MediaCodec hardware decoder much slower with different server configuration?

Question

I've been using the Android MediaCodec in order to (hardware) decode H.264 frames on my Galaxy S4 coming from a Live 555 RTSP live (real-time) stream. After changing my Live 555 server configuration from using ffmpeg (with x264) to encode frames, to using strictly x264 to encode frames, the time to decode frames with MediaCodec takes much longer. Basically, MediaCodec can't keep up with the stream, and displays the video in slow motion, getting slower and slower as time goes on. Going back to ffmpeg isn't a solution for me, because I need the ability to encode into discrete NAL units, rather than a whole frame like ffmpeg does.

I was wondering if this was either: A) An issue with the way my server is encoding NAL units, or B) An issue with my Android client, specifically the way it is receiving and decoding NAL units.

My encoding configuration with x264 is:

x264_param_default_preset(&param,"ultrafast", "zerolatency:fastdecode");
param.i_threads = 1;
param.i_bframe = 1;
param.i_width = image_width;
param.i_height = image_height;
param.i_fps_num = 60;
param.i_fps_den = 1;
param.i_keyint_max = 10;

param.rc.i_rc_method = X264_RC_ABR;
param.rc.i.bitrate = 6000;
param.i_sps_id = 7;
param.b_repeat_headers = 1;
param.b_annexb = 0;

My Android MediaCodec client is set up as follows:

I receive each individual NAL unit on a separate Live 555 RTSP client thread. Each NAL is put into a queue along with its size and presentation time. A separate decoder thread grabs NALs from this queue, and if there are none available, waits until there are.

Some notes:

What I generally see happen is the queue starts filling up with NALs, instead of staying close to empty. So, I know that the decoder thread is not working fast enough. I don't think this is an inherent problem with decoding on an Android phone (for example, processing limitations) because it does the same thing for very low bitrates-- also again, it DID work when I was using ffmpeg to encode. If I omit certain NAL units, the decoder can start to keep up. Since I'm using Cyanogenmod 10.1, bumping up the minimum CPU frequency helps, too.

Edit:

Here is a log of the Android client, as well as a log highlighting the garbage collector specifically--

Entire Logcat:

