Updated videoDecodePerf Sample with host backend (#635)

* set min: number of decoded surfaces to non-zero value for ffmpeg based decoding * added host decoding option to videodecode perf app * fix GetOutputSurfaceInfo() for FFMpeg * updated chagelog * added host decoding option to videodecode perf app * fix GetOutputSurfaceInfo() for FFMpeg * updated chagelog * removed GPU device info printing for HOST backend * fix for review comment [ROCm/rocdecode commit: eeecc3e8c8]
2025-08-28 09:02:13 -07:00
parent 87baa53ebd
commit 74ea277fdf
8 changed files with 113 additions and 39 deletions
@@ -2,7 +2,7 @@

 Full documentation for rocDecode is available at [https://rocm.docs.amd.com/projects/rocDecode/en/latest/](https://rocm.docs.amd.com/projects/rocDecode/en/latest/)

-## rocDecode 1.2.0 (unreleased)
+## rocDecode 1.3.0 (unreleased)

 ### Added

@@ -20,6 +20,7 @@ Full documentation for rocDecode is available at [https://rocm.docs.amd.com/proj
 * rocdecode now uses the Cmake CMAKE_PREFIX_PATH directive.
 * rocdecode - A new avcodec-based decoder built as a separate "rocdecode-host" library
 * The host backend in the `videoDeode`  sample now uses the rocdecode-host library
+* The host backend using rocdecode-host library is added in `videoDecodePerf` sample

 ### Optimized

@@ -40,7 +40,7 @@ if (NOT DEFINED CMAKE_CXX_COMPILER)
 endif()

 # rocdecode Version
-set(VERSION "1.2.0")
+set(VERSION "1.3.0")

 # Set Project Version and Language
 project(rocdecode VERSION ${VERSION} LANGUAGES CXX)
@@ -65,6 +65,7 @@ find_package(HIP QUIET)
 find_package(rocdecode QUIET)
 find_package(rocprofiler-register QUIET)
 find_package(FFmpeg QUIET)
+find_package(Threads QUIET)

 if(HIP_FOUND AND FFMPEG_FOUND AND rocdecode_FOUND AND Threads_FOUND AND rocprofiler-register_FOUND)
    # HIP
@@ -74,7 +75,7 @@ if(HIP_FOUND AND FFMPEG_FOUND AND rocdecode_FOUND AND Threads_FOUND AND rocprofi
                      ${SWSCALE_INCLUDE_DIR} ${AVFORMAT_INCLUDE_DIR})
    set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} ${FFMPEG_LIBRARIES})
    # rocdecode and utils
-    include_directories (${rocdecode_INCLUDE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/../../utils ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/rocvideodecode ${CMAKE_CURRENT_SOURCE_DIR}/..)
+    include_directories (${rocdecode_INCLUDE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/../../utils ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/rocvideodecode ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/ffmpegvideodecode/ ${CMAKE_CURRENT_SOURCE_DIR}/..)
    set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} rocdecode::rocdecode)
    # threads
    set(THREADS_PREFER_PTHREAD_FLAG ON)
@@ -82,10 +83,19 @@ if(HIP_FOUND AND FFMPEG_FOUND AND rocdecode_FOUND AND Threads_FOUND AND rocprofi
    set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} Threads::Threads)
    # rocprofiler-register
    set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} rocprofiler-register::rocprofiler-register)
+    # rocdecodehost
+    find_library(ROCDECODE_HOST_LIBRARY NAMES rocdecodehost HINTS ${ROCM_PATH}/lib)
    # sample app exe
-    list(APPEND SOURCES ${PROJECT_SOURCE_DIR} videodecodeperf.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/rocvideodecode/roc_video_dec.cpp)
+    list(APPEND SOURCES ${PROJECT_SOURCE_DIR} videodecodeperf.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/rocvideodecode/roc_video_dec.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../../utils/ffmpegvideodecode/ffmpeg_video_dec.cpp)
    add_executable(${PROJECT_NAME} ${SOURCES})
+    if(ROCDECODE_HOST_LIBRARY)
+      set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} ${ROCDECODE_HOST_LIBRARY})
+      target_compile_definitions(${PROJECT_NAME} PUBLIC ENABLE_HOST_DECODE=1)
+    else()
+      target_compile_definitions(${PROJECT_NAME} PUBLIC ENABLE_HOST_DECODE=0)
+    endif()
    target_link_libraries(${PROJECT_NAME} ${LINK_LIBRARY_LIST})
+
    # FFMPEG multi-version support
    if(_FFMPEG_AVCODEC_VERSION VERSION_LESS_EQUAL 58.134.100)
      target_compile_definitions(${PROJECT_NAME} PUBLIC USE_AVCODEC_GREATER_THAN_58_134=0)
@@ -35,6 +35,7 @@ THE SOFTWARE.
 #endif
 #include "video_demuxer.h"
 #include "roc_video_dec.h"
+#include "ffmpeg_video_dec.h"
 #include "common.h"

