Merge commit '019ab88a95cb31b698506d90e8ce56695a7f1cc5'

* commit '019ab88a95cb31b698506d90e8ce56695a7f1cc5':
  lavc: add an option for exporting cropping information to the caller

Merged-by: James Almer <jamrial@gmail.com>

doc/APIchanges

@@ -15,6 +15,10 @@ libavutil:     2015-08-28
 
 API changes, most recent first:
 
+2017-xx-xx - xxxxxxx - lavc 57.95.100 / 57.31.0 - avcodec.h
+  Add AVCodecContext.apply_cropping to control whether cropping
+  is handled by libavcodec or the caller.
+
 2017-xx-xx - xxxxxxx - lavu 55.62.100 / 55.30.0 - frame.h
   Add AVFrame.crop_left/right/top/bottom fields for attaching cropping
   information to video frames.

libavcodec/avcodec.h

@@ -3644,6 +3644,33 @@ typedef struct AVCodecContext {
      *             AVCodecContext.get_format callback)
      */
     int hwaccel_flags;
+
+    /**
+     * Video decoding only. Certain video codecs support cropping, meaning that
+     * only a sub-rectangle of the decoded frame is intended for display.  This
+     * option controls how cropping is handled by libavcodec.
+     *
+     * When set to 1 (the default), libavcodec will apply cropping internally.
+     * I.e. it will modify the output frame width/height fields and offset the
+     * data pointers (only by as much as possible while preserving alignment, or
+     * by the full amount if the AV_CODEC_FLAG_UNALIGNED flag is set) so that
+     * the frames output by the decoder refer only to the cropped area. The
+     * crop_* fields of the output frames will be zero.
+     *
+     * When set to 0, the width/height fields of the output frames will be set
+     * to the coded dimensions and the crop_* fields will describe the cropping
+     * rectangle. Applying the cropping is left to the caller.
+     *
+     * @warning When hardware acceleration with opaque output frames is used,
+     * libavcodec is unable to apply cropping from the top/left border.
+     *
+     * @note when this option is set to zero, the width/height fields of the
+     * AVCodecContext and output AVFrames have different meanings. The codec
+     * context fields store display dimensions (with the coded dimensions in
+     * coded_width/height), while the frame fields store the coded dimensions
+     * (with the display dimensions being determined by the crop_* fields).
+     */
+    int apply_cropping;
 } AVCodecContext;
 
 AVRational av_codec_get_pkt_timebase         (const AVCodecContext *avctx);

libavcodec/decode.c

@@ -35,6 +35,7 @@
 #include "libavutil/hwcontext.h"
 #include "libavutil/imgutils.h"
 #include "libavutil/internal.h"
+#include "libavutil/intmath.h"
 
 #include "avcodec.h"
 #include "bytestream.h"
@@ -682,6 +683,112 @@ int attribute_align_arg avcodec_send_packet(AVCodecContext *avctx, const AVPacke
     return 0;
 }
 
+static int calc_cropping_offsets(size_t offsets[4], const AVFrame *frame,
+                                 const AVPixFmtDescriptor *desc)
+{
+    int i, j;
+
+    for (i = 0; frame->data[i]; i++) {
+        const AVComponentDescriptor *comp = NULL;
+        int shift_x = (i == 1 || i == 2) ? desc->log2_chroma_w : 0;
+        int shift_y = (i == 1 || i == 2) ? desc->log2_chroma_h : 0;
+
+        if (desc->flags & (AV_PIX_FMT_FLAG_PAL | AV_PIX_FMT_FLAG_PSEUDOPAL) && i == 1) {
+            offsets[i] = 0;
+            break;
+        }
+
+        /* find any component descriptor for this plane */
+        for (j = 0; j < desc->nb_components; j++) {
+            if (desc->comp[j].plane == i) {
+                comp = &desc->comp[j];
+                break;
+            }
+        }
+        if (!comp)
+            return AVERROR_BUG;
+
+        offsets[i] = (frame->crop_top  >> shift_y) * frame->linesize[i] +
+                     (frame->crop_left >> shift_x) * comp->step;
+    }
+
+    return 0;
+}
+
+static int apply_cropping(AVCodecContext *avctx, AVFrame *frame)
+{
+    const AVPixFmtDescriptor *desc;
+    size_t offsets[4];
+    int i;
+
+    /* make sure we are noisy about decoders returning invalid cropping data */
+    if (frame->crop_left >= INT_MAX - frame->crop_right        ||
+        frame->crop_top  >= INT_MAX - frame->crop_bottom       ||
+        (frame->crop_left + frame->crop_right) >= frame->width ||
+        (frame->crop_top + frame->crop_bottom) >= frame->height) {
+        av_log(avctx, AV_LOG_WARNING,
+               "Invalid cropping information set by a decoder: "
+               "%"SIZE_SPECIFIER"/%"SIZE_SPECIFIER"/%"SIZE_SPECIFIER"/%"SIZE_SPECIFIER" "
+               "(frame size %dx%d). This is a bug, please report it\n",
+               frame->crop_left, frame->crop_right, frame->crop_top, frame->crop_bottom,
+               frame->width, frame->height);
+        frame->crop_left   = 0;
+        frame->crop_right  = 0;
+        frame->crop_top    = 0;
+        frame->crop_bottom = 0;
+        return 0;
+    }
+
+    if (!avctx->apply_cropping)
+        return 0;
+
+    desc = av_pix_fmt_desc_get(frame->format);
+    if (!desc)
+        return AVERROR_BUG;
+
+    /* Apply just the right/bottom cropping for hwaccel formats. Bitstream
+     * formats cannot be easily handled here either (and corresponding decoders
+     * should not export any cropping anyway), so do the same for those as well.
+     * */
+    if (desc->flags & (AV_PIX_FMT_FLAG_BITSTREAM | AV_PIX_FMT_FLAG_HWACCEL)) {
+        frame->width      -= frame->crop_right;
+        frame->height     -= frame->crop_bottom;
+        frame->crop_right  = 0;
+        frame->crop_bottom = 0;
+        return 0;
+    }
+
+    /* calculate the offsets for each plane */
+    calc_cropping_offsets(offsets, frame, desc);
+
+    /* adjust the offsets to avoid breaking alignment */
+    if (!(avctx->flags & AV_CODEC_FLAG_UNALIGNED)) {
+        int min_log2_align = INT_MAX;
+
+        for (i = 0; frame->data[i]; i++) {
+            int log2_align = offsets[i] ? ff_ctz(offsets[i]) : INT_MAX;
+            min_log2_align = FFMIN(log2_align, min_log2_align);
+        }
+
+        if (min_log2_align < 5) {
+            frame->crop_left &= ~((1 << min_log2_align) - 1);
+            calc_cropping_offsets(offsets, frame, desc);
+        }
+    }
+
+    for (i = 0; frame->data[i]; i++)
+        frame->data[i] += offsets[i];
+
+    frame->width      -= (frame->crop_left + frame->crop_right);
+    frame->height     -= (frame->crop_top  + frame->crop_bottom);
+    frame->crop_left   = 0;
+    frame->crop_right  = 0;
+    frame->crop_top    = 0;
+    frame->crop_bottom = 0;
+
+    return 0;
+}
+
 int attribute_align_arg avcodec_receive_frame(AVCodecContext *avctx, AVFrame *frame)
 {
     AVCodecInternal *avci = avctx->internal;
@@ -704,6 +811,14 @@ int attribute_align_arg avcodec_receive_frame(AVCodecContext *avctx, AVFrame *fr
             return ret;
     }
 
