madshi bug tracker - madVR
View Issue Details
0000193madVRbugpublic2014-04-11 16:392015-03-22 14:59
huhn 
madshi 
normalminorhave not tried
closedfixed 
x64windows 77200
87.10
mpc-be 13.1.1 4624
lavfilter 61.1
lavfilter 61.1
Software
auto mode
no
On
all modes
AMD
x9 270
14.3
0000193: image corruption when up and downscale it needed at the same time
when a pixel shader down scaler and bilinear up scaler is used with a 1440x1080 16/9 ar source and both is needed up and downscaling then the picture gets corrupted looks like nearest neighbor is used for upscale.
set up scaling to bilinear and downscaling to a pixel shader scaler like spline 3
open a 1440x1080p 16/9 common TV steam
scale it to 1600x900 or some way a up and a down scaler is needed.
this is just a reminder madshi you said to my you can reproduce this.
No tags attached.
Issue History
2014-04-11 16:39huhnNew Issue
2014-04-11 16:49madshiNote Added: 0000594
2014-04-11 16:49madshiAssigned To => madshi
2014-04-11 16:49madshiStatusnew => feedback
2014-04-11 16:56huhnNote Added: 0000595
2014-04-11 16:56huhnStatusfeedback => assigned
2014-04-11 17:03huhnNote Added: 0000596
2014-04-18 19:21madshiNote Added: 0000604
2014-04-18 19:21madshiStatusassigned => closed
2014-04-18 19:21madshiResolutionopen => fixed
2014-04-23 16:42huhnmadVR Version87.9 => 87.10
2014-04-23 16:42huhnNote Added: 0000612
2014-04-23 16:42huhnStatusclosed => feedback
2014-04-23 16:42huhnResolutionfixed => reopened
2014-04-23 19:42huhnNote Added: 0000613
2014-04-23 19:42huhnStatusfeedback => assigned
2015-03-22 14:59madshiNote Added: 0000808
2015-03-22 14:59madshiStatusassigned => closed
2015-03-22 14:59madshiResolutionreopened => fixed

Notes
(0000594)
madshi   
2014-04-11 16:49   
> you said to my you can reproduce this

That was ages ago, wasn't it? I thought that was already fixed? Does this really still occur with 87.9?
(0000595)
huhn   
2014-04-11 16:56   
one more bug in the same scenario

if bilinear is used for downscaling and nearest neighbor is used for upscaling bilinear is used for both. very unusual case... i have to say

there are more bug when a texture unit and a pixel shader scaler is used at the same time but they are to hard to describe for me. i think they all have the same source.
(0000596)
huhn   
2014-04-11 17:03   
> That was ages ago, wasn't it? I thought that was already fixed? Does this really still occur with 87.9?

it is see here:
spline/spline http://abload.de/img/madvrscalenobugwesry.png [^]
bilinear/spline http://abload.de/img/madvrscalebug14sh6.png [^]

and that's why i remind you
(0000604)
madshi   
2014-04-18 19:21   
Should be fixed in v0.87.10. Please double check when the new build is out. If you can still reproduce the bug, please reopen this bug entry.
(0000612)
huhn   
2014-04-23 16:42   
there is still a problem with this resize case:
  up down
bilinear / nearest = both bilinear or at least nearest neighbor is not used
nearest / bilinear = both bilinear or at least nearest neighbor is not used

not that important if you ask me cause nearest neighbor is not a viable choice to me. but i think it's not wrong to inform you about this. and it looks like this was added by fixing this bug at least 50 % off this.

quote:
> if bilinear is used for downscaling and nearest neighbor is used for
> upscaling bilinear is used for both. very unusual case... i have to say


looks like the bilinear up and nearest down bug was added in this version and the nearest up and bilinear down was added when the first case of this error was fixed. looks like you fixed it "only" for 1 case not both back there. of cause i'm not sure about this.

just to make things more clear nearest neighbor is used in combination with all resizer except bilinear!
(0000613)
huhn   
2014-04-23 19:42   
i double checked 87.9 this version has the same error as 87.10 for both cases so this is NOT introduced in 89.10.
(0000808)
madshi   
2015-03-22 14:59   
Yes, you're right, the bilinear/nearest combination doesn't work correctly. But to be honest, fixing this would make things quite a bit more complicated in my code, and I don't think it's important to fix. So I think I'll keep things as they are now, and close this bug report.