, i,
                                              s->dc_index[i], s->ac_index[i],
                                              s->quant_matrixes[s->quant_index[c]]) < 0) {
@@ -877,7 +899,7 @@ static int mjpeg_decode_scan(MJpegDecodeContext *s, int nb_components, int Ah,
                                        "error y=%d x=%d\n", mb_y, mb_x);
                                 return AVERROR_INVALIDDATA;
                             }
-                            s->dsp.idct_put(ptr, linesize[c], s->block);
+                            s->idsp.idct_put(ptr, linesize[c], s->block);
                         }
                     } else {
                         int block_idx  = s->block_stride[c] * (v * mb_y + y) +
@@ -894,8 +916,8 @@ static int mjpeg_decode_scan(MJpegDecodeContext *s, int nb_components, int Ah,
                             return AVERROR_INVALIDDATA;
                         }
                     }
-                    av_dlog(s->avctx, "mb: %d %d processed\n", mb_y, mb_x);
-                    av_dlog(s->avctx, "%d %d %d %d %d %d %d %d \n",
+                    ff_dlog(s->avctx, "mb: %d %d processed\n", mb_y, mb_x);
+                    ff_dlog(s->avctx, "%d %d %d %d %d %d %d %d \n",
                             mb_x, mb_y, x, y, c, s->bottom_field,
                             (v * mb_y + y) * 8, (h * mb_x + x) * 8);
                     if (++x == h) {
@@ -950,7 +972,9 @@ static int mjpeg_decode_scan_progressive_ac(MJpegDecodeContext *s, int ss,
         init_get_bits(&mb_bitmask_gb, mb_bitmask, s->mb_width * s->mb_height);
 
     if (!Al) {
-        s->coefs_finished[c] |= (1LL << (se + 1)) - (1LL << ss);
+        // s->coefs_finished is a bitmask for coefficients coded
+        // ss and se are parameters telling start and end coefficients
+        s->coefs_finished[c] |= (~0ULL >> (63 - (se - ss))) << ss;
         last_scan = !~s->coefs_finished[c];
     }
 
@@ -990,7 +1014,7 @@ static int mjpeg_decode_scan_progressive_ac(MJpegDecodeContext *s, int ss,
                                                  reference_data + block_offset,
                                                  linesize, 8);
                 } else {
-                    s->dsp.idct_put(ptr, linesize, *block);
+                    s->idsp.idct_put(ptr, linesize, *block);
                     ptr += 8;
                 }
             }
@@ -1129,7 +1153,7 @@ next_field:
         GetBitContext bak = s->gb;
         align_get_bits(&bak);
         if (show_bits(&bak, 16) == 0xFFD1) {
-            av_dlog(s->avctx, "AVRn interlaced picture marker found\n");
+            ff_dlog(s->avctx, "AVRn interlaced picture marker found\n");
             s->gb = bak;
             skip_bits(&s->gb, 16);
             s->bottom_field ^= 1;
@@ -1211,6 +1235,7 @@ static int mjpeg_decode_app(MJpegDecodeContext *s)
 
         s->avctx->sample_aspect_ratio.num = get_bits(&s->gb, 16);
         s->avctx->sample_aspect_ratio.den = get_bits(&s->gb, 16);
+        ff_set_sar(s->avctx, s->avctx->sample_aspect_ratio);
 
         if (s->avctx->debug & FF_DEBUG_PICT_INFO)
             av_log(s->avctx, AV_LOG_INFO,
@@ -1356,7 +1381,7 @@ static int find_marker(const uint8_t **pbuf_ptr, const uint8_t *buf_end)
     }
     val = -1;
 found:
-    av_dlog(NULL, "find_marker skipped %d bytes\n", skipped);
+    ff_dlog(NULL, "find_marker skipped %d bytes\n", skipped);
     *pbuf_ptr = buf_ptr;
     return val;
 }
@@ -1397,7 +1422,7 @@ int ff_mjpeg_find_marker(MJpegDecodeContext *s,
         *unescaped_buf_ptr  = s->buffer;
         *unescaped_buf_size = dst - s->buffer;
         memset(s->buffer + *unescaped_buf_size, 0,
-               FF_INPUT_BUFFER_PADDING_SIZE);
+               AV_INPUT_BUFFER_PADDING_SIZE);
 
