[DM8168] Video Display Latency

Michael Questo

Other Parts Discussed in Thread: TVP5158

DM Champ,

I'm working with a customer who's using 2x DM816x for encode, transport and display. This application will be used for robotic control so glass to glass latency is critical. The following latency measurements have been very helpful in determining the expected latency that will be seen in the end system. All the capture, encode, decode latencis are appealing except for the final Scale/Chroma Con. Display Delay T7–T5 This time is roughly double the capture and encode latency. Can you kindly explain the reason (break down) for this large latency. Is there a way to bypass/accelerate this time?

DM8168 Latency Measurements:

http://processors.wiki.ti.com/index.php/Latency_Measurement_on_Capture_Encode_Decode_Display_Demo

over 11 years ago

0 Michael Questo over 11 years ago

TI__Expert 6000 points

Here's some additional information.

Experimental setup We created a small H264 elementary stream with 3 light frames and 50 dark frames. The first light frame is an I frame, and the first dark frame is an I frame. We then take this file and keep looping it, feeding it into the decoder.

We send a character out the console for each I frame (or approximately 2 characters per second). This way we can see on the scope that a frame has been submitted to the decoder.

Attached to our display is a photo diode so we can see when the display transitions from dark to light and vice versa. These can be viewed on a scope with the character from the console.

We see pretty consistent numbers doing this. There maybe a few millisecond of jitter, but not much.

Would I be correct to assume that this is without Subframes? Yes there are no subframes in the stream. We submit a whole frame at the time to the decoder.

Does this include the ethernet transport delay? No there is no Ethernet delay in these numbers. We submit frames after the Ethernet stack and we do it as fast as the decoder will let us.

Do you know what the subsystem individual latencies are and how have you measured the latency? No yet using this method, but we are expecting the decoder to take about 15ms, and the scaler another 17ms based on work we have already done. This means the rest of the time is being consumed somewhere else. We are assuming this is in the HDVPSS but there may be other things in the stock decoder demo we are using.

What operating parameters have you changed? We change the format to expect 60FPS

Have you tried this with 1080p30? Overall latency was worse at 30 FPS.

We ran another test where we starved the input to the decoder. We ran the display at 60FPS, but I shoved in images to the decoder at only 30FPS. When I did this, the latency from the decoder to the display changed from 212ms to a range of 80-105ms with quite a bit of jitter.

If we take the Decoder at 15ms, and the scaler 4:2:0 -> 4:2:2 at 17ms then we get the a range of 48 - 73ms for the display portion. This would imply the hardware imposes a 3 frame latency on all images being submitted for display. Does this make sense?

Basically we are seeing numbers that look worse than the SDK even after subtracting off the full network delay.

0 Shiju Sivasankaran over 11 years ago in reply to Michael Questo

TI__Genius 13395 points

Michael

First of all, thanks for gathering and sharing all these data

Are you using OMX Non tunnel way of data communication? What is the A8 load? If you use NT and A8 load s high it can affect the latency number very badly

Normally, each processing block takes ~16ms incase of 1080p60. There is an exception with display where it need priming of 3 frames before starting the display, so the worst case it can introduce 3-4 frame delay, which is ~48-64ms

For example, let us look into the encode path delay. Capture + DEi + encode = 16+16+16 = 48ms (excluding the delay introduced by NT path)

In the second display path you can avoid the scalar if you use the latest HDVPSS driver. For this you need to use the 420 path display driver, i guess VFDC do not support this driver. But if 420SP path display driver is integrated with VFDC, then you can avoid SC and that reduces ~16ms

A few other points to reduce the NT path delay

make SR0 non-cached - To avoid a lot of cache invalidation APIs
make bits stream (i guess SR1) region cached
Reduce the messageQ size from 1K (i think this is set as default) to 256 bytes. This will reduce the message copy time

Are they ready to evaluate these usecase with RDK? RDK has better data communication path compared with EZSDK SNT. Also has support to use 420 path display driver and a few techniques to reduce the overall display latency

regards, shiju

0 Oliver King-Smith2 over 11 years ago in reply to Shiju Sivasankaran

Prodigy 110 points

shiju,

Thank you for posting the info. The latency numbers we are seeing are on the RDK. Let me clarify our setup.

We are using the 3.5 RDK on the 8168 Spectrum digital EVM.

We are using an HDMI interface to display data.

We have our own H264 clip we inject into the decoder.

The A8 is lightly loaded.

We believe the decoder and scaler are taking about 30ms to run per image.

The total delay from the decoder to the display is about 90ms.

This means after the decoder and scaler the total delay before we see the image on the screen is 60ms. It is this delay we are trying to understand. We don't know where this time going?

Can you tell us why there is this delay?

What can we do to improve this delay?

With regards to your response above, what does NT refer to?

Oliver

0 Oliver King-Smith2 over 11 years ago in reply to Oliver King-Smith2

Prodigy 110 points

Shiju,

I forgot to mention we are using the standard Decode example program provided with the RDK. We have modified this to inject our video clip into the decoder. We do this from RAM to avoid latency issues with reading from disk.

Oliver

0 Oliver King-Smith2 over 11 years ago in reply to Oliver King-Smith2

Prodigy 110 points

Shiju,

Looking at the demo code, I see an area that might be causing the delay. In demo_vdec_vdis_frames_send.c there appears to be a function VdecVdis_ipcFramesStart where we call

    VdecVdis_ipcFramesInitFrameObj(gVdecVdis_config.ipcFrames.frameObj,
                                   MCFW_IPCFRAMES_MAX_NUM_ALLOC_FRAMES,
                                   &chInfoList);
The MCFW_IPCFRAMES_MAX_NUM_ALLOC_FRAMES is currently set to 4. Is there a reason why this number would be set to 4 instead of 1?