10-15 16:40:03.955: D/DecodeActivity(18859): INFO_OUTPUT_BUFFERS_CHANGED
10-15 16:40:03.995: E/OMX-VDEC-1080P(288): Sync frame received
10-15 16:40:03.995: E/OMX-VDEC-1080P(288):  No color conversion required
10-15 16:40:03.995: E/OMX-VDEC-1080P(288): Get_parameter: OMX_IndexParamPortDefinition: nPortIndex (1), nFrameWidth (1280), nFrameHeight (720), nStride (1280), nSliceHeight (736), nBitrate (-1073741824), xFramerate (0x1e), nBufferSize (1433600), nBufferCountMin (4), nBufferCountActual (8), bBuffersContiguous (1918394328), nBufferAlignment (1075643347), bEnabled (1), bPopulated (1), eCompressionFormat (0x0), eColorFormat (0x7fa30c03)
10-15 16:40:03.995: D/DecodeActivity(18859): New format {height=720, what=1869968451, color-format=2141391875, slice-height=736, crop-left=0, width=1280, crop-bottom=719, crop-top=0, mime=video/raw, stride=1280, crop-right=1279}
10-15 16:40:04.005: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4966) fps(201.369308)
10-15 16:40:04.015: W/IInputConnectionWrapper(1069): showStatusIcon on inactive InputConnection
10-15 16:40:04.025: I/ActivityManager(698): Displayed com.mathieu.alloclient.javadecoder/.MainActivity: +617ms
10-15 16:40:04.025: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4766) fps(209.819550)
10-15 16:40:04.125: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4733) fps(211.282486)
10-15 16:40:04.445: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4711) fps(212.269150)
10-15 16:40:04.495: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4700) fps(212.765961)
10-15 16:40:05.676: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4688) fps(213.310577)
10-15 16:40:06.087: D/dalvikvm(698): WAIT_FOR_CONCURRENT_GC blocked 1ms
10-15 16:40:06.207: D/dalvikvm(698): GC_EXPLICIT freed 4120K, 39% free 24216K/39664K, paused 7ms+9ms, total 117ms
10-15 16:40:06.537: D/ALSADevice(288): standby: handle 0x40024450 h 0x0
10-15 16:40:06.577: D/alsa_ucm(288): snd_use_case_set(): uc_mgr 0x400fbfb0 identifier _verb value Inactive
10-15 16:40:06.577: D/alsa_ucm(288): Set mixer controls for HiFi Lowlatency enable 0
10-15 16:40:06.577: D/alsa_ucm(288): Setting mixer control: SLIMBUS_0_RX Audio Mixer MultiMedia5, value: 0
10-15 16:40:06.577: D/alsa_ucm(288): snd_use_case_set(): uc_mgr 0x400fbfb0 identifier _disdev value Line
10-15 16:40:06.577: D/alsa_ucm(288): disdev: device Line not enabled, no need to disable
10-15 16:40:06.577: D/alsa_ucm(288): snd_use_case_set(): uc_mgr 0x400fbfb0 identifier _disdev value Speaker
10-15 16:40:06.577: D/alsa_ucm(288): Set mixer controls for Speaker enable 0
10-15 16:40:06.577: D/alsa_ucm(288): Setting mixer control: RX5 MIX1 INP1, value: ZERO
10-15 16:40:06.587: D/alsa_ucm(288): Setting mixer control: RX5 MIX1 INP2, value: ZERO
10-15 16:40:06.587: D/alsa_ucm(288): Setting mixer control: LINEOUT2 Volume, value: 0
10-15 16:40:06.587: D/alsa_ucm(288): Setting mixer control: LINEOUT4 Volume, value: 0
10-15 16:40:06.587: D/alsa_ucm(288): Setting mixer control: RX5 Digital Volume, value: 0
10-15 16:40:06.587: D/AudioUsbALSA(288): exitPlaybackThread, mproxypfdPlayback: -1
10-15 16:40:06.587: D/AudioUsbALSA(288): closeDevice handle 0x0
10-15 16:40:06.587: D/AudioUsbALSA(288): closeDevice handle 0x0
10-15 16:40:17.638: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4677) fps(213.812271)
10-15 16:40:17.698: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4644) fps(215.331604)
10-15 16:40:20.681: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4633) fps(215.842865)
10-15 16:40:21.111: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4611) fps(216.872696)
10-15 16:40:25.746: D/dalvikvm(698): GC_CONCURRENT freed 5829K, 41% free 23778K/39664K, paused 10ms+24ms, total 165ms
10-15 16:40:28.448: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:2.800000 hotplug_avg_load_dw: 74
10-15 16:40:28.448: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:243.000000
10-15 16:40:30.841: W/SystemClock(698): time going backwards: prev 16555345563411(ioctl) vs now 16555345441341(ioctl), tid=764
10-15 16:40:44.684: D/dalvikvm(698): GC_CONCURRENT freed 5302K, 41% free 23774K/39664K, paused 6ms+9ms, total 107ms
10-15 16:40:57.467: E/OMX-VDEC-1080P(288): set_frame_rate: frm_int(4577) fps(218.483719)
10-15 16:41:14.383: D/dalvikvm(698): GC_CONCURRENT freed 5371K, 41% free 23768K/39664K, paused 7ms+8ms, total 146ms
10-15 16:41:14.403: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:4.500000 hotplug_avg_load_dw: 89
10-15 16:41:14.403: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:41:14.403: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:41:14.403: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:922.000000
10-15 16:41:27.466: E/kickstart(862): Total bytes received so far: 48
10-15 16:41:27.466: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_HELLO
10-15 16:41:27.466: E/kickstart(862): EVENT: SENDING --> SAHARA_HELLO_RESPONSE
10-15 16:41:27.466: E/kickstart(862): EVENT: sahara_mode                         = 2
10-15 16:41:27.466: E/kickstart(862): EVENT: m_comm->sahara_hello_packet_rx.mode = 2
10-15 16:41:27.466: E/kickstart(862): EVENT: helloRx.mode                        = 2
10-15 16:41:27.466: E/kickstart(862): Total bytes received so far: 64
10-15 16:41:27.466: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_MEMORY_DEBUG
10-15 16:41:27.466: E/kickstart(862): Total bytes received so far: 116
10-15 16:41:27.466: E/kickstart(862): EVENT: 0x46980000, len=000C0000, "m9kefs1", ""
10-15 16:41:27.466: E/kickstart(862): EVENT: STATE <-- SAHARA_WAIT_MEMORY_REGION
10-15 16:41:27.466: E/kickstart(862): EVENT: Saving "/dev/block/platform/msm_sdcc.1/by-name/m9kefs1"
10-15 16:41:27.526: E/kickstart(862): Total bytes received so far: 786548
10-15 16:41:27.526: E/kickstart(862): EVENT: Received: 786432 bytes
10-15 16:41:27.526: E/kickstart(862): EVENT: Writing to disk
10-15 16:41:27.526: E/kickstart(862): EVENT: Successfully wrote to disk
10-15 16:41:27.526: E/kickstart(862): Received file "m9kefs1"
10-15 16:41:27.576: E/kickstart(862): Sync finish Received file "m9kefs1"
10-15 16:41:27.576: E/kickstart(862): 786432 bytes transferred in 0.106s (7.10 MBps)
10-15 16:41:27.576: E/kickstart(862): EVENT: num_debug_entries not >=0
10-15 16:41:27.576: E/kickstart(862): Successfully downloaded files from target 
10-15 16:41:27.576: E/kickstart(862): EVENT: SENDING --> SAHARA_RESET
10-15 16:41:27.576: E/kickstart(862): Total bytes received so far: 786556
10-15 16:41:27.576: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_RESET_RESP
