C64x+ and Davinci codec performance tables

From Texas Instruments Wiki
Jump to: navigation, search

Scope

  • Document provides codec performances for
    • Video Encoder and Decoder (DM6446, DM6467, DM648, C64x)
    • Video Post Processing (DM6467)
    • C64x+ based Resize kernels
  • Assumptions
    • Performance numbers are based on DSP and Coprocessor usage.
    • Performance numbers listed are per channel and they are the peak performances for a given bitstream.
    • C64x+ based performance can be used for devices using only C64x+ core.
    • Multichannel codecs are available only from TI Third Parties
    • System Level Analysis should be done for any application. Such as:
      • DDR bandwidth usage
      • Video Input and Output capabilities of the device.
  • Terminology


DM6467 - 729 Mhz Video Decoder

Video Decoder Profile Resolution Frame Rate Channel Density DDR BW
(MB/s)
HDVICP
(MHz)
CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
H264 BP 1080p 30 2 750 280 320 Upto 20 Mbps riverbed_1920x1080_420p_BP.264 RFS
H264 BP 720p 30 3 330 124 140 Upto 14 Mbps Available RFS
H264 BP D1 30 8 124 46 55 Upto 10 Mbps Available RFS
H264 BP CIF 30 16 46 18 16 Upto 4 Mbps Available RFS
Video Decoder Profile Resolution Frame Rate Channel Density HDVICP
(MHz)
CPU/ DSP (MHz) Input Bit Rate Status Software Channel*
MPEG2 MP@ HL 1080p 30 2 270 10 Upto 80 Mbps Available RFS
MPEG2 MP 720p 30 5 120 5 Upto 60 Mbps Available RFS
MPEG2 MP@ ML D1 30 10 45 2 Upto 15 Mbps Available RFS
MPEG2 MP@ ML CIF 30 16 13 0.5 Upto 4 Mbps Available RFS


DM6467 - 594 MHz Video Decoder

Video Decoder Profile Resolution Frame Rate Channel Density DDR BW
(MB/s)
HDVICP
(MHz)
CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
H264 BP 1080p 30 1 750 280 320 Upto 20 Mbps riverbed_1920x1080_420p_BP.264 RFS
H264 BP 720p 30 2 330 124 140 Upto 14 Mbps Available RFS
H264 BP D1 30 4 124 46 55 Upto 10 Mbps Available RFS
H264 BP CIF 30 12 46 18 16 Upto 4 Mbps Available RFS
Video Decoder Profile Resolution Frame Rate Channel Density HDVICP
(MHz)
CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
MPEG2 MP@ HL 1080p 30 2 270 10 Upto 80 Mbps Available RFS
MPEG2 MP 720p 30 4 120 5 Upto 60 Mbps Available RFS
MPEG2 MP@ ML D1 30 8 45 2 Upto 15 Mbps Available RFS
MPEG2 MP@ ML CIF 30 12 13 0.5 Upto 4 Mbps Available RFS


DM6467 - 729 MHz Video Encoder

Video Encoder Profile High Speed/ Quality Resolution Frame Rate Channel Density HDVICP
(MHz)
CPU/ DSP (MHz) Latency Output Bit Rate Features/Status Software Channel*
H264 BP Speed 720p 30 1 270 339 < 85ms Upto 14 Mbps Available, CABAC coding, No 4 MV, Limited MP, HP with 8x8 transform support RFS
H264 BP Speed D1 30 5 101 135 < 50 ms Upto 10 Mbps Available, No 4 MV, Both HDVICP and DSP RFS
H264 BP Speed CIF 30 16 37 34 < 50 ms Upto 4 Mbps Available, No 4 MV, Both HDVICP and DSP RFS

DM6467 - 594 MHz Video Encoder

Video Encoder Profile High Speed/ Quality Resolution Frame Rate Channel Density HDVICP (MHz) CPU/ DSP (MHz) Latency Output Bit Rate Features/Status Software Channel*
H264 BP Speed 720p 30 1 270 339 < 85ms Upto 14 Mbps Available, CABAC coding, No 4 MV, Limited MP, HP with 8x8 transform support RFS
H264 BP Speed D1 30 4 101 135 < 50 ms Upto 10 Mbps Available, No 4 MV, Both HDVICP and DSP RFS
H264 BP Speed CIF 30 12 37 34 < 50 ms Upto 4 Mbps Available, No 4 MV, Both HDVICP and DSP RFS