Oliver

0 Badri Narayanan over 11 years ago in reply to Oliver King-Smith2

TI__Guru 59700 points

Can you share the h264 elementary stream you are using.. We would like to try it on a UDwroks DVR setup and measure the latency. We have hooks in place to measure latency at each component. The demo needs minor modification to set the walltime. We can measure upto the point driver gives frame to HDVPSS h/w display. We don't have the means to check when it exactly gets displayed on TV

0 Oliver King-Smith2 over 11 years ago in reply to Badri Narayanan

Prodigy 110 points

Badri,

Thank you for your offer of assistance. Please find the attached clip which is approximately 60 frames. The first 3 frames are very light in color, and the next 53 frames are very dark.

Oliver

dark_N_frames.h264.zip

0 Badri Narayanan over 11 years ago in reply to Oliver King-Smith2

TI__Guru 59700 points

Thank you for sharing the stream.How many channels of decode do you want me to try out for latency measurement purpose. Is it single channel ?

0 Shiju Sivasankaran over 11 years ago in reply to Oliver King-Smith2

TI__Genius 13395 points

Oliver

I have mistaken you use OMX SDK framework from TI and this is because of the keyword "OMX" in the post header. NT means Non-tunnel communication, used in OMX context.

Good that you use RDK, which is far better in terms of feature-set and latency

The RDK chain for this particular usecase is

ipcBitsOut(HLOS) (16ms) -> ipcBitsIn (Vid-M3) (16ms) -> DEC link (33ms) -> MPSCLR Link (no delay if resolution is <= HD) -> SWMS Link (33ms) -> Display Link (3*16 ms)

= 16+16+33+0+33+48 = ~150ms

ipcBits links are configured as noNotifyMode mode which could add 16 + 16 ms (worst case) delay in the pipeline. We can avoid this 32ms by enabling the notify mode

As Badri mentioned we will measure the latency with your stream and get back to you

Regards, shiju

0 Michael Questo over 11 years ago in reply to Shiju Sivasankaran

TI__Expert 6000 points

Shiju/Badri,

Thanks for your help so far and sorry for the misnomer in the title. We are only decoding a single stream of video.

How would the latency scale if multiple streams were decoded and potentially mosaiced?

So just to verify my understanding, if we were to set the ipcBits to noNotifyMode then we could essentially eliminate the ipcBitsOut(HLOS) (16ms) -> ipcBitsIn (Vid-M3) (16ms) latencies.

Is there anyway to reduce the Display Link latency? And can you breifly explain the 3x16ms? Does that have to do with the frame pipeline being three frames deep?

0 Shiju Sivasankaran over 11 years ago in reply to Michael Questo

TI__Genius 13395 points

>> How would the latency scale if multiple streams were decoded and potentially mosaiced?

Normally each IP in DM8168 can do 2ch 1080p30 decode or scale in real time. So if you are not overloading the IP, then the max latency will be 33ms, it can be smaller if you do only say one channel SD

>> Is there anyway to reduce the Display Link latency? And can you breifly explain the 3x16ms? Does that have to do with the frame pipeline being three frames deep?

This is not possible as it is a HW + driver limitation. The display queue has a minimum depth of 3 and it required to start the display only after prime 3 buffers/frames. So even a frame is available in the input side of the display it will be dsiplayed only after 3 display interrupts

regards, shiju

0 Badri Narayanan over 11 years ago in reply to Shiju Sivasankaran

TI__Guru 59700 points

Below is the statistics with the clip you shared. This is for one channel 1080P60. The clip has only 56 frames and I don't see any light frame followed by dark frame. Anyhow the content doesn't matter for the latency calculation below

