DM365 - DVSDK 2.10

Environment

The results were taken using a DM365 EVM (Rev C) board with DVSDK 2.10.00.16
The settings for the system are, ARM @297 MHz, HDVICP @243 MHz, DDR @243 MHz, Monta Vista® Linux® 5.0
NFS mounted File system is used
YUV 420 Semi Planar (NV12) chroma format is used
No quality analysis was performed for the results
H.264 Video Encoder/Decoder: High Profile [HP] upto Level 3.1 Compliant.
MPEG4 Video Encoder: Simple Profile [SP] Level 0, 1, 2, 3 Compliant.
MPEG4 Video Decoder: Restricted Simple Profile [SP] Level 0, 1, 2, 3 Compliant.

Results

**Table 1**: Decoding Performance results. (*Click on a footnote to read its content*)
DECODING PERFORMANCE STATISTICS [ms/frame]
*TEST METHOD^()**				DSP ONLY		FILE to FILE		OOB DEMO			GSTREAMER		CONDITIONS
CODEC	VIDEO SIZE	RATE¹	MV²	AVG	MAX	AVG	MAX	FPS³	AVG	MAX	AVG	MAX	Clip Used⁴	PRESET
H264DEC	720P	4Mbps	N/A	N/A	N/A	26.05	29.10	N/A	30.73	32.81	26.04	28.44	`parkrun`	Default
H264DEC	D1	2Mbps	N/A	N/A	N/A	11.74	13.17	N/A	11.87	13.28	11.68	13.10	`Shields`	Default
H264DEC	CIF	512Kbps	N/A	N/A	N/A	5.07	5.59	N/A	5.20	5.63	4.98	5.32	`Akiyo`	Default
MPEG4DEC	720P	3Mbps	N/A	N/A	N/A	19.03	19.57	N/A	25.34	27.04	19.77	10.86	`Stockholm`	Default
MPEG4DEC	D1	6Mbps	N/A	N/A	N/A	10.33	10.70	N/A	10.43	11.31	10.40	10.61	`Football`	Default
MPEG4DEC	CIF	1.5Mbps	N/A	N/A	N/A	3.85	4.52	N/A	3.85	4.24	3.90	4.01	`Bus`	Default

**Table 2**: Encoding Performance results. (*Click on a footnote to read its content*)
ENCODING PERFORMANCE STATISTICS [ms/frame]
*TEST METHOD^()**				DSP ONLY		FILE to FILE		OOB DEMO			GSTREAMER		CONDITIONS
CODEC	VIDEO SIZE	RATE¹	MV²	AVG	MAX	AVG	MAX	FPS³	AVG	MAX	AVG	MAX	Clip Used⁴	PRESET
H264ENC	720P	3Mbps	N/A	N/A	N/A	28.96	39.45	N/A	30.58	39.78	N/A	N/A	`Parkrun`	High Speed
H264ENC	D1	1Mbps	N/A	N/A	N/A	11.99	15.94	N/A	13.65	17.45	N/A	N/A	`Shields`	High Speed
H264ENC	CIF	100Kbps	N/A	N/A	N/A	5.44	6.52	N/A	N/A	N/A	N/A	N/A	`Akiyo`	High Speed
MPEG4ENC	720P	2Mbps	N/A	N/A	N/A	20.04	20.75	N/A	21.94	23.76	N/A	N/A	`Stockholm`	High Speed
MPEG4ENC	D1	1Mbps	N/A	N/A	N/A	8.57	8.81	N/A	10.29	10.60	N/A	N/A	`Football`	High Speed
MPEG4ENC	CIF	512Kbps	N/A	N/A	N/A	3.71	4.11	N/A	N/A	N/A	N/A	N/A	`Bus`	High Speed

Analysis

Figure 1 shows the change in decoder process time with different processor loading for H.264 stream with 720P resolution at 30fps.

Figure 1: Video decoder processing delay for 720P resolution H.264 stream at 30fps.

In case of decoder, the process time there is about 18% increase in process/decode time with decode demo application for 720P resolution clip. Addition of other components like OSD, speech(G.711 speech with 64 kbit/s bitrate) does not show significant performance change for DM365. The codec performance with gstreamer is almost same as file_to_file case. This looks like the ideal case, but the output video quality of the decode demo application is better.

Figure 2 shows the change in encoder process time with different processor loading for H.264 stream with 720P resolution at 30fps.

Figure 2: Video encoder processing delay for 720P resolution H.264 stream at 30fps.

In case of encoder, there is about 5% increase in process time with capture+display+OSD+speech(G.711 speech with 64 kbit/s bitrate) loading from a file to file encoding for 720P encoding. One possible reason for relatively lower penalty in encode demo application might be, in case of 720P resolution encoding there is no framecopy in encode demo application.

Conclusions

Implementation of larger FIFO queue between video and capture/display threads absorbs the video processing time irregularities, but it might come with a performance and memory penalty
Framecopy in the application should be avoided as much as possible; even if it uses DMA based transfer to copy frames, it might cause increase in process time
The process times for encode/decode sometimes crosses the expected process time for a 30fps stream; maintaining queues between process thread and capture/display thread helps achieve the final 30fps goal

Note

The DVSDK software version used for this test is 2.10.00.16. But the results should not change significantly for DVSDK 2.10 GA (DVSDK version 2.10.00.17).
These performance measurement are taken while the codec software is still changing. The final version of the software is expected to have better performance characteristics than shown here.