DM6446 - 594 MHz Video Decoder

Video Decoder Profile Resolution Frame Rate Channel Density CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
H264 BP CIF 30 4 121 upto 2 Mbps traffic_multiple_slice_aso.264,YUV422ILE, only ASO present on all frames RFS
H264 BP D1 30 1 336 upto 4 Mbps sheilds_720x480_1MV.264 , YUV422ILE RFS
MPEG4 SP CIF 30 8 64 upto 4 Mbps Available, non erroneous streams without DP/RVLC/error concealment/post-processing routines
cif_high_100f_fixedqp20_nofilter.yuv
RFS
MPEG4 SP VGA 30 1 340 upto 4 Mbps Deblocking Derining VGA
oreman_vga_dp0.yuv, MPEG4 (YUV420)
RFS
MPEG4 SP D1 30 2 199 upto 10 Mbps non erroneous streams without DP/RVLC/error concealment/post-processing routines RFS
MPEG2 MP@ML CIF 30 8 60 upto 4 Mbps Available RFS
MPEG2 MP@ML D1 30 2 211 upto 10 Mbps Available RFS
MPEG2 MP 720p 30 1 561 upto 14 Mbps Available RFS


DM6446 - 594 MHz Video Encoder

Video Encoder Profile High Speed/ Quality Resolution Latency Frame Rate Channel Density CPU/ DSP (MHz) Output Bit Rate Features/Status Software Channel*
H264 BP Speed CIF < 50 ms 30 3 146 upto 2 Mbps Without 4-MV and early-skip support RFS
H264 BP Speed D1 < 50 ms 30 1 385 upto 4 Mbps Without 4-MV and early-skip support RFS
H264 BP Quality CIF < 50 ms 30 2 146 upto 2 Mbps mobile_p352x288_30fps_420pl_300fr IntraPeriod=1sec (encoding Frame rate=30 fps), rcAlgo=PLR4_RC,
NFAVG=100
RFS
H264 BP Quality D1 < 50 ms 30 1 452 upto 4 Mbps sheilds_720x480_422i_252frames.yuv, rcAlgo= PLR4_RC, NFAVG=252 RFS
MPEG4 SP Speed CIF < 50 ms 30 6 83 upto 2 Mbps Available, Without RPS ME RFS
MPEG4 SP Speed D1 < 50 ms 30 2 256 upto 5 Mbps Without RPS ME,
Fire_420.yuv, YUV420/NTSC 1MV, HPI on, UMV on
RFS
MPEG4 SP Speed 720p < 85 ms 25 1 533 upto 12 Mbps Available, half-pel ME disabled, higher frequency of I-frames RFS
MPEG4 SP Quality CIF < 50 ms 30 5 95 upto 2 Mbps Available, With RPS ME + RC improvements RFS
MPEG4 SP Quality D1 < 50 ms 30 1 290 upto 5 Mbps Available, With RPS ME + RC improvements RFS
JPEG 422 BL, Seq   D1   30 3 165 upto 5 Mbps Available RFS


DM648 - 900MHz Video Decoder

Video Decoder Profile Resolution Frame Rate Channel Density CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
H264 BP CIF 30 7 112 Upto 2096 traffic_multiple_slice_aso.264, YUV422ILE, only ASO
present on all frames, CIF (352x288)
RFS
H264 BP D1 30 3 293 Upto 4096 foreman_vga_1mbps_100f.264, YUV422ILE, 4 MV, VGA 201 293
(640x480), @ 1 mbps
RFS
MPEG4 SP CIF 30 16 72 Upto 4096 cif_high_fixedqp20_nofilter.yuv YUV420 RFS
MPEG4 SP D1 30 4 212 upto 10 Mbps hp_720x480.yuv, MPEG4, YUV420. The performance with Deringing, Deblocking enabled 354 MHz RFS
MPEG4 SP 720p 30 1 563 upto 14 Mbps Available RFS
MPEG2 MP CIF 30 16 50 Upto 4096 Available RFS
MPEG2 MP D1 30 4 190 upto 10 Mbps Available RFS