+    if (avctx->codec_type == AVMEDIA_TYPE_VIDEO) {
+        ret = apply_cropping(avctx, frame);
+        if (ret < 0) {
+            av_frame_unref(frame);
+            return ret;
+        }
+    }
+
     avctx->frame_number++;
 
     return 0;
@@ -1611,7 +1726,8 @@ static int get_buffer_internal(AVCodecContext *avctx, AVFrame *frame, int flags)
         validate_avframe_allocation(avctx, frame);
 
 end:
-    if (avctx->codec_type == AVMEDIA_TYPE_VIDEO && !override_dimensions) {
+    if (avctx->codec_type == AVMEDIA_TYPE_VIDEO && !override_dimensions &&
+        !(avctx->codec->caps_internal & FF_CODEC_CAP_EXPORTS_CROPPING)) {
         frame->width  = avctx->width;
         frame->height = avctx->height;
     }

libavcodec/internal.h

@@ -58,6 +58,12 @@
  * skipped due to the skip_frame setting.
  */
 #define FF_CODEC_CAP_SKIP_FRAME_FILL_PARAM  (1 << 3)
+/**
+ * The decoder sets the cropping fields in the output frames manually.
+ * If this cap is set, the generic code will initialize output frame
+ * dimensions to coded rather than display values.
+ */
+#define FF_CODEC_CAP_EXPORTS_CROPPING       (1 << 4)
 
 #ifdef TRACE
 #   define ff_tlog(ctx, ...) av_log(ctx, AV_LOG_TRACE, __VA_ARGS__)

libavcodec/options_table.h

@@ -561,6 +561,7 @@ static const AVOption avcodec_options[] = {
 #if FF_API_SIDEDATA_ONLY_PKT
 {"side_data_only_packets", NULL, OFFSET(side_data_only_packets), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, A|V|E },
 #endif
+{"apply_cropping", NULL, OFFSET(apply_cropping), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, V | D },
 {"skip_alpha", "Skip processing alpha", OFFSET(skip_alpha), AV_OPT_TYPE_BOOL, {.i64 = 0 }, 0, 1, V|D },
 {"field_order", "Field order", OFFSET(field_order), AV_OPT_TYPE_INT, {.i64 = AV_FIELD_UNKNOWN }, 0, 5, V|D|E, "field_order" },
 {"progressive", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = AV_FIELD_PROGRESSIVE }, 0, 0, V|D|E, "field_order" },

libavcodec/version.h

@@ -28,8 +28,8 @@
 #include "libavutil/version.h"
 
 #define LIBAVCODEC_VERSION_MAJOR  57
-#define LIBAVCODEC_VERSION_MINOR  94
+#define LIBAVCODEC_VERSION_MINOR  95
-#define LIBAVCODEC_VERSION_MICRO 101
+#define LIBAVCODEC_VERSION_MICRO 100
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \

tests/ref/fate/api-mjpeg-codec-param

@@ -149,6 +149,7 @@ stream=0, decode=0
     sub_text_format=1
     refcounted_frames=false
     side_data_only_packets=true
+    apply_cropping=true
     skip_alpha=false
     field_order=0
     dump_separator=
@@ -307,6 +308,7 @@ stream=0, decode=1
     sub_text_format=1
     refcounted_frames=false
     side_data_only_packets=true
+    apply_cropping=true
     skip_alpha=false
     field_order=0
     dump_separator=

tests/ref/fate/api-png-codec-param

@@ -149,6 +149,7 @@ stream=0, decode=0
     sub_text_format=1
     refcounted_frames=false
     side_data_only_packets=true
+    apply_cropping=true
     skip_alpha=false
     field_order=0
     dump_separator=
@@ -307,6 +308,7 @@ stream=0, decode=1
     sub_text_format=1
     refcounted_frames=false
     side_data_only_packets=true
+    apply_cropping=true
     skip_alpha=false
     field_order=0
     dump_separator=