[m3vpss ]
[m3vpss ] *** [MP_SCLR0 ] Statistics ***
[m3vpss ]
[m3vpss ] Total Frames Received : 2720
[m3vpss ] Total Frames Forwarded : 2720
[m3vpss ]
[m3vpss ]
[m3vpss ] CH | In Recv In Reject Processed Latency(DRV) Processed Rejected
[m3vpss ] Num | FPS     FPS       FPS        Min / Max    Frames     Frames
[m3vpss ] -------------------------------------------------------------------
[m3vpss ]
[m3vpss ] *** [SWMS0] Mosaic Statistics ***
[m3vpss ]
[m3vpss ] Elasped Time: 34 secs
[m3vpss ]
[m3vpss ] Output Request FPS   : 60 fps (2066 frames)
[m3vpss ] Output Actual FPS   : 60 fps (2066 frames)
[m3vpss ] Output Drop    FPS   : 0 fps (0 frames)
[m3vpss ] Output Reject FPS   : 0 fps (0 frames)
[m3vpss ] Scaling Internal     : 16 ms
[m3vpss ] Scaling Internal min : 16 ms
[m3vpss ] Scaling Internal max : 17 ms
[m3vpss ]
[m3vpss ] Win | Window Repeat Drop Recv Que FID Invlid Acc Event          Invalid   Que Reject Reject Latency   OutBufCopy InBufCopy
[m3vpss ] Num | FPS    FPS    FPS FPS FPS FPS        Count (Max/Min)    CH Frames Frames     Frames Min / Max FPS        FPS
[m3vpss ] ---------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |     60      0    0    0   60          0        0 ( 0/255)         0          0      0 35 / 40          0         0
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS0] Mosaic Parameters ***
[m3vpss ]
[m3vpss ] Output FPS: 60
[m3vpss ]
[m3vpss ] Win | Ch | Input      | Input          | Input         | Input       | Output     | Output         | Output        | Output      | Low Cost | SWMS | Data | Blank |
[m3vpss ] Num | Num | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | ON / OFF | Inst | Format| Frame |
[m3vpss ] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |   0 |    0,    0 | 1920 x   1080 | 16384 / 32768 | TILED       |    0,    0 | 1920 x   1080 | 3840 /      0 | NON-TILED   |      OFF |    0 | 420SP |   OFF |
[m3vpss ]
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS1] Mosaic Statistics ***
[m3vpss ]
[m3vpss ] Elasped Time: 34 secs
[m3vpss ]
[m3vpss ] Output Request FPS   : 60 fps (2066 frames)
[m3vpss ] Output Actual FPS   : 60 fps (2066 frames)
[m3vpss ] Output Drop    FPS   : 0 fps (0 frames)
[m3vpss ] Output Reject FPS   : 0 fps (0 frames)
[m3vpss ] Scaling Internal     : 16 ms
[m3vpss ] Scaling Internal min : 16 ms
[m3vpss ] Scaling Internal max : 17 ms
[m3vpss ]
[m3vpss ] Win | Window Repeat Drop Recv Que FID Invlid Acc Event          Invalid   Que Reject Reject Latency   OutBufCopy InBufCopy
[m3vpss ] Num | FPS    FPS    FPS FPS FPS FPS        Count (Max/Min)    CH Frames Frames     Frames Min / Max FPS        FPS
[m3vpss ] ---------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |     60      0    0    0   60          0        0 ( 0/255)         0          0      0 35 / 41          0         0
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS1] Mosaic Parameters ***
[m3vpss ]
[m3vpss ] Output FPS: 60
[m3vpss ]
[m3vpss ] Win | Ch | Input      | Input          | Input         | Input       | Output     | Output         | Output        | Output      | Low Cost | SWMS | Data | Blank |
[m3vpss ] Num | Num | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | ON / OFF | Inst | Format| Frame |
[m3vpss ] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |   0 |    0,    0 | 1920 x   1080 | 16384 / 32768 | TILED       |    0,    0 | 1920 x   1080 | 3840 /      0 | NON-TILED   |      OFF |    0 | 420SP |   OFF |
[m3vpss ]
[m3vpss ]
[m3vpss ] 3725213: DISPLAY: HDDAC(BP0) : 59 fps, Latency (Min / Max) = ( 50 / 51 ), Callback Interval (Min / Max) = ( 16 / 17 ) DropCount:0 DispLatency (Min / Max) = ( 33 / 42 ) !!!
[m3vpss ] 3725213: DISPLAY DRV: HDDAC(BP0) : Q:[2801] Display:[2811], Repeat:[11], DQ:[2799]
[m3vpss ] 3725213: DISPLAY: UNDERFLOW COUNT: HDMI(BP0) 2066, HDDAC(BP0) 2066, DVO2(BP1) 2066, SDDAC(SEC1) 4133
[m3vpss ] 3725213: SYSTEM : FREE SPACE : System Heap      = 67320 B, Mbx = 10239 msgs)
[m3vpss ] 3725213: SYSTEM : FREE SPACE : SR0 Heap         = 16831744 B (16 MB)
[m3vpss ] 3725213: SYSTEM : FREE SPACE : Frame Buffer     = 599780224 B (571 MB)
[m3vpss ] 3725213: SYSTEM : FREE SPACE : Bitstream Buffer = 289430400 B (276 MB)
[m3vpss ] TILER_STATS: CNT :8BIT
[m3vpss ] TILER_STATS: CNT RESOLUTION:    16384 x 8192
[m3vpss ] TILER_STATS: BUCKET RESOLUTION: 2048 x 1662
[m3vpss ] TILER_STATS: NUM FREE BUCKETS: 37
[m3vpss ] TILER_STATS: NUM USED BUCKETS: 3
[m3vpss ] TILER_STATS: TOTAL FREE AREA:   126959616 (94 %)
[m3vpss ] TILER_STATS: TOTAL USED AREA:   7225344 (5 %)
[m3vpss ] TILER_STATS: TOTAL WASTE AREA: 0 (0 %)
[m3vpss ] TILER_STATS: MAX WIDTH RECT:    2048 x 456
[m3vpss ] TILER_STATS: MAX HEIGHT RECT:    2048 x 456