DM648 - 900 MHz Video Encoder

Video Encoder Profile High Speed/ Quality Resolution Latency Frame Rate Channel Density CPU/ DSP (MHz) Output Bit Rate Features/Status Software Channel*
H264 BP speed CIF < 50 ms 30 6 118 upto 2 Mbps Available, NO 4-MV RFS
H264 BP speed D1 < 50 ms 30 2 399 upto 4 Mbps Tennis.yuv, YUV420/704x480 RFS
H264 BP Quality CIF < 50 ms 30 6 130 upto 2 Mbps mobile.yuv, YUV420/352x288 @ 768 kbps @ 30 fps
with 1 MV, QPI, LPF, UMV, high quality preset, 100
frames
RFS
H264 BP Quality D1 < 50 ms 30 1 435 upto 4 Mbps ti_commercial_720x576.yuv, 1 MV, QPI, LPF, UMV-enabled RFS
MPEG4 SP Quality CIF < 50 ms 30 6 103 upto 2 Mbps Available, RPS (ME) + RC improvements and at lower bitrates RFS
MPEG4 SP Quality D1 < 50 ms 30 2 352 upto 5 Mbps Fire_420.yuv, YUV420/NTSC D1 @ 4 Mbps with 1MV, HPI on. UMV on. RFS
MPEG4 SP Speed CIF < 50 ms 30 7 106 upto 2 Mbps Mobile.yuv, YUV420/CIF @ 512 Kbps with 1MV, HPI on. UMV on. RFS
MPEG4 SP Speed D1 < 50 ms 30 2 282 upto 5 Mbps Available on DM6446, Without RPS ME,
Fire_420.yuv, YUV420/NTSC D1 with 1MV, HPI on, UMV on
RFS
JPEG 422 BL, Seq 768x512 768x512   30 3 40 Mpixel/sec   Available, YUV 422 input RFS


C64x+ Video Decoder

Video Decoder Profile Resolution Frame Rate CPU/ DSP (MHz) Input Bit Rate Features/Status Software Channel*
H264 BP CIF 30 121 upto 2 Mbps traffic_multiple_slice_aso.264,YUV422ILE, only ASO present on all frames RFS
H264 BP D1 30 336 upto 4 Mbps sheilds_720x480_1MV_1_5Mbps.264 , YUV422ILE, 1MV, D1 (720x480) @ 220 336
1.5Mbps
RFS
MPEG4 SP CIF 30 64 upto 4 Mbps Available, non erroneous streams without DP/RVLC/error concealment/post-processing routines
cif_high_256kbps_100f_fixedqp20_nofilter.yuv, MPEG4,
YUV420, CIF @ 256 kbps
RFS
MPEG4 SP VGA 30 340 upto 4 Mbps Deblocking Derining VGA
oreman_vga_dp0.yuv, MPEG4 (YUV420, VGA @ 3 mbps)
RFS
MPEG4 SP D1 30 199 upto 10 Mbps non erroneous streams without DP/RVLC/error concealment/post-processing routines RFS
MPEG4 SP 720p 30 529 upto 14 Mbps non erroneous streams without DP/RVLC/error concealment/post-processing routines RFS
MPEG2 MP@ML CIF 30 60 upto 4 Mbps Available RFS
MPEG2 MP@ML D1 30 211 upto 10 Mbps Available RFS
MPEG2 MP 720p 30 561 upto 14 Mbps Under development RFS


Important Notes

referring to the current C64x+ codecs download page: [1]

  • the current implementation of C64x+ H264 decoder v2.00.002 does not support DM648 or C6455 since the values for the QDMA internal addresses are hard coded, and the memory map for these devices is different than the memory map for the DM644x/DM646x/DM643x
Workaround: for C645x, the DM648 native version can be used, though, since the decoder does not use VICP resources, and the memory map is similar. This H264 decoder version 1.12 is provided in the DVSDK 1.11 for DM648
  • the current implementation of C64x+ MPEG4 SP Decode – v2.00.011 does not support DM648 or C6455 since the values for the QDMA internal addresses are hard coded, and the memory map for these devices is different than the memory map for the DM644x/DM646x/DM643x
Workaround: for C645x, the DM648 native version can be used, though, since the decoder does not use VICP resources, and the memory map is similar. This Mpeg-4 decoder version 1.11 is provided in the DVSDK 1.11 for DM648


  • the current implementation of C64x+ MPEG2 Decode – v2.00.010 does not support DM648 or C6455 since the values for the QDMA internal addresses are hard coded, and the memory map for these devices is different than the memory map for the DM644x/DM646x/DM643x
Workaround: Mpeg-2 is not offered for DM648 devices so it is not possible to use it
  • Jpeg decode and audio decode should not have any issue since they do not use QDMA
  • the download page will need to be changed accordingly


