]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/libx264.c
avcodec/libx264: add a check for the prft AVCodecContext export_side_data flag
[ffmpeg] / libavcodec / libx264.c
index a3493f393da2a4f786949834fc1397ea9e0f1d84..a08fe0ce76d7b7579f21fac6fb3a65c46d8f4587 100644 (file)
@@ -25,6 +25,7 @@
 #include "libavutil/mem.h"
 #include "libavutil/pixdesc.h"
 #include "libavutil/stereo3d.h"
+#include "libavutil/time.h"
 #include "libavutil/intreadwrite.h"
 #include "avcodec.h"
 #include "internal.h"
 // blocks of pixels (with respect to the luma plane)
 #define MB_SIZE 16
 
+typedef struct X264Opaque {
+    int64_t reordered_opaque;
+    int64_t wallclock;
+} X264Opaque;
+
 typedef struct X264Context {
     AVClass        *class;
     x264_param_t    params;
@@ -95,10 +101,16 @@ typedef struct X264Context {
     int scenechange_threshold;
     int noise_reduction;
 
-    char *x264_params;
+    AVDictionary *x264_params;
 
     int nb_reordered_opaque, next_reordered_opaque;
-    int64_t *reordered_opaque;
+    X264Opaque *reordered_opaque;
+
+    /**
+     * If the encoder does not support ROI then warn the first time we
+     * encounter a frame with ROI side data.
+     */
+    int roi_warned;
 } X264Context;
 
 static void X264_log(void *p, int level, const char *fmt, va_list args)
@@ -285,16 +297,19 @@ static int X264_frame(AVCodecContext *ctx, AVPacket *pkt, const AVFrame *frame,
     int nnal, i, ret;
     x264_picture_t pic_out = {0};
     int pict_type;
-    int64_t *out_opaque;
+    int bit_depth;
+    int64_t wallclock = 0;
+    X264Opaque *out_opaque;
     AVFrameSideData *sd;
 
     x264_picture_init( &x4->pic );
     x4->pic.img.i_csp   = x4->params.i_csp;
 #if X264_BUILD >= 153
-    if (x4->params.i_bitdepth > 8)
+    bit_depth = x4->params.i_bitdepth;
 #else
-    if (x264_bit_depth > 8)
+    bit_depth = x264_bit_depth;
 #endif
+    if (bit_depth > 8)
         x4->pic.img.i_csp |= X264_CSP_HIGH_DEPTH;
     x4->pic.img.i_plane = avfmt2_num_planes(ctx->pix_fmt);
 
@@ -306,7 +321,10 @@ static int X264_frame(AVCodecContext *ctx, AVPacket *pkt, const AVFrame *frame,
 
         x4->pic.i_pts  = frame->pts;
 
-        x4->reordered_opaque[x4->next_reordered_opaque] = frame->reordered_opaque;
+        x4->reordered_opaque[x4->next_reordered_opaque].reordered_opaque = frame->reordered_opaque;
+        x4->reordered_opaque[x4->next_reordered_opaque].wallclock = wallclock;
+        if (ctx->export_side_data & AV_CODEC_EXPORT_DATA_PRFT)
+            x4->reordered_opaque[x4->next_reordered_opaque].wallclock = av_gettime();
         x4->pic.opaque = &x4->reordered_opaque[x4->next_reordered_opaque];
         x4->next_reordered_opaque++;
         x4->next_reordered_opaque %= x4->nb_reordered_opaque;
@@ -354,56 +372,67 @@ static int X264_frame(AVCodecContext *ctx, AVPacket *pkt, const AVFrame *frame,
         sd = av_frame_get_side_data(frame, AV_FRAME_DATA_REGIONS_OF_INTEREST);
         if (sd) {
             if (x4->params.rc.i_aq_mode == X264_AQ_NONE) {
-                av_log(ctx, AV_LOG_WARNING, "Adaptive quantization must be enabled to use ROI encoding, skipping ROI.\n");
+                if (!x4->roi_warned) {
+                    x4->roi_warned = 1;
+                    av_log(ctx, AV_LOG_WARNING, "Adaptive quantization must be enabled to use ROI encoding, skipping ROI.\n");
+                }
             } else {
                 if (frame->interlaced_frame == 0) {
                     int mbx = (frame->width + MB_SIZE - 1) / MB_SIZE;
                     int mby = (frame->height + MB_SIZE - 1) / MB_SIZE;
+                    int qp_range = 51 + 6 * (bit_depth - 8);
                     int nb_rois;
-                    AVRegionOfInterest* roi;
-                    float* qoffsets;
+                    const AVRegionOfInterest *roi;
+                    uint32_t roi_size;
+                    float *qoffsets;
+
+                    roi = (const AVRegionOfInterest*)sd->data;
+                    roi_size = roi->self_size;
+                    if (!roi_size || sd->size % roi_size != 0) {
+                        av_log(ctx, AV_LOG_ERROR, "Invalid AVRegionOfInterest.self_size.\n");
+                        return AVERROR(EINVAL);
+                    }
+                    nb_rois = sd->size / roi_size;
+
                     qoffsets = av_mallocz_array(mbx * mby, sizeof(*qoffsets));
                     if (!qoffsets)
                         return AVERROR(ENOMEM);
 