10-15 16:41:27.576: E/kickstart(862): Sahara protocol completed
10-15 16:41:27.576: E/kickstart(862): EVENT: STATE <-- SAHARA_WAIT_HELLO
10-15 16:41:27.746: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:2.100000 hotplug_avg_load_dw: 79
10-15 16:41:27.746: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:244.000000
10-15 16:41:49.598: D/dalvikvm(698): GC_CONCURRENT freed 5339K, 41% free 23775K/39664K, paused 10ms+7ms, total 134ms
10-15 16:42:17.225: I/ActivityManager(698): Start proc com.cyanogenmod.lockclock for service com.cyanogenmod.lockclock/.weather.WeatherUpdateService: pid=18954 uid=10028 gids={50028, 3003, 1028}
10-15 16:42:17.865: D/WeatherXmlParser(18954): Weather updated: WeatherInfo for Santa Barbara@ Tue Oct 15 16:42:17 PDT 2013: Fair(34), temperature 29°C, low 11°, high 27°, humidity 14%, wind 11km/h at W
10-15 16:42:17.945: I/ActivityManager(698): No longer want com.google.android.apps.uploader (pid 13565): empty #17
10-15 16:42:24.021: D/dalvikvm(698): GC_CONCURRENT freed 5277K, 41% free 23770K/39664K, paused 8ms+11ms, total 106ms
10-15 16:42:35.983: W/ThrottleService(698): unable to find stats for iface rmnet0
10-15 16:42:59.476: D/dalvikvm(698): GC_CONCURRENT freed 5345K, 41% free 23770K/39664K, paused 6ms+13ms, total 168ms
10-15 16:43:02.098: E/kickstart(862): Total bytes received so far: 48
10-15 16:43:02.098: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_HELLO
10-15 16:43:02.098: E/kickstart(862): EVENT: SENDING --> SAHARA_HELLO_RESPONSE
10-15 16:43:02.098: E/kickstart(862): EVENT: sahara_mode                         = 2
10-15 16:43:02.098: E/kickstart(862): EVENT: m_comm->sahara_hello_packet_rx.mode = 2
10-15 16:43:02.098: E/kickstart(862): EVENT: helloRx.mode                        = 2
10-15 16:43:02.098: E/kickstart(862): Total bytes received so far: 64
10-15 16:43:02.098: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_MEMORY_DEBUG
10-15 16:43:02.108: E/kickstart(862): Total bytes received so far: 116
10-15 16:43:02.108: E/kickstart(862): EVENT: 0x46980000, len=000C0000, "m9kefs2", ""
10-15 16:43:02.108: E/kickstart(862): EVENT: STATE <-- SAHARA_WAIT_MEMORY_REGION
10-15 16:43:02.108: E/kickstart(862): EVENT: Saving "/dev/block/platform/msm_sdcc.1/by-name/m9kefs2"
10-15 16:43:02.158: E/kickstart(862): Total bytes received so far: 786548
10-15 16:43:02.158: E/kickstart(862): EVENT: Received: 786432 bytes
10-15 16:43:02.158: E/kickstart(862): EVENT: Writing to disk
10-15 16:43:02.168: E/kickstart(862): EVENT: Successfully wrote to disk
10-15 16:43:02.168: E/kickstart(862): Received file "m9kefs2"
10-15 16:43:02.218: E/kickstart(862): Sync finish Received file "m9kefs2"
10-15 16:43:02.218: E/kickstart(862): 786432 bytes transferred in 0.113s (6.65 MBps)
10-15 16:43:02.218: E/kickstart(862): EVENT: num_debug_entries not >=0
10-15 16:43:02.218: E/kickstart(862): Successfully downloaded files from target 
10-15 16:43:02.218: E/kickstart(862): EVENT: SENDING --> SAHARA_RESET
10-15 16:43:02.218: E/kickstart(862): Total bytes received so far: 786556
10-15 16:43:02.218: E/kickstart(862): EVENT: RECEIVED <-- SAHARA_RESET_RESP
10-15 16:43:02.218: E/kickstart(862): Sahara protocol completed
10-15 16:43:02.218: E/kickstart(862): EVENT: STATE <-- SAHARA_WAIT_HELLO
10-15 16:43:17.563: W/SystemClock(698): time going backwards: prev 16722067121029(ioctl) vs now 16722066968441(ioctl), tid=764
10-15 16:43:34.610: D/dalvikvm(698): GC_CONCURRENT freed 5359K, 41% free 23772K/39664K, paused 9ms+9ms, total 113ms
10-15 16:43:36.452: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:3.900000 hotplug_avg_load_dw: 100
10-15 16:43:36.452: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:43:36.452: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:43:36.452: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:194.000000
10-15 16:44:09.965: D/dalvikvm(698): GC_CONCURRENT freed 5369K, 41% free 23779K/39664K, paused 8ms+8ms, total 103ms
10-15 16:44:09.965: D/dalvikvm(698): WAIT_FOR_CONCURRENT_GC blocked 30ms
10-15 16:44:10.605: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:2.800000 hotplug_avg_load_dw: 133
10-15 16:44:10.605: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:244.000000
10-15 16:44:12.707: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:4.100000 hotplug_avg_load_dw: 116
10-15 16:44:12.707: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:44:12.707: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:44:12.707: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:97.000000
10-15 16:44:14.008: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:1.700000 hotplug_avg_load_dw: 140
10-15 16:44:14.008: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:240.000000
10-15 16:44:14.759: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:4.900000 hotplug_avg_load_dw: 132
10-15 16:44:14.759: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:44:14.759: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:44:14.759: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:97.000000
10-15 16:44:15.360: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:2.300000 hotplug_avg_load_dw: 139
10-15 16:44:15.360: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:243.000000
10-15 16:44:15.560: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:3.900000 hotplug_avg_load_dw: 133
10-15 16:44:15.560: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:44:15.560: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:44:15.560: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:95.000000
10-15 16:44:16.361: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:2.500000 hotplug_avg_load_dw: 132
10-15 16:44:16.361: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:242.000000
10-15 16:44:16.661: E/MP-Decision(1385): num online cores: 3 reqd : 4 available : 4 rq_depth:3.700000 hotplug_avg_load_dw: 127
10-15 16:44:16.661: E/MP-Decision(1385): UP cpu:1 core_idx:1 Nw:1.900000 Tw:140 total_time_up:0.000000
10-15 16:44:16.661: E/MP-Decision(1385): UP cpu:2 core_idx:2 Nw:2.700000 Tw:90 total_time_up:0.000000
10-15 16:44:16.661: E/MP-Decision(1385): UP cpu:3 core_idx:3 Nw:3.500000 Tw:90 total_time_up:97.000000
10-15 16:44:20.605: E/MP-Decision(1385): num online cores: 4 reqd : 3 available : 4 rq_depth:1.700000 hotplug_avg_load_dw: 122
10-15 16:44:20.605: E/MP-Decision(1385): DOWN cpu:3 core_idx:3 Ns:3.100000 Ts:240 total_time_down:244.000000
10-15 16:44:20.615: W/ProcessStats(698): Skipping unknown process pid 19038
10-15 16:44:20.615: W/ProcessStats(698): Skipping unknown process pid 19041