To help in porting each application using a specific codec on a different platform please refer to application note sprab73 It explains very clearly what the changes are across devices and what needs to be done

C64x+ Video Encoder

Video Encoder Profile High Speed/ Quality Resolution Latency Frame Rate CPU/ DSP (MHz) Output Bit Rate Features/Status Software Channel*
H264 BP speed CIF < 50 ms 30 165 upto 2 Mbps Available, NO 4-MV RFS
H264 BP speed D1 < 50 ms 30 600 upto 4 Mbps Available, NO 4-MV RFS
H264 BP Quality CIF < 50 ms 30 168 upto 2 Mbps mobile.yuv, YUV422/CIF High Quality Preset RFS
H264 BP Quality D1 < 50 ms 30 674 upto 4 Mbps ti_commercial_720x576.yuv,
1 MV, QPI, LPF, UMV enabled, High
Quality Preset, 75 frames
RFS
MPEG4 SP Quality CIF < 50 ms 30 100 upto 2 Mbps Mobile.yuv, YUV420
with 1MV, HPI on. UMV on.
RFS
MPEG4 SP Quality D1 < 50 ms 30 359 upto 5 Mbps Fire_422.yuv, YUV422/NTSC 1MV, HPI on. UMV on. RFS
MPEG4 SP Speed CIF < 50 ms 30 92 upto 2 Mbps Mobile.yuv, YUV420/CIF 1MV, HPI on. UMV on. RFS
MPEG4 SP Speed D1 < 50 ms 30 332 upto 5 Mbps Fire_422.yuv, YUV422/NTSC 1MV, HPI on. UMV on. RFS


Important notes

referring to the C64x+ codecs download page: [2]

  • the current implementation of C64x+ MPEG4 Encode– v2.02.00.02, C64x+ H.264 BP Encode – 02.02.01, C64x+ MPEG2 Decode – v2.00.010 do not support DM648 or C645x since the values for the QDMA internal addresses are hard coded, and the memory map for these devices is different than the memory map for the DM644x/DM646x/DM643x

Workaround: since the H264 and Mpeg-4 native encoders for DM648 use the VICP, they cannot be used on C645x, so it is not possible to use them

  • Jpeg encode and audio encoders should not have any issue since they do not use QDMA
  • the download page will need to be changed accordingly

To help in porting each application using a specific codec on a different platform please refer to application note sprab73 It explains very clearly what the changes are across devices and what needs to be done

C64x+ Based Resizing Kernels

  • Resize Performance is simulated using horizontal and vertical kernels. The resizing kernels in the Image Lib can be used to generate resizing algorithms. The resizing algorithms are not available from TI in the Image library.
  • Optimized Polyphase filter kernel is available for upto 64 filter taps
    • Sub partitioned as 4 tap – 16 phase, 8 tap – 8 phase, 12 tap - 5 phase and 16 tap - 4 phase polyphase filter
    • Number of tabs can increment in multiple of 4 upto 16 taps.
  • Assumptions
    • Data is for YUY420 Image with 16 bit input and 16 bit coefficient.
    • Data is in the internal memory.
    • Sequence of Downsizing
      • First horizontal resizing and then vertical resizing.
      • This is to enable better performance in vertical filtering.
    • Sequence of Upsizing
      • First vertical resizing and then horizontal resizing.
      • This is to enable better performance in vertical filtering.
  • DSP is running at 594 MHz.
  • The performance numbers are in Microseconds.


Input Output 320 x 240 720 x 480 1280 x 720
16-tap
1920 x 1080 9299 21350 48113
1280 x 720 6913 16033 x
720 x 480 5323 x x
12-tap  
1920 x 1080 7556 16299 35321
1280 x 720 5750 12291 x
720 x 480 4550 x x
8-tap      
1920 x 1080 5806 11242 21557
1280 x 720 4584 8543 x
720 x 480 3770 x x
Upscale      
Output Input 320 x 240 720 x 480 1280 x 720
16 -tap
1920 x 1080 48962 54438 72093
1280 x 720 21811 26919 x
720 x 480 10054 x x
12 -tap      
1920 x 1080 48873 52976 58737
1280 x 720 19181 23126 x
720 x 480 7793 x x
8-tap      
1920 x 1080 25690 28425 32256
1280 x 720 11464 14235 x
720 x 480 5527 x x


See Also