VP9: Added support for video size change (scaling down) on inter frames. (#508)

* * rocDecode/VP9: Added support for video size change (scaling down) on inter frames.
 - A new reconfigure mode is added, where we keep the existing VAAPI surfaces and context, and work on the smaller images and store the images in the surfaces.

* * rocDecode/VP9 resize: Added changes based on review comments.

api/rocparser.h

@@ -46,6 +46,16 @@ extern "C" {
 typedef void *RocdecVideoParser;
 typedef uint64_t RocdecTimeStamp;
 
+/***************************************************************/
+//! \enum RocdecReconfigOptions
+//! Decoder reconfigure options for VP9 decode
+//! Used in RocdecVideoFormat structure
+/***************************************************************/
+typedef enum {
+    ROCDEC_RECONFIG_NEW_SURFACES  = 0,      /**< Allocate new surfaces during decoder reconfig                             */
+    ROCDEC_RECONFIG_KEEP_SURFACES = 1,      /**< Keep existing surfaces during decoder reconfig                            */
+} RocdecReconfigOptions;
+
 /**
  * @brief ROCDEC_VIDEO_FORMAT struct
  * @ingroup group_rocdec_struct
@@ -89,7 +99,6 @@ typedef struct {
         int right;  /**< OUT: right position of display rect   */
         int bottom; /**< OUT: bottom position of display rect  */
     } display_area;
-
     rocDecVideoChromaFormat chroma_format; /**< OUT:  Chroma format                   */
     uint32_t bitrate;                      /**< OUT: video bitrate (bps, 0=unknown)   */
                                             /**
@@ -111,7 +120,8 @@ typedef struct {
         uint8_t transfer_characteristics;  /**< OUT: opto-electronic transfer characteristic of the source picture */
         uint8_t matrix_coefficients;       /**< OUT: used in deriving luma and chroma signals from RGB primaries   */
     } video_signal_description;
-    uint32_t seqhdr_data_length; /**< OUT: Additional bytes following (RocdecVideoFormatEx)                  */
+    uint32_t seqhdr_data_length;           /**< OUT: Additional bytes following (RocdecVideoFormatEx)                  */
+    uint32_t reconfig_options;             /**< OUT: Refer to RocdecReconfigOptions */
 } RocdecVideoFormat;
 
 /****************************************************************/

src/parser/av1_parser.cpp

@@ -230,6 +230,7 @@ ParserResult Av1VideoParser::NotifyNewSequence(Av1SequenceHeader *p_seq_header,
     video_format_params_.display_aspect_ratio.x = disp_width / gcd;
     video_format_params_.display_aspect_ratio.y = disp_height / gcd;
 
+    video_format_params_.reconfig_options = ROCDEC_RECONFIG_NEW_SURFACES;
     video_format_params_.video_signal_description = {0};
     video_format_params_.seqhdr_data_length = 0;
 

src/parser/avc_parser.cpp

@@ -387,6 +387,7 @@ ParserResult AvcVideoParser::NotifyNewSps(AvcSeqParameterSet *p_sps) {
     video_format_params_.display_aspect_ratio.x = disp_width / gcd;
     video_format_params_.display_aspect_ratio.y = disp_height / gcd;
 
+    video_format_params_.reconfig_options = ROCDEC_RECONFIG_NEW_SURFACES;
     if (p_sps->vui_parameters_present_flag) {
         video_format_params_.video_signal_description.video_format = p_sps->vui_seq_parameters.video_format;
         video_format_params_.video_signal_description.video_full_range_flag = p_sps->vui_seq_parameters.video_full_range_flag;

src/parser/hevc_parser.cpp

@@ -190,6 +190,7 @@ int HevcVideoParser::FillSeqCallbackFn(HevcSeqParamSet* sps_data) {
     video_format_params_.display_aspect_ratio.x = disp_width / gcd;
     video_format_params_.display_aspect_ratio.y = disp_height / gcd;
 
+    video_format_params_.reconfig_options = ROCDEC_RECONFIG_NEW_SURFACES;
     if (sps_data->vui_parameters_present_flag) {
         video_format_params_.video_signal_description.video_format = sps_data->vui_parameters.video_format;
         video_format_params_.video_signal_description.video_full_range_flag = sps_data->vui_parameters.video_full_range_flag;

src/parser/vp9_parser.cpp

@@ -35,6 +35,9 @@ Vp9VideoParser::Vp9VideoParser() {
     InitDpb();
     num_frames_in_chunck_ = 1;
     frame_sizes_.assign(1, 0);
+    curr_surface_width_ = 0;
+    curr_surface_height_ = 0;
+    reconfig_option_ = ROCDEC_RECONFIG_NEW_SURFACES;
 }
 
 Vp9VideoParser::~Vp9VideoParser() {
@@ -88,6 +91,9 @@ ParserResult Vp9VideoParser::ParsePictureData(const uint8_t *p_stream, uint32_t
         }
         // Init Roc decoder for the first time or reconfigure the existing decoder
         if (new_seq_activated_) {
+            if ((ret = FlushDpb()) != PARSER_OK) {
+                return ret;
+            }
             if ((ret = NotifyNewSequence(&uncompressed_header_)) != PARSER_OK) {
                 return ret;
             }
@@ -212,8 +218,8 @@ ParserResult Vp9VideoParser::NotifyNewSequence(Vp9UncompressedHeader *p_uncomp_h
 
     video_format_params_.display_area.left = 0;
     video_format_params_.display_area.top = 0;
-    video_format_params_.display_area.right = p_uncomp_header->render_size.render_width;
-    video_format_params_.display_area.bottom = p_uncomp_header->render_size.render_height;
+    video_format_params_.display_area.right = p_uncomp_header->frame_size.frame_width;
+    video_format_params_.display_area.bottom = p_uncomp_header->frame_size.frame_height;
     video_format_params_.bitrate = 0;
 