[m3vpss ] TILER_STATS: RES ALLOC COUNT [1080P] : 3
[m3vpss ] TILER_STATS: CNT :16BIT
[m3vpss ] TILER_STATS: CNT RESOLUTION:    32768 x 4096
[m3vpss ] TILER_STATS: BUCKET RESOLUTION: 2048 x 832
[m3vpss ] TILER_STATS: NUM FREE BUCKETS: 77
[m3vpss ] TILER_STATS: NUM USED BUCKETS: 3
[m3vpss ] TILER_STATS: TOTAL FREE AREA:   130605056 (97 %)
[m3vpss ] TILER_STATS: TOTAL USED AREA:   3612672 (2 %)
[m3vpss ] TILER_STATS: TOTAL WASTE AREA: 0 (0 %)
[m3vpss ] TILER_STATS: MAX WIDTH RECT:    2048 x 228
[m3vpss ] TILER_STATS: MAX HEIGHT RECT:    2048 x 228
[m3vpss ] TILER_STATS: RES ALLOC COUNT [1080P] : 3
[m3vpss ] 3725216: SYSTEM : FREE SPACE : Tiler 8-bit      = 126959616 B (121 MB) - TILER ON
[m3vpss ] 3725216: SYSTEM : FREE SPACE : Tiler 16-bit     = 130605056 B (124 MB) - TILER ON
[m3vpss ] 3725216: DISPLAY: DVO2(BP1) : 59 fps, Latency (Min / Max) = ( 50 / 50 ), Callback Interval (Min / Max) = ( 16 / 17 ) DropCount:0 DispLatency (Min / Max) = ( 36 / 42 ) !!!
[m3vpss ] 3725217: DISPLAY DRV: DVO2(BP1) : Q:[2802] Display:[2808], Repeat:[7], DQ:[2800]
[m3video]      3731217: HDVICP-ID:0
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :52 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :52 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :54 %
[m3video]               totalAcq2acqDelay :45 %
[m3video]               totalElapsedTime in msec :   51358
[m3video]               numAccessCnt:    3081
[m3video]              IVA-FPS :      60
[m3video]              Average time spent per frame in microsec:    8667
[m3video]      3731218: HDVICP-ID:1
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :0 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :0 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :0 %
[m3video]               totalAcq2acqDelay :0 %
[m3video]               totalElapsedTime in msec :       0
[m3video]               numAccessCnt:       0
[m3video]              IVA-FPS :       0
[m3video]              Average time spent per frame in microsec:       0
[m3video]      3731218: HDVICP-ID:2
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :0 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :0 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :0 %
[m3video]               totalAcq2acqDelay :0 %
[m3video]               totalElapsedTime in msec :       0
[m3video]               numAccessCnt:       0
[m3video]              IVA-FPS :       0
[m3video]              Average time spent per frame in microsec:       0
[m3video]
[m3video] *** DECODE Statistics ***
[m3video]
[m3video] Elasped Time           : 34 secs
[m3video]
[m3video]
[m3video] CH | In Recv In User Out
[m3video] Num | FPS     Skip FPS FPS
[m3video] -----------------------------------
[m3video]    0 |      60        0 60
[m3video]
[m3video] Multi Channel Decode Average Submit Batch Size
[m3video] Max Submit Batch Size : 24
[m3video] IVAHD_0 Average Batch Size : 1
[m3video] IVAHD_0 Max achieved Batch Size : 1
[m3video] IVAHD_1 Average Batch Size : 0
[m3video] IVAHD_1 Max achieved Batch Size : 0
[m3video] IVAHD_2 Average Batch Size : 0
[m3video] IVAHD_2 Max achieved Batch Size : 0
[m3video]
[m3video] Multi Channel Decode Batch break Stats
[m3video] Total Number of Batches created: 2066
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 100 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video] Total Number of Batches created: 0
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 0 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video] Total Number of Batches created: 0
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 0 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video]
[m3vpss ]
[m3vpss ] 3734222: LOAD: CPU: 7.8% HWI: 2.2%, SWI:1.0%
[m3vpss ]
[m3vpss ] 3734222: LOAD: TSK: IPC_IN_M30          : 0.3%
[m3vpss ] 3734222: LOAD: TSK: DISPLAY0            : 0.5%
[m3vpss ] 3734222: LOAD: TSK: DISPLAY1            : 0.3%
[m3vpss ] 3734222: LOAD: TSK: DUP0                : 0.2%
[m3vpss ] 3734222: LOAD: TSK: SWMS0               : 1.2%
[m3vpss ] 3734223: LOAD: TSK: SWMS1               : 1.2%
[m3vpss ] 3734223: LOAD: TSK: MP_SCLR_FWD_Q0      : 0.2%
[m3vpss ] 3734223: LOAD: TSK: MISC                : 0.7%
[m3vpss ]
[m3video]
[m3video] 3734724: LOAD: CPU: 5.0% HWI: 0.7%, SWI:0.5%
[m3video]
[m3video] 3734724: LOAD: TSK: IPC_OUT_M30         : 0.7%
[m3video] 3734724: LOAD: TSK: IPC_BITS_IN0        : 0.3%
[m3video] 3734724: LOAD: TSK: DEC0                : 1.2%
[m3video] 3734724: LOAD: TSK: DEC_PROCESS_TSK_0   : 1.2%
[m3video] 3734724: LOAD: TSK: MISC                : 0.4%
[m3video]
[c6xdsp ]
[c6xdsp ] 3732311: LOAD: CPU: 0.2% HWI: 0.0%, SWI:0.0%
[c6xdsp ]
[c6xdsp ] 3732311: LOAD: TSK: MISC                : 0.2%
[c6xdsp ]

