Perf sample: Added decoded frame number and display delay options to video decode performance sample. (#405)

* * rocDecode/Perf sample: Added decoded frame number and display delay options to video decode performance sample. Also changed default thread number from 4 to 1.

* * rocDecode: Added max number of decoded frames option to decode sample script. This is useful to do partial decoding test on long streams.

* * rocDecode: Updated README.md.

* * rocDecode: Minor correction.

samples/videoDecodePerf/README.md

@@ -30,7 +30,9 @@ make -j
 
 ```shell
 ./videodecodeperf -i <input video file [required]> 
-                  -t <number of threads [optional - default:4]>
+                  -t <number of threads [optional - default:1]>
+                  -f <Number of decoded frames - specify the number of pictures to be decoded [optional]>
+                  -disp_delay <display delay - specify the number of frames to be delayed for display [optional]>
                   -d <Device ID (>= 0) [optional - default:0]>
                   -z <force_zero_latency - Decoded frames will be flushed out for display immediately [optional]>
 ```

samples/videoDecodePerf/videodecodeperf.cpp

@@ -37,7 +37,7 @@ THE SOFTWARE.
 #include "roc_video_dec.h"
 #include "common.h"
 
-void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, double *pn_fps) {
+void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, double *pn_fps, int num_decoded_frames) {
     int n_video_bytes = 0, n_frame_returned = 0, n_frame = 0;
     uint8_t *p_video = nullptr;
     int64_t pts = 0;
@@ -48,6 +48,9 @@ void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, doubl
         demuxer->Demux(&p_video, &n_video_bytes, &pts);
         n_frame_returned = p_dec->DecodeFrame(p_video, n_video_bytes, 0, pts);
         n_frame += n_frame_returned;
+        if (num_decoded_frames && num_decoded_frames <= n_frame) {
+            break;
+        }
     } while (n_video_bytes);
 
     auto end_time = std::chrono::high_resolution_clock::now();
@@ -64,7 +67,7 @@ void DecProc(RocVideoDecoder *p_dec, VideoDemuxer *demuxer, int *pn_frame, doubl
 void ShowHelpAndExit(const char *option = NULL) {
     std::cout << "Options:" << std::endl
     << "-i Input File Path - required" << std::endl
-    << "-t Number of threads (>= 1) - optional; default: 4" << std::endl
+    << "-t Number of threads (>= 1) - optional; default: 1" << std::endl
     << "-d Device ID (>= 0)  - optional; default: 0" << std::endl
     << "-z force_zero_latency (force_zero_latency, Decoded frames will be flushed out for display immediately); optional;" << std::endl;
     exit(0);
@@ -74,10 +77,13 @@ int main(int argc, char **argv) {
 
     std::string input_file_path;
     int device_id = 0;
-    int n_thread = 4;
+    int n_thread = 1;
     Rect *p_crop_rect = nullptr;
     OutputSurfaceMemoryType mem_type = OUT_SURFACE_MEM_NOT_MAPPED;        // set to decode only for performance
     bool b_force_zero_latency = false;
+    uint32_t num_decoded_frames = 0;  // default value is 0, meaning decode the entire stream
+    int disp_delay = 0;
+
     // Parse command-line arguments
     if(argc <= 1) {
         ShowHelpAndExit();
@@ -113,6 +119,20 @@ int main(int argc, char **argv) {
             }
             continue;
         }
+        if (!strcmp(argv[i], "-disp_delay")) {
+            if (++i == argc) {
+                ShowHelpAndExit("-disp_delay");
+            }
+            disp_delay = atoi(argv[i]);
+            continue;
+        }
+        if (!strcmp(argv[i], "-f")) {
+            if (++i == argc) {
+                ShowHelpAndExit("-d");
+            }
+            num_decoded_frames = atoi(argv[i]);
+            continue;
+        }
         if (!strcmp(argv[i], "-z")) {
             if (i == argc) {
                 ShowHelpAndExit("-z");
@@ -177,7 +197,7 @@ 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));
+            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));
             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;
@@ -207,7 +227,7 @@ int main(int argc, char **argv) {
         }
 
