-cpu-used=3

[GoogleCodeExporter]

What is the expected behavior?
cpu-used=3 should produce higher quality than cpu-used=4

What do you see instead?

cpu-used=2
Stream 0 PSNR (Overall/Avg/Y/U/V) 38.725 38.917 37.774 41.922 44.330

cpu-used=3 
Stream 0 PSNR (Overall/Avg/Y/U/V) 33.816 34.946 33.732 38.211 41.483

cpu-used=4
Stream 0 PSNR (Overall/Avg/Y/U/V) 38.001 38.277 37.086 41.560 44.128

What version are you using?
aug 29 VP9 Encoder v1.2.0-4156-ge80bf80

On what operating system?
win64

Can you reproduce using the vpxdec or vpxenc tools? What command line are you 
using?

vpxenc --cpu-used=3 -t 1 -w 640 -h 360 --fps=30000/1001 --target-bitrate=400 
bear.640x360_30Hz_P420.yuv -o bear0.tmp.vp9.webm --fpf=bear_vp9.fpf -p 2 
--pass=2 --codec=vp9 --good --lag-in-frames=25 --min-q=0 --max-q=63 
--end-usage=vbr --auto-alt-ref=1 --kf-max-dist=9999 --kf-min-dist=0 
--drop-frame=0 --static-thresh=0 --bias-pct=50 --minsection-pct=0 
--maxsection-pct=2000 --arnr-maxframes=7 --arnr-strength=5 --arnr-type=3 
--sharpness=0 --undershoot-pct=100 -v --psnr 

Please provide any additional information below.

Original issue reported on code.google.com by fbarch...@google.com on 1 Sep 2013 at 5:39

[GoogleCodeExporter]

Overall PSNR is 34

e:\mediatests\bear>vpxenc --cpu-used=3 -t 1 -w 640 -h 360 --fps=30000/1001 
--target-bitrate=400 bear.640x360_30Hz_P420.yuv -o bear0.tmp.vp9.we
bm --fpf=bear_vp9.fpf -p 2 --pass=2 --codec=vp9 --good --lag-in-frames=25 
--min-q=0 --max-q=63 --end-usage=vbr --auto-alt-ref=1 --kf-max-dist=
9999 --kf-min-dist=0 --drop-frame=0 --static-thresh=0 --bias-pct=50 
--minsection-pct=0 --maxsection-pct=2000 --arnr-maxframes=7 --arnr-strengt
h=5 --arnr-type=3 --sharpness=0 --undershoot-pct=100 -v --psnr
Pass 2/2 frame   82/58    131678B 3004001 us 27.30 fps [ETA  0:00:01] 
←[K32.409 31.137 35.707 40.105    1138F32.092 30.790 35.631 40.075    12
69F32.030 30.755 35.317 39.805    4189F31.755 30.463 35.198 39.660    
1116F31.529 30.202 35.296 39.736    1022F31.383 30.037 35.301 39.909
 962F31.294 29.947 35.212 39.845     769F31.219 29.867 35.227 39.703     580F31.266 29.920 35.238 39.628     678F31.542 30.230 35.187 39.581
   774F31.445 30.109 35.292 39.766    1074F31.606 30.291 35.245 39.759     910F31.404 30.078 35.152 39.635    1028F31.492 30.156 35.437 39.559
    1046F31.893 30.581 35.658 39.595     834F31.986 30.698 35.527 39.532     755F31.580 30.256 35.465 39.401     753F31.603 30.288 35.411 39.3
08     869F31.321 30.001 35.201 39.027    1341F31.510 30.217 35.087 39.100     
766F31.335 30.021 35.150 39.015     895F31.392 30.071 35.314 39
Pass 2/2 frame   82/82    156501B   15268b/f  457594b/s 3759042 us (21.81 
fps)←[K505F
Stream 0 PSNR (Overall/Avg/Y/U/V) 34.257 35.365 34.152 38.606 41.870

But Luma drops to 29.8 on frame 65.

vpxdec --codec=vp9 --limit=10000 --i420 bear0.vp9.webm -o 
bear_vp9.webm.640x360_30Hz_P420.yuv

>psnr -s 640 360 bear_vp9.webm.640x360_30Hz_P420.yuv bear.640x360_30Hz_P420.yuv
Frame    PSNR-Y          PSNR-U          PSNR-V          PSNR-All        Frame
Global:  32.987857       37.720038       41.414968       34.256602         82
Avg:     34.152023       38.606474       41.869965       35.364642         82
Min:     29.866801       35.087410       39.013800       31.218834         65

Original comment by fbarch...@google.com on 6 Sep 2013 at 6:44

[GoogleCodeExporter]

I think Paul reported similar observations and is working on get the speed 
steps sorted out.

Original comment by ya...@google.com on 9 Sep 2013 at 5:56

[GoogleCodeExporter]

Using VP9 build from Sept 6, v1.2.0-4213-gd1268c5

cpu-used     PSNR   Time to Encode
--cpu-used=0 40.758 112788 ms 
--cpu-used=1 40.001 16087 ms 
--cpu-used=2 38.877 4887 ms
--cpu-used=3 34.579 3263 ms   <-- bug
--cpu-used=4 38.381 2631 ms
--cpu-used=5 40.571 55703 ms

Original comment by fbarch...@google.com on 9 Sep 2013 at 9:54

[GoogleCodeExporter]

Original comment by ya...@google.com on 17 Oct 2013 at 10:33

• Changed state: Assigned

[GoogleCodeExporter]

Speeds above 2 were still experimental and being actively used to try things 
out when this was posted.

As of now Speed 0-3 are reasonably stable but above that things may still 
behave a bit oddly.

Original comment by paulwilk...@google.com on 18 Oct 2013 at 8:40

• Changed state: Fixed