         av_log(s->avctx, AV_LOG_DEBUG, "escaping removed %td bytes\n",
                (buf_end - *buf_ptr) - (dst - s->buffer));
@@ -1440,7 +1465,7 @@ int ff_mjpeg_find_marker(MJpegDecodeContext *s,
         *unescaped_buf_ptr  = dst;
         *unescaped_buf_size = (bit_count + 7) >> 3;
         memset(s->buffer + *unescaped_buf_size, 0,
-               FF_INPUT_BUFFER_PADDING_SIZE);
+               AV_INPUT_BUFFER_PADDING_SIZE);
     } else {
         *unescaped_buf_ptr  = *buf_ptr;
         *unescaped_buf_size = buf_end - *buf_ptr;
@@ -1452,6 +1477,7 @@ int ff_mjpeg_find_marker(MJpegDecodeContext *s,
 int ff_mjpeg_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
                           AVPacket *avpkt)
 {
+    AVFrame     *frame = data;
     const uint8_t *buf = avpkt->data;
     int buf_size       = avpkt->size;
     MJpegDecodeContext *s = avctx->priv_data;
@@ -1573,8 +1599,17 @@ eoi_parser:
                 if (s->bottom_field == !s->interlace_polarity)
                     goto not_the_end;
             }
-            if ((ret = av_frame_ref(data, s->picture_ptr)) < 0)
+            if ((ret = av_frame_ref(frame, s->picture_ptr)) < 0)
                 return ret;
+            if (s->flipped) {
+                int i;
+                for (i = 0; frame->data[i]; i++) {
+                    int h = frame->height >> ((i == 1 || i == 2) ?
+                                              s->pix_desc->log2_chroma_h : 0);
+                    frame->data[i] += frame->linesize[i] * (h - 1);
+                    frame->linesize[i] *= -1;
+                }
+            }
             *got_frame = 1;
 
             if (!s->lossless &&
@@ -1643,7 +1678,10 @@ av_cold int ff_mjpeg_decode_end(AVCodecContext *avctx)
     MJpegDecodeContext *s = avctx->priv_data;
     int i, j;
 
-    if (s->picture_ptr)
+    if (s->picture) {
+        av_frame_free(&s->picture);
+        s->picture_ptr = NULL;
+    } else if (s->picture_ptr)
         av_frame_unref(s->picture_ptr);
 
     av_free(s->buffer);
@@ -1678,25 +1716,27 @@ static const AVClass mjpegdec_class = {
 
 AVCodec ff_mjpeg_decoder = {
     .name           = "mjpeg",
+    .long_name      = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_MJPEG,
     .priv_data_size = sizeof(MJpegDecodeContext),
     .init           = ff_mjpeg_decode_init,
     .close          = ff_mjpeg_decode_end,
     .decode         = ff_mjpeg_decode_frame,
-    .capabilities   = CODEC_CAP_DR1,
-    .long_name      = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"),
+    .capabilities   = AV_CODEC_CAP_DR1,
     .priv_class     = &mjpegdec_class,
+    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
 };
 
 AVCodec ff_thp_decoder = {
     .name           = "thp",
+    .long_name      = NULL_IF_CONFIG_SMALL("Nintendo Gamecube THP video"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_THP,
     .priv_data_size = sizeof(MJpegDecodeContext),
     .init           = ff_mjpeg_decode_init,
     .close          = ff_mjpeg_decode_end,
     .decode         = ff_mjpeg_decode_frame,
-    .capabilities   = CODEC_CAP_DR1,
-    .long_name      = NULL_IF_CONFIG_SMALL("Nintendo Gamecube THP video"),
+    .capabilities   = AV_CODEC_CAP_DR1,
+    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
 };