 void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, int *pn_pic_dec, double *pn_fps, double *pn_fps_dec, int max_num_frames, OutputSurfaceMemoryType mem_type) {
@@ -98,6 +99,7 @@ int main(int argc, char **argv) {
    bool b_force_zero_latency = false;
    uint32_t max_num_frames = 0;  // max number of frames to be decoded. default value is 0, meaning decode the entire stream
    int disp_delay = 1;
+    int backend = 0;

    // Parse command-line arguments
    if(argc <= 1) {
@@ -162,6 +164,13 @@ int main(int argc, char **argv) {
            mem_type = static_cast<OutputSurfaceMemoryType>(atoi(argv[i]));
            continue;
        }
+        if (!strcmp(argv[i], "-backend")) {
+            if (++i == argc) {
+                ShowHelpAndExit("-backend");
+            }
+            backend = atoi(argv[i]);
+            continue;
+        }
        ShowHelpAndExit(argv[i]);
    }
    
@@ -219,7 +228,22 @@ int main(int argc, char **argv) {
            } else {
                v_device_id[i] = i % hip_vis_dev_count;
            }
-            std::unique_ptr<RocVideoDecoder> dec(new RocVideoDecoder(v_device_id[i], mem_type, rocdec_codec_id, b_force_zero_latency, p_crop_rect, false, disp_delay));
+            std::unique_ptr<RocVideoDecoder> dec;
+            if (!backend) { // gpu backend
+                dec = std::make_unique<RocVideoDecoder>(v_device_id[i], mem_type, rocdec_codec_id, b_force_zero_latency, p_crop_rect, false, disp_delay);
+            } else {
+            #if ENABLE_HOST_DECODE
+                std::cout << "info: RocDecode is using CPU backend!" << std::endl;
+                uint32_t max_width = demuxer->GetWidth();
+                uint32_t max_height =demuxer->GetHeight();
+                mem_type = OUT_SURFACE_MEM_HOST_COPIED;
+                dec = std::make_unique<FFMpegVideoDecoder>(v_device_id[i], mem_type, rocdec_codec_id, b_force_zero_latency, p_crop_rect, false, disp_delay, max_width, max_height);
+            #else
+                std::cout << "Error: RocDecode HOST library is not found and backend is not supported!" << std::endl;
+                return 0;
+            #endif
+            }
+
            if (!dec->CodecSupported(v_device_id[i], rocdec_codec_id, demuxer->GetBitDepth())) {
                std::cerr << "Codec not supported on GPU, skipping this file!" << std::endl;
                continue;
@@ -246,7 +270,7 @@ int main(int argc, char **argv) {

        for (int i = 0; i < n_thread; i++) {
            v_viddec[i]->GetDeviceinfo(device_name, gcn_arch_name, pci_bus_id, pci_domain_id, pci_device_id);
-            std::cout << "info: stream " << i << " using GPU device " << v_device_id[i] << " - " << device_name << "[" << gcn_arch_name << "] on PCI bus " <<
+            if (!backend) std::cout << "info: stream " << i << " using GPU device " << v_device_id[i] << " - " << device_name << "[" << gcn_arch_name << "] on PCI bus " <<
            std::setfill('0') << std::setw(2) << std::right << std::hex << pci_bus_id << ":" << std::setfill('0') << std::setw(2) <<
            std::right << std::hex << pci_domain_id << "." << pci_device_id << std::dec << std::endl;
            std::cout << "info: decoding started for thread " << i << " ,please wait!" << std::endl;
@@ -45,11 +45,16 @@ static inline float GetChromaWidthFactor(rocDecVideoSurfaceFormat surface_format

