On 20/05/2011 17:11, Stephen Newbold wrote:
So, found the issue. As part of my pixel engine I'm comparing
distanceSquared between 2 points to obtain the minimum distance from a
set of points. I'm then only finding the square root of the smallest in
order to try and speed things up. I thought this was a pretty standard
thing? But for some reason this is where my artifacts are coming from.
If I compare actual distances, everything works correctly. I'm not
seeing a noticeable performance hit from doing this as my point set is
small.
Possibly rounding errors. Whenever comparing floating point values you
can't rely on equality but need to test for equivalence by including a
small threshold (often called epsilon), so:
if (a - b < 1e-5)
match
else
no match
This is why it took me ages to find the problem, I just assumed this
code would be fine as it was so simple.
Still have no idea why this glitches, or why updating _validate() (when
changing any knob value) outputs different artifacts, but I'm glad its
sorted. Thanks for all the suggestions.
Steve
Stephen Newbold wrote:
Hi Colin,
Yeah random speckles. When I get back to work tomorrow I'll do as you
suggest. Can't really share code as it's a work tool, but thanks for
all the advice and I'll strip it right back until I find the culprit.
Cheers,
Steve
On 19 May 2011, at 20:23, "Colin Doncaster" <co...@peregrinelabs.com
<mailto:co...@peregrinelabs.com>> wrote:
Hi Steve,
Just to be certain we're talking about the same thing, is it safe to
assume that the artifacts you're seeing are noise like speckles that
change frame to frame/on refresh vs some equally random though more
structured issue?
It still sounds like it's an issue with the output row not being set
to 0 - can you isolate it to one or more channels, ie. just in red,
or just red and green, etc.? You're not creating any new rows
yourself that haven't been set to 0?
I would say comment everything out except for your memset, and then
slowly add them back in until you see the artifacts. It's really hard
to help more without seeing code. :)
*/
*/
*/--
Colin Doncaster/*
*/
*/
*/Peregrine Labs/*
/*
/*
*/
*/
www.peregrinelabs.com <http://www.peregrinelabs.com>
/*
/*
/*
/*
On 2011-05-19, at 1:23 PM, Stephen Newbold wrote:
I might be wrong when I said the artifacts were consistent,
artifacts do actually seem to be random, which makes more sense but
is still annoying.
Steve
Stephen Newbold wrote:
Hi Colin,
Of course, rookie error! That has stopped the crashing but hasn't
stopped the artifacts. Does this mean I can rule out issues with
the output buffer? I think all I can do is get the pixel output
down to a manageable size so I can check exactly what values are
getting output and any variables associated with them. Sounds like
a fun evening!
Steve
Colin Doncaster wrote:
It sounds like you need to zero out the row ( nuke doesn't do this
for you ) prior using it - thus you're memset is heading down the
right path you're call is wrong,
memset( buffer +x, 0, (r-x) * sizeof(float));
you have to remember that although the pointer is pointing to [0]
that may not be safe if Nuke's only managing x to r.
This should hopefully avoid the crash.
cheers
*/
*/
*/--
Colin Doncaster/*
*/
*/
*/Peregrine Labs/*
/*
/*
*/
*/
www.peregrinelabs.com <http://www.peregrinelabs.com>
/*
/*
/*
/*
On 2011-05-19, at 12:28 PM, Stephen Newbold wrote:
Hi,
I'm wondering if you guys can give me some pointers in where I
might look to iron out a glitch in my DrawIop. So everything is
generally working as I want it, but I'm getting very small
glitches in the output. Could be in my code but I'm noticing the
glitches change when I use any knob, even those unconnected with
the output which is a bit odd. I've even added a null knob that
does absolutely nothing just to test this. Change this knob and
my artifacts change.
So, what I'm trying to do first is work out why my output (albeit
with artifacts) is changing when I move unrelated controls. I'm
assuming certain things get updated on any knob change? I know it
hard without seeing code, but have you guys got any ideas what is
called/updated each time a knob is changed? If it helps, things
don't change when changing to a different frame, only when using
a knob. Also, the change in artifacts will change consistently,
in other words if I move a slider back and forth, the result will
change consistently back and forth, same artifact pattern for
each value.
I've also tried emptying the output buffer with memset for each
row before filling it again to see if this affected things, but
this is crashing nuke,
memset(buffer, 0, (R - X) * sizeof(float));
This has kind of got me stumped.
Steve
--
Stephen Newbold
Compositing Lead - Film
MPC
127 Wardour Street
Soho, London, W1F 0NL
Main - + 44 (0) 20 7434 3100
www.moving-picture.com <http://www.moving-picture.com>
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk
<mailto:Nuke-dev@support.thefoundry.co.uk>,
http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
------------------------------------------------------------------------
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk
<mailto:Nuke-dev@support.thefoundry.co.uk>,http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
--
Stephen Newbold
Compositing Lead - Film
MPC
127 Wardour Street
Soho, London, W1F 0NL
Main - + 44 (0) 20 7434 3100
www.moving-picture.com <http://www.moving-picture.com>
------------------------------------------------------------------------
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk
<mailto:Nuke-dev@support.thefoundry.co.uk>,http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
--
Stephen Newbold
Compositing Lead - Film
MPC
127 Wardour Street
Soho, London, W1F 0NL
Main - + 44 (0) 20 7434 3100
www.moving-picture.com <http://www.moving-picture.com>
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk
<mailto:Nuke-dev@support.thefoundry.co.uk>,
http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk
<mailto:Nuke-dev@support.thefoundry.co.uk>,
http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
------------------------------------------------------------------------
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk,http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
--
Stephen Newbold
Compositing Lead - Film
MPC
127 Wardour Street
Soho, London, W1F 0NL
Main - + 44 (0) 20 7434 3100
www.moving-picture.com
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk, http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
--
Wouter Klouwen, Software Engineer
The Foundry, 6th Floor, The Communications Building,
48 Leicester Square, London, WC2H 7LT, UK
T: +442079686828 - F: +442074341550 - thefoundry.co.uk
The Foundry Visionmongers Ltd - Reg.d in England and Wales No: 4642027
_______________________________________________
Nuke-dev mailing list
Nuke-dev@support.thefoundry.co.uk, http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