     // Dispaly aspect ratio
@@ -223,6 +229,7 @@ ParserResult Vp9VideoParser::NotifyNewSequence(Vp9UncompressedHeader *p_uncomp_h
     video_format_params_.display_aspect_ratio.x = disp_width / gcd;
     video_format_params_.display_aspect_ratio.y = disp_height / gcd;
 
+    video_format_params_.reconfig_options = reconfig_option_;
     video_format_params_.video_signal_description = {0};
     video_format_params_.seqhdr_data_length = 0;
 
@@ -546,12 +553,27 @@ ParserResult Vp9VideoParser::ParseUncompressedHeader(uint8_t *p_stream, size_t s
 
     p_uncomp_header->header_size_in_bytes = Parser::ReadBits(p_stream, offset, 16);
 
+    // Arbitrary size change is only supported on key frames. For other frame types, particularly inter-coded frames, only size down is
+    // supported where the existing surface can be re-used.
     if (pic_width_ != p_uncomp_header->frame_size.frame_width || pic_height_ != p_uncomp_header->frame_size.frame_height) {
         pic_width_ = p_uncomp_header->frame_size.frame_width;
         pic_height_ = p_uncomp_header->frame_size.frame_height;
+        if (p_uncomp_header->frame_type == kVp9KeyFrame) {
+            curr_surface_width_ = pic_width_;
+            curr_surface_height_ = pic_height_;
+            reconfig_option_ = ROCDEC_RECONFIG_NEW_SURFACES; // Normal mode: free existing surfaces and allocate new surfaces.
+        } else {
+            if (pic_width_ <= curr_surface_width_ && pic_height_ <= curr_surface_height_) {
+                reconfig_option_ = ROCDEC_RECONFIG_KEEP_SURFACES; // Keep the existing surfaces
+            } else {
+                ERR("VP9 video size (up) change on non-key frames is not supported. Decode errors can occur.");
+                curr_surface_width_ = pic_width_;
+                curr_surface_height_ = pic_height_;
+                reconfig_option_ = ROCDEC_RECONFIG_NEW_SURFACES; // Normal mode: free existing surfaces and allocate new surfaces.
+            }
+        }
         new_seq_activated_ = true;
     }
-
     uncomp_header_size_ = (offset + 7) >> 3;
     return PARSER_OK;
 }

src/parser/vp9_parser.h

@@ -90,6 +90,12 @@ protected:
     DecodedPictureBuffer dpb_buffer_;
     Vp9Picture curr_pic_;
 
+    // Current surface size, used to support size change (down) on inter frames, where we keep the
+    // previously allocated surfaces and use them to store the smaller images.
+    uint32_t curr_surface_width_;
+    uint32_t curr_surface_height_;
+    uint32_t reconfig_option_;
+
     /*! \brief Function to parse one picture bit stream received from the demuxer.
      * \param [in] p_stream A pointer of <tt>uint8_t</tt> for the input stream to be parsed
      * \param [in] pic_data_size Size of the input stream

utils/rocvideodecode/roc_video_dec.cpp

@@ -528,15 +528,19 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
         }
     }
 
-    if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL || out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED) {
-        GetSurfaceStrideInternal(video_surface_format_, coded_width_, coded_height_, &surface_stride_, &surface_vstride_);
-    } else {
-        surface_stride_ = target_width_ * byte_per_pixel_;
+    if (p_video_format->reconfig_options == ROCDEC_RECONFIG_NEW_SURFACES) {
+        if (out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL || out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED) {
+            GetSurfaceStrideInternal(video_surface_format_, coded_width_, coded_height_, &surface_stride_, &surface_vstride_);
+        } else {
+            surface_stride_ = target_width_ * byte_per_pixel_;
+        }
     }
     chroma_height_ = static_cast<int>(ceil(target_height_ * GetChromaHeightFactor(video_surface_format_)));
     num_chroma_planes_ = GetChromaPlaneCount(video_surface_format_);
     if (p_video_format->chroma_format == rocDecVideoChromaFormat_Monochrome) num_chroma_planes_ = 0;
-    chroma_vstride_ = static_cast<int>(std::ceil(surface_vstride_ * GetChromaHeightFactor(video_surface_format_)));
+    if (p_video_format->reconfig_options == ROCDEC_RECONFIG_NEW_SURFACES) {
+        chroma_vstride_ = static_cast<int>(std::ceil(surface_vstride_ * GetChromaHeightFactor(video_surface_format_)));
+    }
     // Fill output_surface_info_
     output_surface_info_.output_width = target_width_;
     output_surface_info_.output_height = target_height_;
@@ -590,8 +594,9 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
         ROCDEC_THROW("Reconfigurition of the decoder detected but the decoder was not initialized previoulsy!", ROCDEC_NOT_SUPPORTED);
         return 0;
     }
-    ROCDEC_API_CALL(rocDecReconfigureDecoder(roc_decoder_, &reconfig_params));
-
+    if (p_video_format->reconfig_options == ROCDEC_RECONFIG_NEW_SURFACES) {
+        ROCDEC_API_CALL(rocDecReconfigureDecoder(roc_decoder_, &reconfig_params));
+    }
 
     input_video_info_str_.str("");
     input_video_info_str_.clear();
@@ -908,7 +913,6 @@ bool RocVideoDecoder::ReleaseInternalFrames() {
     return true;
 }
 
-
 void RocVideoDecoder::SaveFrameToFile(std::string output_file_name, void *surf_mem, OutputSurfaceInfo *surf_info, size_t rgb_image_size) {
     uint8_t *hst_ptr = nullptr;
     bool is_rgb = (rgb_image_size != 0);