Add session overhead functions & modify perf app (#366)

* add session overhead funcitons & modify perf app

* remove class:: for function calls

* add session overhead funcitons & modify perf app

* remove class:: for function calls

* review comments

* fix comments

* duration to double

* update perf sample for thread_id for overhead

* remove debug statements

* revoew comments

* return value chnage to double

* remove session ID functions

* check session id validity

samples/videoDecodePerf/videodecodeperf.cpp

@@ -52,9 +52,9 @@ void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, doubl
 
     auto end_time = std::chrono::high_resolution_clock::now();
     auto time_per_decode = std::chrono::duration<double, std::milli>(end_time - start_time).count();
-
+    auto session_overhead = p_dec->GetDecoderSessionOverHead(std::this_thread::get_id());
     // Calculate average decoding time
-    total_dec_time = time_per_decode;
+    total_dec_time = time_per_decode - session_overhead;
     double average_decoding_time = total_dec_time / n_frame;
     double n_fps = 1000 / average_decoding_time;
     *pn_fps = n_fps;

utils/rocvideodecode/roc_video_dec.cpp

@@ -52,6 +52,7 @@ RocVideoDecoder::RocVideoDecoder(int device_id, OutputSurfaceMemoryType out_mem_
 
 
 RocVideoDecoder::~RocVideoDecoder() {
+    auto start_time = StartTimer();
     if (curr_sei_message_ptr_) {
         delete curr_sei_message_ptr_;
         curr_sei_message_ptr_ = nullptr;
@@ -100,6 +101,8 @@ RocVideoDecoder::~RocVideoDecoder() {
         fp_out_ = nullptr;
     }
 
+    double elapsed_time = StopTimer(start_time);
+    AddDecoderSessionOverHead(std::this_thread::get_id(), elapsed_time);
 }
 
 static const char * GetVideoCodecString(rocDecVideoCodec e_codec) {
@@ -247,6 +250,7 @@ static void GetSurfaceStrideInternal(rocDecVideoSurfaceFormat surface_format, ui
 *  0: fail, 1: succeeded, > 1: override dpb size of parser (set by CUVIDPARSERPARAMS::max_num_decode_surfaces while creating parser)
 */
 int RocVideoDecoder::HandleVideoSequence(RocdecVideoFormat *p_video_format) {
+    auto start_time = StartTimer();
     input_video_info_str_.str("");
     input_video_info_str_.clear();
     input_video_info_str_ << "Input Video Information" << std::endl
@@ -415,6 +419,8 @@ int RocVideoDecoder::HandleVideoSequence(RocdecVideoFormat *p_video_format) {
     std::cout << input_video_info_str_.str();
 
     ROCDEC_API_CALL(rocDecCreateDecoder(&roc_decoder_, &videoDecodeCreateInfo));
+    double elapsed_time = StopTimer(start_time);
+    AddDecoderSessionOverHead(std::this_thread::get_id(), elapsed_time);
     return num_decode_surfaces;
 }
 
@@ -587,7 +593,6 @@ int RocVideoDecoder::ReconfigureDecoder(RocdecVideoFormat *p_video_format) {
     std::cout << input_video_info_str_.str();
 
     is_decoder_reconfigured_ = true;
-
     return 1;
 }
 
@@ -1125,3 +1130,11 @@ bool RocVideoDecoder::InitHIP(int device_id) {
     HIP_API_CALL(hipStreamCreate(&hip_stream_));
     return true;
 }
+
+std::chrono::_V2::system_clock::time_point RocVideoDecoder::StartTimer() {
+    return std::chrono::_V2::system_clock::now();
+}
+
+double RocVideoDecoder::StopTimer(const std::chrono::_V2::system_clock::time_point &start_time) {
+    return std::chrono::duration<double, std::milli>(std::chrono::_V2::system_clock::now() - start_time).count();
+}

utils/rocvideodecode/roc_video_dec.h

@@ -34,6 +34,8 @@ THE SOFTWARE.
 #include <stdexcept>
 #include <exception>
 #include <cstring>
+#include <unordered_map>
+#include <chrono>
 #include <hip/hip_runtime.h>
 extern "C" {
 #include "libavutil/md5.h"
@@ -76,7 +78,6 @@ typedef enum OutputSurfaceMemoryType_enum {
 #endif
 #define ERR(X) std::cerr << "[ERR] "  << " {" << __func__ <<"} " << " " << X << std::endl;
 
-
 class RocVideoDecodeException : public std::exception {
 public:
 
@@ -357,8 +358,17 @@ class RocVideoDecoder {
          */
         int32_t GetNumOfFlushedFrames() { return num_frames_flushed_during_reconfig_;}
 
+        // Session overhead refers to decoder initialization and deinitialization time
+        void AddDecoderSessionOverHead(std::thread::id session_id, double duration) { session_overhead_[session_id] += duration; }
+        double GetDecoderSessionOverHead(std::thread::id session_id) {
+            if (session_overhead_.find(session_id) != session_overhead_.end()) {
+                return session_overhead_[session_id];
+            } else {
+                return 0;
+            }
+         }
+
     private:
-        int decoder_session_id_; // Decoder session identifier. Used to gather session level stats.
         /**
          *   @brief  Callback function to be registered for getting a callback when decoding of sequence starts
          */
@@ -420,6 +430,18 @@ class RocVideoDecoder {
          */
         bool InitHIP(int device_id);
 
+        /**
+         * @brief Function to get start time
+         * 
+         */
+        std::chrono::_V2::system_clock::time_point StartTimer();
+
+        /**
+         * @brief Function to get elapsed time
+         * 
+         */
+        double StopTimer(const std::chrono::_V2::system_clock::time_point &start_time);
+
         int num_devices_;
         int device_id_;
         RocdecVideoParser rocdec_parser_ = nullptr;
@@ -469,4 +491,6 @@ class RocVideoDecoder {
         bool is_decoder_reconfigured_ = false;
         std::string current_output_filename = "";
         uint32_t extra_output_file_count_ = 0;
+        std::thread::id decoder_session_id_; // Decoder session identifier. Used to gather session level stats.
+        std::unordered_map<std::thread::id, double> session_overhead_; // Records session overhead of initialization+deinitialization time. Format is (thread id, duration)
 };