Garbage collector entries in Logcat:

10-15 16:40:06.087: D/dalvikvm(698): WAIT_FOR_CONCURRENT_GC blocked 1ms
10-15 16:40:06.207: D/dalvikvm(698): GC_EXPLICIT freed 4120K, 39% free 24216K/39664K,   paused 7ms+9ms, total 117ms
10-15 16:40:25.746: D/dalvikvm(698): GC_CONCURRENT freed 5829K, 41% free 23778K/39664K, paused 10ms+24ms, total 165ms
10-15 16:40:44.684: D/dalvikvm(698): GC_CONCURRENT freed 5302K, 41% free 23774K/39664K, paused 6ms+9ms, total 107ms
10-15 16:41:14.383: D/dalvikvm(698): GC_CONCURRENT freed 5371K, 41% free 23768K/39664K, paused 7ms+8ms, total 146ms
10-15 16:41:49.598: D/dalvikvm(698): GC_CONCURRENT freed 5339K, 41% free 23775K/39664K, paused 10ms+7ms, total 134ms
10-15 16:42:24.021: D/dalvikvm(698): GC_CONCURRENT freed 5277K, 41% free 23770K/39664K, paused 8ms+11ms, total 106ms
10-15 16:42:59.476: D/dalvikvm(698): GC_CONCURRENT freed 5345K, 41% free 23770K/39664K, paused 6ms+13ms, total 168ms
10-15 16:43:34.610: D/dalvikvm(698): GC_CONCURRENT freed 5359K, 41% free 23772K/39664K, paused 9ms+9ms, total 113ms
10-15 16:44:09.965: D/dalvikvm(698): GC_CONCURRENT freed 5369K, 41% free 23779K/39664K, paused 8ms+8ms, total 103ms
10-15 16:44:09.965: D/dalvikvm(698): WAIT_FOR_CONCURRENT_GC blocked 30ms
10-15 16:44:24.389: D/dalvikvm(698): GC_EXPLICIT freed 3618K, 41% free 23768K/39664K, paused 12ms+11ms, total 123ms