 FFMpegVideoDecoder::FFMpegVideoDecoder(int device_id, OutputSurfaceMemoryType out_mem_type, rocDecVideoCodec codec, bool force_zero_latency,
              const Rect *p_crop_rect, bool extract_user_sei_Message, uint32_t disp_delay,  int max_width, int max_height, uint32_t clk_rate) :
-              RocVideoDecoder(device_id, out_mem_type, codec, force_zero_latency, p_crop_rect, extract_user_sei_Message, disp_delay, max_width, max_height, clk_rate) {
+               RocVideoDecoder(device_id, out_mem_type, codec, force_zero_latency, p_crop_rect, extract_user_sei_Message, disp_delay, max_width, max_height, clk_rate, true) {

    if ((out_mem_type_ == OUT_SURFACE_MEM_DEV_INTERNAL) || (out_mem_type_ == OUT_SURFACE_MEM_NOT_MAPPED)) {
        ROCDEC_THROW("Unsupported output memory type", ROCDEC_INVALID_PARAMETER);
    }
+    if (out_mem_type_ == OUT_SURFACE_MEM_DEV_COPIED) {
+        if (!InitHIP(device_id_)) {
+            ROCDEC_THROW("Failed to initilize the HIP", ROCDEC_DEVICE_INVALID);
+        }
+    }
    // many of the decoder parameters are hardcoded below for just creating the decoder.
    // In the handlevideosequence callback, the decoder will get reconfigured to the actual parameters in the sequence header
    RocDecoderHostCreateInfo create_info = {};
@@ -74,7 +79,19 @@ FFMpegVideoDecoder::FFMpegVideoDecoder(int device_id, OutputSurfaceMemoryType ou
    create_info.pfn_display_picture = FFMpegHandlePictureDisplayProc;
    create_info.pfn_get_sei_msg = nullptr;        // tobe supported in future
    ROCDEC_API_CALL(rocDecCreateDecoderHost(&roc_decoder_, &create_info));
-
+    // set disp_width and height to non_zero values for it doesn't trigger decoding error before actual start of decoding
+    disp_width_ = max_width;
+    disp_height_ = max_height;
+    // fill output_surface_info_
+    output_surface_info_.output_width = max_width;
+    output_surface_info_.output_height = max_height;
+    output_surface_info_.output_pitch  = max_width * 2;     // bytes_per_pixel 2
+    output_surface_info_.output_vstride = max_height;
+    output_surface_info_.bit_depth = bitdepth_minus_8_ + 8;
+    output_surface_info_.bytes_per_pixel = 2;
+    output_surface_info_.surface_format = rocDecVideoSurfaceFormat_P016;
+    output_surface_info_.num_chroma_planes = 2;
+    output_surface_info_.mem_type = OUT_SURFACE_MEM_HOST_COPIED;
 }


@@ -201,6 +218,15 @@ int FFMpegVideoDecoder::HandleVideoSequence(RocdecVideoFormatHost *format_host)
    return num_decode_surfaces;
 }

+bool FFMpegVideoDecoder::GetOutputSurfaceInfo(OutputSurfaceInfo **surface_info) {
+    if (!disp_width_ || !disp_height_) {
+        std::cerr << "ERROR: FFMpegVideo is not intialized" << std::endl;
+        return false;
+    }
+    *surface_info = &output_surface_info_;
+    return true;
+}
+
 /**
 * @brief function to reconfigure decoder if there is a change in sequence params.
 *
@@ -72,13 +72,22 @@ class FFMpegVideoDecoder: public RocVideoDecoder {
         * @param num_decoded_pics - nummber of pictures decoded in this call
         * @return int - num of frames to display
         */
-        int DecodeFrame(const uint8_t *data, size_t size, int pkt_flags, int64_t pts = 0, int *num_decoded_pics = nullptr);
+        int DecodeFrame(const uint8_t *data, size_t size, int pkt_flags, int64_t pts = 0, int *num_decoded_pics = nullptr) override;
+        
+        /**
+         * @brief Get the pointer to the Output Image Info
+         * 
+         * @param surface_info ptr to output surface info
+         * @return true
+         * @return false
+         */
+        bool GetOutputSurfaceInfo(OutputSurfaceInfo **surface_info) override;