=============
Run-Time Menu
=============

1: Capture Settings
2: Encode Settings
3: Decode Settings
4: Display Settings
5: Audio Capture <TVP5158> & Encode <AAC-LC, G711> demo
6: Change Playback Channel <valid only if capture/playback is active>
7: Audio encode demo <File In/Out>
8: Audio decode demo <File In/Out>

i: Print detailed system information
s: Core Status: Active/In-active
f: Switch IVA Channel Map

e: Stop Demo

Enter Choice: i

[m3vpss ]
[m3vpss ] *** [MP_SCLR0 ] Statistics ***
[m3vpss ]
[m3vpss ] Total Frames Received : 4687
[m3vpss ] Total Frames Forwarded : 4687
[m3vpss ]
[m3vpss ]
[m3vpss ] CH | In Recv In Reject Processed Latency(DRV) Processed Rejected
[m3vpss ] Num | FPS     FPS       FPS        Min / Max    Frames     Frames
[m3vpss ] -------------------------------------------------------------------
[m3vpss ]
[m3vpss ] *** [SWMS0] Mosaic Statistics ***
[m3vpss ]
[m3vpss ] Elasped Time: 32 secs
[m3vpss ]
[m3vpss ] Output Request FPS   : 61 fps (1967 frames)
[m3vpss ] Output Actual FPS   : 61 fps (1967 frames)
[m3vpss ] Output Drop    FPS   : 0 fps (0 frames)
[m3vpss ] Output Reject FPS   : 0 fps (0 frames)
[m3vpss ] Scaling Internal     : 16 ms
[m3vpss ] Scaling Internal min : 16 ms
[m3vpss ] Scaling Internal max : 17 ms
[m3vpss ]
[m3vpss ] Win | Window Repeat Drop Recv Que FID Invlid Acc Event          Invalid   Que Reject Reject Latency   OutBufCopy InBufCopy
[m3vpss ] Num | FPS    FPS    FPS FPS FPS FPS        Count (Max/Min)    CH Frames Frames     Frames Min / Max FPS        FPS
[m3vpss ] ---------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |     61      0    0    0   61          0        0 ( 0/255)         0          0      0 35 / 40          0         0
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS0] Mosaic Parameters ***
[m3vpss ]
[m3vpss ] Output FPS: 60
[m3vpss ]
[m3vpss ] Win | Ch | Input      | Input          | Input         | Input       | Output     | Output         | Output        | Output      | Low Cost | SWMS | Data | Blank |
[m3vpss ] Num | Num | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | ON / OFF | Inst | Format| Frame |
[m3vpss ] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |   0 |    0,    0 | 1920 x   1080 | 16384 / 32768 | TILED       |    0,    0 | 1920 x   1080 | 3840 /      0 | NON-TILED   |      OFF |    0 | 420SP |   OFF |
[m3vpss ]
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS1] Mosaic Statistics ***
[m3vpss ]
[m3vpss ] Elasped Time: 32 secs
[m3vpss ]
[m3vpss ] Output Request FPS   : 61 fps (1967 frames)
[m3vpss ] Output Actual FPS   : 61 fps (1967 frames)
[m3vpss ] Output Drop    FPS   : 0 fps (0 frames)
[m3vpss ] Output Reject FPS   : 0 fps (0 frames)
[m3vpss ] Scaling Internal     : 16 ms
[m3vpss ] Scaling Internal min : 16 ms
[m3vpss ] Scaling Internal max : 17 ms
[m3vpss ]
[m3vpss ] Win | Window Repeat Drop Recv Que FID Invlid Acc Event          Invalid   Que Reject Reject Latency   OutBufCopy InBufCopy
[m3vpss ] Num | FPS    FPS    FPS FPS FPS FPS        Count (Max/Min)    CH Frames Frames     Frames Min / Max FPS        FPS
[m3vpss ] ---------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |     61      0    0    0   61          0        0 ( 0/255)         0          0      0 35 / 41          0         0
[m3vpss ]
[m3vpss ]
[m3vpss ] *** [SWMS1] Mosaic Parameters ***
[m3vpss ]
[m3vpss ] Output FPS: 60
[m3vpss ]
[m3vpss ] Win | Ch | Input      | Input          | Input         | Input       | Output     | Output         | Output        | Output      | Low Cost | SWMS | Data | Blank |
[m3vpss ] Num | Num | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | Start X, Y | Width x Height | Pitch Y / C   | Memory Type | ON / OFF | Inst | Format| Frame |
[m3vpss ] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------
[m3vpss ]    0 |   0 |    0,    0 | 1920 x   1080 | 16384 / 32768 | TILED       |    0,    0 | 1920 x   1080 | 3840 /      0 | NON-TILED   |      OFF |    0 | 420SP |   OFF |
[m3vpss ]
[m3vpss ]
[m3vpss ] 3757994: DISPLAY: HDDAC(BP0) : 59 fps, Latency (Min / Max) = ( 50 / 52 ), Callback Interval (Min / Max) = ( 16 / 17 ) DropCount:0 DispLatency (Min / Max) = ( 41 / 42 ) !!!
[m3vpss ] 3757994: DISPLAY DRV: HDDAC(BP0) : Q:[4768] Display:[4778], Repeat:[11], DQ:[4766]
[m3vpss ] 3757995: DISPLAY: UNDERFLOW COUNT: HDMI(BP0) 1967, HDDAC(BP0) 1967, DVO2(BP1) 1967, SDDAC(SEC1) 3934
[m3vpss ] 3757995: SYSTEM : FREE SPACE : System Heap      = 67320 B, Mbx = 10238 msgs)
[m3vpss ] 3757995: SYSTEM : FREE SPACE : SR0 Heap         = 16831744 B (16 MB)
[m3vpss ] 3757995: SYSTEM : FREE SPACE : Frame Buffer     = 599780224 B (571 MB)
[m3vpss ] 3757995: SYSTEM : FREE SPACE : Bitstream Buffer = 289430400 B (276 MB)
[m3vpss ] TILER_STATS: CNT :8BIT
[m3vpss ] TILER_STATS: CNT RESOLUTION:    16384 x 8192
[m3vpss ] TILER_STATS: BUCKET RESOLUTION: 2048 x 1662
[m3vpss ] TILER_STATS: NUM FREE BUCKETS: 37
[m3vpss ] TILER_STATS: NUM USED BUCKETS: 3
[m3vpss ] TILER_STATS: TOTAL FREE AREA:   126959616 (94 %)
[m3vpss ] TILER_STATS: TOTAL USED AREA:   7225344 (5 %)
[m3vpss ] TILER_STATS: TOTAL WASTE AREA: 0 (0 %)
[m3vpss ] TILER_STATS: MAX WIDTH RECT:    2048 x 456
[m3vpss ] TILER_STATS: MAX HEIGHT RECT:    2048 x 456