score 0 · Answer 1 · answered Oct 15 '13 at 22:21

0

This easily could be a problem with your Android client code (it's all Java, isn't it?) especially if your logcat shows that garbage collector is activated too often.

answered Oct 15 '13 at 22:21

Alex Cohn

56,089
9
113
307

You could definitely be right-- I'm still learning how to optimize Android. It's not all Java, however. I'm working with JNI, so the actual RTSP client is C++, which creates a Java object (NAL class) every time a new NAL unit is to be delivered to the queue. I posted my entire logcat as well as just the dalvikvm above. Could you take a look and see if there are any red flags? – mathieujofis Oct 16 '13 at 00:02
From your log we learn that every ~30 sec there is a ~100 ms pause to collect the objects. This *could* explain the slowdown. At any rate, creating new Java object for every NAL unit that arrives from the socket looks as very inefficient approach. But then, how are they assembled for MediaCodec? You can use [the profiler](http://developer.android.com/tools/debugging/debugging-tracing.html) to learn more about the bottlenecks in your app. – Alex Cohn Oct 16 '13 at 07:18

score 0 · Answer 2 · answered Oct 19 '13 at 11:02

0

I guess that changes of the server configuration in its turn changes rtp stream which produces higher data volume. And it in its turn reveals issues on the client: for example you can keep static pools for data instead of allocating new one each time.

answered Oct 19 '13 at 11:02

user2399321

605
7
13

MediaCodec hardware decoder much slower with different server configuration?

2 Answers2