        /**
         * @brief This function returns a decoded frame and timestamp. This should be called in a loop fetching all the available frames
         * 
         */
-        uint8_t* GetFrame(int64_t *pts);
+        uint8_t* GetFrame(int64_t *pts) override;

        /**
         * @brief function to release frame after use by the application: Only used with "OUT_SURFACE_MEM_DEV_INTERNAL"
@@ -88,7 +97,7 @@ class FFMpegVideoDecoder: public RocVideoDecoder {
         * @return true      - success
         * @return false     - falied
         */
-        bool ReleaseFrame(int64_t pTimestamp, bool b_flushing = false);
+        bool ReleaseFrame(int64_t pTimestamp, bool b_flushing = false) override;

        /**
         * @brief Helper function to dump decoded output surface to file
@@ -98,12 +107,17 @@ class FFMpegVideoDecoder: public RocVideoDecoder {
         * @param surf_info         - surface info
         * @param rgb_image_size    - image size for rgb (optional). A non_zero value indicates the surf_mem holds an rgb interleaved image and the entire size will be dumped to file
         */
-        void SaveFrameToFile(std::string output_file_name, void *surf_mem, OutputSurfaceInfo *surf_info, size_t rgb_image_size = 0);
+        void SaveFrameToFile(std::string output_file_name, void *surf_mem, OutputSurfaceInfo *surf_info, size_t rgb_image_size = 0) override;

        /**
        *   @brief  This function is used to get the current frame size based on pixel format.
        */
-        virtual int GetFrameSize() {CHECK_ZERO("Display width", disp_width_); return ((disp_width_ * disp_height_) + ((chroma_height_ * chroma_width_) * num_chroma_planes_)) * byte_per_pixel_; }
+        virtual int GetFrameSize() override {CHECK_ZERO("Display width", disp_width_); return ((disp_width_ * disp_height_) + ((chroma_height_ * chroma_width_) * num_chroma_planes_)) * byte_per_pixel_; }
+
+        /**
+         *   @brief  This function reconfigure decoder if there is a change in sequence params.
+         */
+        int ReconfigureDecoder(RocdecVideoFormat *p_video_format) override;

    private:
        /**
@@ -138,9 +152,4 @@ class FFMpegVideoDecoder: public RocVideoDecoder {
         */
        int GetSEIMessage(RocdecSeiMessageInfo *p_sei_message_info) { return RocVideoDecoder::GetSEIMessage(p_sei_message_info);};

-        /**
-         *   @brief  This function reconfigure decoder if there is a change in sequence params.
-         */
-        int ReconfigureDecoder(RocdecVideoFormat *p_video_format);
-
 };
@@ -23,30 +23,34 @@ THE SOFTWARE.
 #include "roc_video_dec.h"