[m3vpss ] TILER_STATS: RES ALLOC COUNT [1080P] : 3
[m3vpss ] TILER_STATS: CNT :16BIT
[m3vpss ] TILER_STATS: CNT RESOLUTION:    32768 x 4096
[m3vpss ] TILER_STATS: BUCKET RESOLUTION: 2048 x 832
[m3vpss ] TILER_STATS: NUM FREE BUCKETS: 77
[m3vpss ] TILER_STATS: NUM USED BUCKETS: 3
[m3vpss ] TILER_STATS: TOTAL FREE AREA:   130605056 (97 %)
[m3vpss ] TILER_STATS: TOTAL USED AREA:   3612672 (2 %)
[m3vpss ] TILER_STATS: TOTAL WASTE AREA: 0 (0 %)
[m3vpss ] TILER_STATS: MAX WIDTH RECT:    2048 x 228
[m3vpss ] TILER_STATS: MAX HEIGHT RECT:    2048 x 228
[m3vpss ] TILER_STATS: RES ALLOC COUNT [1080P] : 3
[m3vpss ] 3757997: SYSTEM : FREE SPACE : Tiler 8-bit      = 126959616 B (121 MB) - TILER ON
[m3vpss ] 3757997: SYSTEM : FREE SPACE : Tiler 16-bit     = 130605056 B (124 MB) - TILER ON
[m3vpss ] 3757999: DISPLAY: DVO2(BP1) : 59 fps, Latency (Min / Max) = ( 50 / 50 ), Callback Interval (Min / Max) = ( 16 / 17 ) DropCount:0 DispLatency (Min / Max) = ( 41 / 42 ) !!!
[m3vpss ] 3757999: DISPLAY DRV: DVO2(BP1) : Q:[4769] Display:[4775], Repeat:[7], DQ:[4767]
[m3video]      3763999: HDVICP-ID:0
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :52 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :52 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :54 %
[m3video]               totalAcq2acqDelay :45 %
[m3video]               totalElapsedTime in msec :   84141
[m3video]               numAccessCnt:    5048
[m3video]              IVA-FPS :      60
[m3video]              Average time spent per frame in microsec:    8667
[m3video]      3764000: HDVICP-ID:1
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :0 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :0 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :0 %
[m3video]               totalAcq2acqDelay :0 %
[m3video]               totalElapsedTime in msec :       0
[m3video]               numAccessCnt:       0
[m3video]              IVA-FPS :       0
[m3video]              Average time spent per frame in microsec:       0
[m3video]      3764000: HDVICP-ID:2
[m3video] All percentage figures are based off totalElapsedTime
[m3video]               totalAcquire2wait :0 %
[m3video]               totalWait2Isr :0 %
[m3video]               totalIsr2Done :0 %
[m3video]               totalWait2Done :0 %
[m3video]               totalDone2Release :0 %
[m3video]               totalAcquire2Release :0 %
[m3video]               totalAcq2acqDelay :0 %
[m3video]               totalElapsedTime in msec :       0
[m3video]               numAccessCnt:       0
[m3video]              IVA-FPS :       0
[m3video]              Average time spent per frame in microsec:       0
[m3video]
[m3video] *** DECODE Statistics ***
[m3video]
[m3video] Elasped Time           : 32 secs
[m3video]
[m3video]
[m3video] CH | In Recv In User Out
[m3video] Num | FPS     Skip FPS FPS
[m3video] -----------------------------------
[m3video]    0 |      61        0 61
[m3video]
[m3video] Multi Channel Decode Average Submit Batch Size
[m3video] Max Submit Batch Size : 24
[m3video] IVAHD_0 Average Batch Size : 1
[m3video] IVAHD_0 Max achieved Batch Size : 1
[m3video] IVAHD_1 Average Batch Size : 0
[m3video] IVAHD_1 Max achieved Batch Size : 0
[m3video] IVAHD_2 Average Batch Size : 0
[m3video] IVAHD_2 Max achieved Batch Size : 0
[m3video]
[m3video] Multi Channel Decode Batch break Stats
[m3video] Total Number of Batches created: 1967
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 100 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video] Total Number of Batches created: 0
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 0 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video] Total Number of Batches created: 0
[m3video] All numbers are based off total number of Batches created
[m3video]       Batch breaks due to batch sizeexceeding limit: 0 %
[m3video]       Batch breaks due to ReqObj Que being empty: 0 %
[m3video]       Batch breaks due to changed resolution class: 0 %
[m3video]       Batch breaks due to interlace and progressivecontent mix: 0 %
[m3video]       Batch breaks due to channel repeat: 0 %
[m3video]       Batch breaks due to different codec: 0 %
[m3video]
[m3vpss ]
[m3vpss ] 3767004: LOAD: CPU: 7.8% HWI: 2.2%, SWI:1.0%
[m3vpss ]
[m3vpss ] 3767004: LOAD: TSK: IPC_IN_M30          : 0.3%
[m3vpss ] 3767004: LOAD: TSK: DISPLAY0            : 0.5%
[m3vpss ] 3767004: LOAD: TSK: DISPLAY1            : 0.3%
[m3vpss ] 3767005: LOAD: TSK: DUP0                : 0.2%
[m3vpss ] 3767005: LOAD: TSK: SWMS0               : 1.2%
[m3vpss ] 3767005: LOAD: TSK: SWMS1               : 1.2%
[m3vpss ] 3767005: LOAD: TSK: MP_SCLR_FWD_Q0      : 0.2%
[m3vpss ] 3767005: LOAD: TSK: MISC                : 0.7%
[m3vpss ]
[m3video]
[m3video] 3767506: LOAD: CPU: 5.0% HWI: 0.7%, SWI:0.5%
[m3video]
[m3video] 3767506: LOAD: TSK: IPC_OUT_M30         : 0.6%
[m3video] 3767506: LOAD: TSK: IPC_BITS_IN0        : 0.3%
[m3video] 3767506: LOAD: TSK: DEC0                : 1.2%
[m3video] 3767506: LOAD: TSK: DEC_PROCESS_TSK_0   : 1.2%
[m3video] 3767506: LOAD: TSK: MISC                : 0.5%
[m3video]
[c6xdsp ]
[c6xdsp ] 3765092: LOAD: CPU: 0.2% HWI: 0.0%, SWI:0.0%
[c6xdsp ]
[c6xdsp ] 3765092: LOAD: TSK: MISC                : 0.2%