-                    nb_rois = sd->size / sizeof(AVRegionOfInterest);
-                    roi = (AVRegionOfInterest*)sd->data;
-                    for (int count = 0; count < nb_rois; count++) {
-                        int starty = FFMIN(mby, roi->top / MB_SIZE);
-                        int endy   = FFMIN(mby, (roi->bottom + MB_SIZE - 1)/ MB_SIZE);
-                        int startx = FFMIN(mbx, roi->left / MB_SIZE);
-                        int endx   = FFMIN(mbx, (roi->right + MB_SIZE - 1)/ MB_SIZE);
+                    // This list must be iterated in reverse because the first
+                    // region in the list applies when regions overlap.
+                    for (int i = nb_rois - 1; i >= 0; i--) {
+                        int startx, endx, starty, endy;
                         float qoffset;
 
+                        roi = (const AVRegionOfInterest*)(sd->data + roi_size * i);
+
+                        starty = FFMIN(mby, roi->top / MB_SIZE);
+                        endy   = FFMIN(mby, (roi->bottom + MB_SIZE - 1)/ MB_SIZE);
+                        startx = FFMIN(mbx, roi->left / MB_SIZE);
+                        endx   = FFMIN(mbx, (roi->right + MB_SIZE - 1)/ MB_SIZE);
+
                         if (roi->qoffset.den == 0) {
                             av_free(qoffsets);
-                            av_log(ctx, AV_LOG_ERROR, "AVRegionOfInterest.qoffset.den should not be zero.\n");
+                            av_log(ctx, AV_LOG_ERROR, "AVRegionOfInterest.qoffset.den must not be zero.\n");
                             return AVERROR(EINVAL);
                         }
                         qoffset = roi->qoffset.num * 1.0f / roi->qoffset.den;
-                        qoffset = av_clipf(qoffset, -1.0f, 1.0f);
-
-                        // 25 is a number that I think it is a possible proper scale value.
-                        qoffset = qoffset * 25;
+                        qoffset = av_clipf(qoffset * qp_range, -qp_range, +qp_range);
 
                         for (int y = starty; y < endy; y++) {
                             for (int x = startx; x < endx; x++) {
                                 qoffsets[x + y*mbx] = qoffset;
                             }
                         }
-
-                        if (roi->self_size == 0) {
-                            av_free(qoffsets);
-                            av_log(ctx, AV_LOG_ERROR, "AVRegionOfInterest.self_size should be set to sizeof(AVRegionOfInterest).\n");
-                            return AVERROR(EINVAL);
-                        }
-                        roi = (AVRegionOfInterest*)((char*)roi + roi->self_size);
                     }
 
                     x4->pic.prop.quant_offsets = qoffsets;
                     x4->pic.prop.quant_offsets_free = av_free;
                 } else {
-                    av_log(ctx, AV_LOG_WARNING, "interlaced_frame not supported for ROI encoding yet, skipping ROI.\n");
+                    if (!x4->roi_warned) {
+                        x4->roi_warned = 1;
+                        av_log(ctx, AV_LOG_WARNING, "interlaced_frame not supported for ROI encoding yet, skipping ROI.\n");
+                    }
                 }
             }
         }
@@ -424,7 +453,8 @@ static int X264_frame(AVCodecContext *ctx, AVPacket *pkt, const AVFrame *frame,
     out_opaque = pic_out.opaque;
     if (out_opaque >= x4->reordered_opaque &&
         out_opaque < &x4->reordered_opaque[x4->nb_reordered_opaque]) {
-        ctx->reordered_opaque = *out_opaque;
+        ctx->reordered_opaque = out_opaque->reordered_opaque;
+        wallclock = out_opaque->wallclock;
     } else {
         // Unexpected opaque pointer on picture output
         ctx->reordered_opaque = 0;
@@ -454,6 +484,8 @@ FF_ENABLE_DEPRECATION_WARNINGS
     pkt->flags |= AV_PKT_FLAG_KEY*pic_out.b_keyframe;
     if (ret) {
         ff_side_data_set_encoder_stats(pkt, (pic_out.i_qpplus1 - 1) * FF_QP2LAMBDA, NULL, 0, pict_type);
+        if (wallclock)
+            ff_side_data_set_prft(pkt, wallclock);
 