 RocVideoDecoder::RocVideoDecoder(int device_id, OutputSurfaceMemoryType out_mem_type, rocDecVideoCodec codec, bool force_zero_latency,
-              const Rect *p_crop_rect, bool extract_user_sei_Message, uint32_t disp_delay, int max_width, int max_height, uint32_t clk_rate) :
+              const Rect *p_crop_rect, bool extract_user_sei_Message, uint32_t disp_delay, int max_width, int max_height, uint32_t clk_rate, bool skip_init) :
              device_id_{device_id}, out_mem_type_(out_mem_type), codec_id_(codec), b_force_zero_latency_(force_zero_latency), 
              b_extract_sei_message_(extract_user_sei_Message), disp_delay_(disp_delay), max_width_ (max_width), max_height_(max_height) {
-    if (!InitHIP(device_id_)) {
-        ROCDEC_THROW("Failed to initilize the HIP", ROCDEC_DEVICE_INVALID);
-    }
+
    if (p_crop_rect) crop_rect_ = *p_crop_rect;
    if (b_extract_sei_message_) {
        fp_sei_ = fopen("rocdec_sei_message.txt", "wb");
        curr_sei_message_ptr_ = new RocdecSeiMessageInfo;
        memset(&sei_message_display_q_, 0, sizeof(sei_message_display_q_));
    }
-    // create rocdec videoparser
-    RocdecParserParams parser_params = {};
-    parser_params.codec_type = codec_id_;
-    parser_params.max_num_decode_surfaces = 1; // let the parser to determine the decode buffer pool size
-    parser_params.clock_rate = clk_rate;
-    parser_params.max_display_delay = disp_delay_;
-    parser_params.user_data = this;
-    parser_params.pfn_sequence_callback = HandleVideoSequenceProc;
-    parser_params.pfn_decode_picture = HandlePictureDecodeProc;
-    parser_params.pfn_display_picture = b_force_zero_latency_ ? NULL : HandlePictureDisplayProc;
-    parser_params.pfn_get_sei_msg = b_extract_sei_message_ ? HandleSEIMessagesProc : NULL;
-    ROCDEC_API_CALL(rocDecCreateVideoParser(&rocdec_parser_, &parser_params));
+    // derived class can skip the following initialization by setting skip_init flag
+    if (!skip_init) {
+        if (!InitHIP(device_id_)) {
+            ROCDEC_THROW("Failed to initilize the HIP", ROCDEC_DEVICE_INVALID);
+        }
+        // create rocdec videoparser
+        RocdecParserParams parser_params = {};
+        parser_params.codec_type = codec_id_;
+        parser_params.max_num_decode_surfaces = 1; // let the parser to determine the decode buffer pool size
+        parser_params.clock_rate = clk_rate;
+        parser_params.max_display_delay = disp_delay_;
+        parser_params.user_data = this;
+        parser_params.pfn_sequence_callback = HandleVideoSequenceProc;
+        parser_params.pfn_decode_picture = HandlePictureDecodeProc;
+        parser_params.pfn_display_picture = b_force_zero_latency_ ? NULL : HandlePictureDisplayProc;
+        parser_params.pfn_get_sei_msg = b_extract_sei_message_ ? HandleSEIMessagesProc : NULL;
+        ROCDEC_API_CALL(rocDecCreateVideoParser(&rocdec_parser_, &parser_params));
+    }
 }


@@ -76,7 +80,6 @@ RocVideoDecoder::~RocVideoDecoder() {
        delete curr_video_format_ptr_;
        curr_video_format_ptr_ = nullptr;
    }
-
    std::lock_guard<std::mutex> lock(mtx_vp_frame_);
    if (out_mem_type_ != OUT_SURFACE_MEM_DEV_INTERNAL) {
        for (auto &p_frame : vp_frames_) {
@@ -100,11 +103,11 @@ RocVideoDecoder::~RocVideoDecoder() {
            std::cerr << "ERROR: hipStream_Destroy failed! (" << hip_status << ")" << std::endl;
        }
    }
+
    if (fp_out_) {
        fclose(fp_out_);
        fp_out_ = nullptr;
    }
-
    double elapsed_time = StopTimer(start_time);
    AddDecoderSessionOverHead(std::this_thread::get_id(), elapsed_time);
 }
@@ -219,7 +219,8 @@ class RocVideoDecoder {
        */
        RocVideoDecoder(int device_id,  OutputSurfaceMemoryType out_mem_type, rocDecVideoCodec codec, bool force_zero_latency = false,
                          const Rect *p_crop_rect = nullptr, bool extract_user_SEI_Message = false, uint32_t disp_delay = 0, int max_width = 0, int max_height = 0,
-                          uint32_t clk_rate = 1000);
+                          uint32_t clk_rate = 1000, bool skip_init = false);
+
        virtual ~RocVideoDecoder();

        rocDecVideoCodec GetCodecId() { return codec_id_; }
@@ -291,7 +292,7 @@ class RocVideoDecoder {
         * @return true 
         * @return false 
         */
-        bool GetOutputSurfaceInfo(OutputSurfaceInfo **surface_info);
+        virtual bool GetOutputSurfaceInfo(OutputSurfaceInfo **surface_info);

        /**
         * @brief Function to set the Reconfig Params object
@@ -486,7 +487,7 @@ class RocVideoDecoder {
        bool b_force_recofig_flush_ = false;
        int32_t num_frames_flushed_during_reconfig_ = 0;
        hipDeviceProp_t hip_dev_prop_;
-        hipStream_t hip_stream_;
+        hipStream_t hip_stream_ = nullptr;
        rocDecVideoChromaFormat video_chroma_format_ = rocDecVideoChromaFormat_420;
        rocDecVideoSurfaceFormat video_surface_format_ = rocDecVideoSurfaceFormat_NV12;
        RocdecSeiMessageInfo *curr_sei_message_ptr_ = nullptr;