I have highlighted the items of interest in bold.

As you can see from the stats latency from app sending the frame to SwMs outputting the composited frame is (min/max) (35 / 41 )

Latency from SwmS outputting the frame to display freeing the frame after display is 50 ms.

So this means effective total latency from when frame is sent to when frame starts display is:

Min = 35 + 50 - 16 = 69 ms

Max = 41 + 50 - 16 = 75 ms.

There is potential to reduce the latency further by 16 ms if you don't use SwMs and instead use scaler if you don't require composited view. This is because swms will operate at periodic intervals to enable composition from multiple windows which is not the case with scaler.

This stream is of low complexity so the decode processing time is very less as you can see in the stats 8667 usec avg to decode a frame. This could increase by 8msec for more complex streams.

Attached is the patch with list of all changes done to RDK so that you can try the same at your end. I am not sure if patch will apply cleanly on RDK 3.5 as it is based on latest codebase.I can anyway explain the changes in the patch if you require to replicate the setup at your end.

0131.Decode_display_latency_reduction.patch.txt

diff --git a/demos/mcfw_api_demos/mcfw_demo/demo_vdec_vdis_bits_rd.c b/demos/mcfw_api_demos/mcfw_demo/demo_vdec_vdis_bits_rd.c
index afd3854..def1a21 100755
--- a/demos/mcfw_api_demos/mcfw_demo/demo_vdec_vdis_bits_rd.c
+++ b/demos/mcfw_api_demos/mcfw_demo/demo_vdec_vdis_bits_rd.c
@@ -95,6 +95,7 @@ static void VdecVdis_bitsRdFillEmptyBuf(VCODEC_BITSBUF_S *pEmptyBuf)
 }
 
 #define VDEC_VDIS_FRAME_DURATION_MS (33)
+UInt64 Avsync_getWallTime();
 
 static Void VdecVdis_setFrameTimeStamp(VCODEC_BITSBUF_S *pEmptyBuf)
 {
@@ -102,6 +103,7 @@ static Void VdecVdis_setFrameTimeStamp(VCODEC_BITSBUF_S *pEmptyBuf)
       gVdecVdis_config.frameCnt[pEmptyBuf->chnId] * VDEC_VDIS_FRAME_DURATION_MS;
     pEmptyBuf->lowerTimeStamp = (UInt32)(curTimeStamp & 0xFFFFFFFF);
     pEmptyBuf->upperTimeStamp = (UInt32)((curTimeStamp >> 32)& 0xFFFFFFFF);
+    pEmptyBuf->timestamp = Avsync_getWallTime();
     if (0 == gVdecVdis_config.frameCnt[pEmptyBuf->chnId])
     {
         UInt32 displayChId;
@@ -148,7 +150,7 @@ static Void *VdecVdis_bitsRdSendFxn(Void * prm)
     OSA_semWait(&gVdecVdis_obj.thrStartSem,OSA_TIMEOUT_FOREVER);
     while (FALSE == gVdecVdis_obj.thrExit)
     {
-        OSA_waitMsecs(MCFW_IPCBITS_SENDFXN_PERIOD_MS);
+        OSA_waitMsecs(1);
 
         for (i = 0; i < gVdecVdis_config.numChannels; i++)
         {
diff --git a/mcfw/src_linux/mcfw_api/ti_vdis.c b/mcfw/src_linux/mcfw_api/ti_vdis.c
index af5e6a0..a0e0f61 100755
--- a/mcfw/src_linux/mcfw_api/ti_vdis.c
+++ b/mcfw/src_linux/mcfw_api/ti_vdis.c
@@ -801,7 +801,7 @@ Int32 Vdis_setMosaicParamsDefault(VDIS_DEV vdDevId, VDIS_MOSAIC_S *psVdMosaicPar
     /* Assign mosaic layout number and number of windows */
     vdisLayoutPrm.numWin = psVdMosaicParam->numberOfWindows;
     vdisLayoutPrm.onlyCh2WinMapChanged = psVdMosaicParam->onlyCh2WinMapChanged;
-    vdisLayoutPrm.outputFPS = psVdMosaicParam->outputFPS;
+    vdisLayoutPrm.outputFPS = 60;
 
     /* Assign each windows coordinates, size and mapping */
     for(winId=0; winId<vdisLayoutPrm.numWin; winId++)
diff --git a/mcfw/src_linux/mcfw_api/usecases/ti81xx/multich_vdec_vdis.c b/mcfw/src_linux/mcfw_api/usecases/ti81xx/multich_vdec_vdis.c
index 75e1b46..1549e7a 100755
--- a/mcfw/src_linux/mcfw_api/usecases/ti81xx/multich_vdec_vdis.c
+++ b/mcfw/src_linux/mcfw_api/usecases/ti81xx/multich_vdec_vdis.c
@@ -219,7 +219,7 @@ Void mulich_vdec_vdis_set_avsync_vidque_prm(Avsync_SynchConfigParams *queCfg,
         &&
         (gVsysModuleContext.vsysConfig.enableAVsync))
     {
-        queCfg->avsyncEnable = TRUE;
+        queCfg->avsyncEnable = FALSE;
     }
     else
     {
@@ -399,9 +399,9 @@ Void MultiCh_createVdecVdis()
     }
 
     ipcBitsOutHostPrm.baseCreateParams.outQueParams[0].nextLink= gVdecModuleContext.ipcBitsInRTOSId;
-    ipcBitsOutHostPrm.baseCreateParams.notifyNextLink       = FALSE;
+    ipcBitsOutHostPrm.baseCreateParams.notifyNextLink       = TRUE;
     ipcBitsOutHostPrm.baseCreateParams.notifyPrevLink       = FALSE;
-    ipcBitsOutHostPrm.baseCreateParams.noNotifyMode         = TRUE;
+    ipcBitsOutHostPrm.baseCreateParams.noNotifyMode         = FALSE;
     ipcBitsOutHostPrm.baseCreateParams.numOutQue            = 1;
     ipcBitsOutHostPrm.inQueInfo.numCh                       = gVdecModuleContext.vdecConfig.numChn;
 
@@ -430,16 +430,16 @@ Void MultiCh_createVdecVdis()
         ipcBitsOutHostPrm.chMaxReqBufSize[i] = 
                 (ipcBitsOutHostPrm.inQueInfo.chInfo[i].width * ipcBitsOutHostPrm.inQueInfo.chInfo[i].height); 
         ipcBitsOutHostPrm.totalBitStreamBufferSize [i] = 
-                (ipcBitsOutHostPrm.chMaxReqBufSize[i] * BIT_BUF_LENGTH_LIMIT_FACTOR_HD);
+                (ipcBitsOutHostPrm.chMaxReqBufSize[i] * 1);
 
     }
 
     ipcBitsInVideoPrm.baseCreateParams.inQueParams.prevLinkId    = gVdecModuleContext.ipcBitsOutHLOSId;
     ipcBitsInVideoPrm.baseCreateParams.inQueParams.prevLinkQueId = 0;
     ipcBitsInVideoPrm.baseCreateParams.outQueParams[0].nextLink  = gVdecModuleContext.decId;
-    ipcBitsInVideoPrm.baseCreateParams.noNotifyMode              = TRUE;
+    ipcBitsInVideoPrm.baseCreateParams.noNotifyMode              = FALSE;
     ipcBitsInVideoPrm.baseCreateParams.notifyNextLink            = TRUE;
-    ipcBitsInVideoPrm.baseCreateParams.notifyPrevLink            = FALSE;
+    ipcBitsInVideoPrm.baseCreateParams.notifyPrevLink            = TRUE;
     ipcBitsInVideoPrm.baseCreateParams.numOutQue                 = 1;
 