 #if FF_API_CODED_FRAME
 FF_DISABLE_DEPRECATION_WARNINGS
@@ -587,6 +619,10 @@ static av_cold int X264_init(AVCodecContext *avctx)
     PARSE_X264_OPT("weightp", wpredp);
 
     if (avctx->bit_rate) {
+        if (avctx->bit_rate / 1000 > INT_MAX || avctx->rc_max_rate / 1000 > INT_MAX) {
+            av_log(avctx, AV_LOG_ERROR, "bit_rate and rc_max_rate > %d000 not supported by libx264\n", INT_MAX);
+            return AVERROR(EINVAL);
+        }
         x4->params.rc.i_bitrate   = avctx->bit_rate / 1000;
         x4->params.rc.i_rc_method = X264_RC_ABR;
     }
@@ -821,8 +857,13 @@ FF_ENABLE_DEPRECATION_WARNINGS
     x4->params.vui.i_sar_height = sh;
     x4->params.i_timebase_den = avctx->time_base.den;
     x4->params.i_timebase_num = avctx->time_base.num;
-    x4->params.i_fps_num = avctx->time_base.den;
-    x4->params.i_fps_den = avctx->time_base.num * avctx->ticks_per_frame;
+    if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {
+        x4->params.i_fps_num = avctx->framerate.num;
+        x4->params.i_fps_den = avctx->framerate.den;
+    } else {
+        x4->params.i_fps_num = avctx->time_base.den;
+        x4->params.i_fps_den = avctx->time_base.num * avctx->ticks_per_frame;
+    }
 
     x4->params.analyse.b_psnr = avctx->flags & AV_CODEC_FLAG_PSNR;
 
@@ -864,19 +905,14 @@ FF_ENABLE_DEPRECATION_WARNINGS
         }
     }
 
-    if (x4->x264_params) {
-        AVDictionary *dict    = NULL;
-        AVDictionaryEntry *en = NULL;
-
-        if (!av_dict_parse_string(&dict, x4->x264_params, "=", ":", 0)) {
-            while ((en = av_dict_get(dict, "", en, AV_DICT_IGNORE_SUFFIX))) {
-                if (x264_param_parse(&x4->params, en->key, en->value) < 0)
-                    av_log(avctx, AV_LOG_WARNING,
-                           "Error parsing option '%s = %s'.\n",
-                            en->key, en->value);
-            }
 
-            av_dict_free(&dict);
+    {
+        AVDictionaryEntry *en = NULL;
+        while (en = av_dict_get(x4->x264_params, "", en, AV_DICT_IGNORE_SUFFIX)) {
+           if (x264_param_parse(&x4->params, en->key, en->value) < 0)
+               av_log(avctx, AV_LOG_WARNING,
+                      "Error parsing option '%s = %s'.\n",
+                       en->key, en->value);
         }
     }
 
@@ -886,7 +922,7 @@ FF_ENABLE_DEPRECATION_WARNINGS
     if (avctx->max_b_frames < 0)
         avctx->max_b_frames = 0;
 
-    avctx->bit_rate = x4->params.rc.i_bitrate*1000;
+    avctx->bit_rate = x4->params.rc.i_bitrate*1000LL;
 
     x4->enc = x264_encoder_open(&x4->params);
     if (!x4->enc)
@@ -923,8 +959,8 @@ FF_ENABLE_DEPRECATION_WARNINGS
     if (!cpb_props)
         return AVERROR(ENOMEM);
     cpb_props->buffer_size = x4->params.rc.i_vbv_buffer_size * 1000;
-    cpb_props->max_bitrate = x4->params.rc.i_vbv_max_bitrate * 1000;
-    cpb_props->avg_bitrate = x4->params.rc.i_bitrate         * 1000;
+    cpb_props->max_bitrate = x4->params.rc.i_vbv_max_bitrate * 1000LL;
+    cpb_props->avg_bitrate = x4->params.rc.i_bitrate         * 1000LL;
 
     // Overestimate the reordered opaque buffer size, in case a runtime
     // reconfigure would increase the delay (which it shouldn't).
@@ -1088,7 +1124,7 @@ static const AVOption options[] = {
     { "sc_threshold", "Scene change threshold",                           OFFSET(scenechange_threshold), AV_OPT_TYPE_INT, { .i64 = -1 }, INT_MIN, INT_MAX, VE },
     { "noise_reduction", "Noise reduction",                               OFFSET(noise_reduction), AV_OPT_TYPE_INT, { .i64 = -1 }, INT_MIN, INT_MAX, VE },
 
-    { "x264-params",  "Override the x264 configuration using a :-separated list of key=value parameters", OFFSET(x264_params), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE },
+    { "x264-params",  "Override the x264 configuration using a :-separated list of key=value parameters", OFFSET(x264_params), AV_OPT_TYPE_DICT, { 0 }, 0, 0, VE },
     { NULL },
 };
 
@@ -1152,7 +1188,11 @@ AVCodec ff_libx264_encoder = {
     .priv_class       = &x264_class,
     .defaults         = x264_defaults,
     .init_static_data = X264_init_static,
+#if X264_BUILD >= 158
+    .caps_internal    = FF_CODEC_CAP_INIT_CLEANUP | FF_CODEC_CAP_INIT_THREADSAFE,
+#else
     .caps_internal    = FF_CODEC_CAP_INIT_CLEANUP,
+#endif
     .wrapper_name     = "libx264",
 };
 #endif
@@ -1179,6 +1219,11 @@ AVCodec ff_libx264rgb_encoder = {
     .priv_class     = &rgbclass,
     .defaults       = x264_defaults,
     .pix_fmts       = pix_fmts_8bit_rgb,
+#if X264_BUILD >= 158
+    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP | FF_CODEC_CAP_INIT_THREADSAFE,
+#else
+    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP,
+#endif
     .wrapper_name   = "libx264",
 };
 #endif