         for (int i = 0; i < n_thread; i++) {
-            v_thread.push_back(std::thread(DecProc, v_viddec[i].get(), v_demuxer[i].get(), &v_frame[i], &v_fps[i]));
+            v_thread.push_back(std::thread(DecProc, v_viddec[i].get(), v_demuxer[i].get(), &v_frame[i], &v_fps[i], num_decoded_frames));
         }
 
         for (int i = 0; i < n_thread; i++) {

test/testScripts/README.md

@@ -43,7 +43,9 @@ optional arguments:
   --sample_mode SAMPLE_MODE
                         The sample to run - optional (default:0 [range:0-1] 0: videoDecode, 1: videoDecodePerf)
   --num_threads NUM_THREADS
-                        The number of threads is only for the videoDecodePerf sample (sample_mode = 1) - optional (default:4)
+                        The number of threads is only for the videoDecodePerf sample (sample_mode = 1) - optional (default:1)
+  --max_num_decoded_frames MAX_NUM_DECODED_FRAMES
+                        The max number of decoded frames. Useful for partial decoding of a long stream. - optional (default:0, meaning no limit)
 ```
 
 * **run_rocDecode_Conformance.py**

test/testScripts/run_rocDecodeSamples.py

@@ -64,8 +64,10 @@ parser.add_argument('--files_directory',    type=str, default='',
                     help='The path to a dirctory containing one or more supported files for decoding (e.g., mp4, mov, etc.) - required')
 parser.add_argument('--sample_mode',          type=int, default=0,
                     help='The sample to run - optional (default:0 [range:0-1] 0: videoDecode, 1: videoDecodePerf)')
-parser.add_argument('--num_threads',          type=int, default=4,
-                    help='The number of threads is only for the videoDecodePerf sample (sample_mode = 1) - optional (default:4)')
+parser.add_argument('--num_threads',          type=int, default=1,
+                    help='The number of threads is only for the videoDecodePerf sample (sample_mode = 1) - optional (default:1)')
+parser.add_argument('--max_num_decoded_frames',          type=int, default=0,
+                    help='The max number of decoded frames. Useful for partial decoding of a long stream. - optional (default:0, meaning no limit)')
 
 args = parser.parse_args()
 
@@ -75,6 +77,7 @@ filesDir = args.files_directory
 filesDirPath = Path(filesDir)
 sampleMode = args.sample_mode
 numThreads = args.num_threads
+maxNumFrames = args.max_num_decoded_frames
 
 print("\nrunrocDecodeTests V"+__version__+"\n")
 
@@ -113,7 +116,7 @@ if os.path.exists(resultsPath+'/rocDecode_test_results.csv'):
 
 if sampleMode == 0:
     for current_file in iter_files(filesDirPath):
-        os.system(run_rocDecode_app+' -i '+str(current_file)+' -d '+str(gpuDeviceID)+' | tee -a '+resultsPath+'/rocDecode_output.log')
+        os.system(run_rocDecode_app+' -i '+str(current_file)+' -d '+str(gpuDeviceID)+' -f '+str(maxNumFrames)+' | tee -a '+resultsPath+'/rocDecode_output.log')
         print("\n\n")
 
     orig_stdout = sys.stdout
@@ -144,7 +147,7 @@ if sampleMode == 0:
     os.system(runAwk_csv)
 elif sampleMode == 1:
     for current_file in iter_files(filesDirPath):
-        os.system(run_rocDecode_app+' -i '+str(current_file)+' -t '+str(numThreads)+' | tee -a '+resultsPath+'/rocDecode_output.log')
+        os.system(run_rocDecode_app+' -i '+str(current_file)+' -t '+str(numThreads)+' -f '+str(maxNumFrames)+' | tee -a '+resultsPath+'/rocDecode_output.log')
         print("\n\n")
 
     orig_stdout = sys.stdout