     for (i=0; i<ipcBitsOutHostPrm.inQueInfo.numCh; i++)
@@ -452,7 +452,7 @@ Void MultiCh_createVdecVdis()
             decPrm.chCreateParams[i].format                 = IVIDEO_MJPEG;
 
         decPrm.chCreateParams[i].numBufPerCh
-                         = gVdecModuleContext.vdecConfig.decChannelParams[i].numBufPerCh;
+                         = 3;
         decPrm.chCreateParams[i].profile                = IH264VDEC_PROFILE_ANY;
         decPrm.chCreateParams[i].displayDelay
                          = gVdecModuleContext.vdecConfig.decChannelParams[i].displayDelay;
@@ -639,10 +639,12 @@ Void MultiCh_createVdecVdis()
         swMsPrm[i].enableLayoutGridDraw = gVdisModuleContext.vdisConfig.enableLayoutGridDraw;
 
         MultiCh_swMsGetDefaultLayoutPrm(vdDevId, &swMsPrm[i], FALSE);    /* both from 0-16 chnl */
+        swMsPrm[i].layoutPrm.outputFPS = 60;
 
         displayPrm[i].inQueParams[0].prevLinkId    = gVdisModuleContext.swMsId[i];
         displayPrm[i].inQueParams[0].prevLinkQueId = 0;
         displayPrm[i].displayRes                = swMsPrm[i].initOutRes;
+        displayPrm[i].queueInISRFlag = TRUE;
         if (i == 1)
 #if defined(TI_814X_BUILD) || defined(TI_8107_BUILD)
                         displayPrm[i].displayRes            = gVdisModuleContext.vdisConfig.deviceParams[VDIS_DEV_SD].resolution;

0 Oliver King-Smith2 over 11 years ago in reply to Badri Narayanan

Prodigy 110 points

Badri and Shiju

Thank you for the timing info. Yes we know this is an easy to decode stream. We expect to use subframes in the final version.

After some digging, I think the code that is creating the framebuffers might be here

static int ti81xxfb_create_framebuffers(struct ti81xxfb_device *fbdev)
{
   int i, r;

   fbdev->num_fbs = 0;
   TFBDBG("create %d fbs\n", CONFIG_FB_TI81XX_NUM_FBS);

   /* allocate fb_info */
   for (i = 0; i < CONFIG_FB_TI81XX_NUM_FBS; i++) {

Where CONFIG_FB_TI81XX_NUM_FBS is set to 3. Do you know if this is where the hold up is?

You mention the 3 frame buffers is a hard / driver limitation. Does anyone know if this path can be shortened and if so what are the consequences of doing this?

Oliver

0 Oliver King-Smith2 over 11 years ago in reply to Oliver King-Smith2

Prodigy 110 points

I see in the HDVPSS UserGuide that

FVID2 Start

This API is used by the application to start the display operation. This is a blocking call and returns after starting the display operation. Before starting the display operation, the application has to prime at least 1 buffer with the driver using queue API. When driver is openned under VPS_GRPX_NON_FRAME_BUFFER_MODE, typically 3 buffers are used - 1 used by application and 2 buffers are queued with the driver at any given time. When driver is openned under VPS_GRPX_FRAME_BUFFER_MODE, typically 1 buffer is used.

Would it help latency if we opened the buffer in VPS_GRPX_FRAME_BUFFER_MODE?

Oliver

0 Badri Narayanan over 11 years ago in reply to Oliver King-Smith2

TI__Guru 59700 points

I don't understand the question.Are you measuring latency for the video pipeline or graphics pipeline.FBDEV is for graphics .The priming buffer length for FBDEV has no impact on video display latency and vice-versa.

0 Oliver King-Smith2 over 11 years ago in reply to Badri Narayanan

Prodigy 110 points

Badri,

We are interested in Video pipeline latency, not graphics. Sorry for the confusion. We are trying to figure out why there is a 3 buffer delay. The 48ms delay that this imposes is painful. I assume this is something on the M3?

Oliver

0 Badri Narayanan over 11 years ago in reply to Oliver King-Smith2

TI__Guru 59700 points

The three buffer for display is minimum .You need 2 buffers ( Ping-Pong) so that on display interrupt the next frame begins display without underrun. A third buffer is required because frames are queued to display driver in a separate thread. The three buffer delay can be reduced to two buffers if display frames are queued in the display ISR instead of the thread context. This is what is done in RDK as you can see in the logs I shared where the display latency is ~ 33ms (2 frames @ 60 fps). Further reduction in display latency is not possible per my understanding. I will ask the HDVPSS driver expert to confirm on this thread.

0 Oliver King-Smith2 over 11 years ago in reply to Badri Narayanan

Prodigy 110 points

Badri

Thanks for getting back to me. I may have misunderstood this line

[m3vpss ] 3725213: DISPLAY: HDDAC(BP0) : 59 fps, Latency (Min / Max) = ( 50 / 51 ), Callback Interval (Min / Max) = ( 16 / 17 ) DropCount:0 DispLatency (Min / Max) = ( 33 / 42 ) !!!

I saw this meant the total latency for the display step was 50/51 ms? Are you saying it was 33/42?

Oliver

0 Badri Narayanan over 11 years ago in reply to Oliver King-Smith2

TI__Guru 59700 points

As I explained in the previous post 50/51 is the latency measure at the point where the display has completed display of frame and frees the frame back to previous component. This means frame started display one VSYNC previously which means display latency is 50 - 16 ~ 34 ms.

0 Brijesh Jadav over 11 years ago in reply to Badri Narayanan

TI__Guru**** 418385 points

Hi,

The minimum latency that we could achieve by queueing/dequeueing from the ISR, is two frame time, ie 33/34 ms for 60fps. Here latency is difference betwee Dequeue and queue operation. Buffer actually starts getting displayed after 16ms.

Regards,

Brijesh Jadav

0 Philipp Schrader over 11 years ago in reply to Brijesh Jadav

Processors

Processors forum

[DM8168] Video Display Latency