Peter Nyboer

I'm starting to jump into shaders (again), and I'm taking a look at the td.repos shader. I'm sorta confused, though, as it seems that the repos matrix really only needs to be 1 plane, as the 2nd plane seems ignored - see adjoining patch. Is that a mistake in the helper patch ("meant to be 1 plane") or a mistake in the shader ("oh, forgot to...")? 

I'd love to figure out how to make an analog of jit.repos, where plane 0 maps x, and plane 1 maps y. As it is, it seems to apply the same distortion to x and y using plane 0...

I'm starting to jump into shaders (again), and I'm taking a look at the td.repos shader.  I'm sorta confused, though, as it seems that the repos matrix really only needs to be 1 plane, as the 2nd plane seems ignored - see adjoining patch.  Is that a mistake in the helper patch ("meant to be 1 plane") or a mistake in the shader ("oh, forgot to...")?  
I'd love to figure out how to make an analog of jit.repos, where plane 0 maps x, and plane 1 maps y. As it is, it seems to apply the same distortion to x and y using plane 0...

repos-shader

the result you are seeing is probably related to how are planes in 

jit.matrix mapped to texture. If you supply repos shader with 4-plane 

(argb) matrix, 'red' and 'green' plane define how much pixels move 

the result you are seeing is probably related to how are planes in  
jit.matrix mapped to texture. If you supply repos shader with 4-plane 
(argb) matrix, 'red' and 'green' plane define how much pixels move  
horizontally and verticaly.
checkout patch pasted below.

Yes, I goofed. By default, a 2-plane matrix get passed to a texture as

lumalpha color mode, which means that inside of a shader these planes

get assigned as (lum,lum,lum,alpha). So, the code for accessing the

vector should be "look.ra" or "look.xw" instead of "look.xy". I hope

that makes sense. The following shader code should work.

distortion based on position map in second inlet

Hi Peter,
Yes, I goofed.  By default, a 2-plane matrix get passed to a texture as
lumalpha color mode, which means that inside of a shader these planes
get assigned as (lum,lum,lum,alpha).  So, the code for accessing the
vector should be "look.ra" or "look.xw" instead of "look.xy".  I hope
that makes sense.  The following shader code should work.

    distortion based on position map in second inlet

> the result you are seeing is probably related to how are planes in

> jit.matrix mapped to texture. If you supply repos shader with 4-plane

> (argb) matrix, 'red' and 'green' plane define how much pixels move

> #P flonum 736 209 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;

> #P newex 613 382 79 196617 pak 0. 0. 0. 0.;

> #P newex 613 405 96 131137545 jit.matrix 4 char 10 10;

> #P flonum 684 363 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;

> #P flonum 591 362 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;

> #P hidden newex 395 57 48 131137545 loadbang;

> #P newex 413 189 100 131137545 pak boundmode 0. 0.;

> #P newex 337 189 75 131137545 pak mode 0. 0.;

> #P flonum 315 163 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;

> #P flonum 280 163 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;

> #P newex 257 189 69 131137545 pak amt 0. 0.;

> #P newex 257 217 76 131137545 prepend param;

> #P message 168 202 67 131137545 getparamlist;

> #P newex 21 145 70 131137545 jit.window fa;

> #P newex 158 425 182 131137545 jit.gl.videoplane fa @scale 1.333 1. 1.;

> #P newex 21 123 121 131137545 jit.gl.render fa @ortho 2;

> #P newex 21 79 58 131137545 t b b erase;

> #P newex 158 382 149 131137545 jit.gl.slab fa @file td.repos.jxs;

> #P flonum 302 55 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;

> #P message 302 75 42 131137545 rate $1;

> #P message 199 76 74 131137545 read dozer.mov;

> #P flonum 62 37 35 9 0.5 0 1 2001 0 0 0 221 221 221 222 222 222 0 0 0;

On 3/28/07, nesa  wrote:
> Hello Peter,
>
> the result you are seeing is probably related to how are planes in
> jit.matrix mapped to texture. If you supply repos shader with 4-plane
> (argb) matrix, 'red' and 'green' plane define how much pixels move
> horizontally and verticaly.
> checkout patch pasted below.
>
> cheers,
> nesa
>
> #P window setfont "Sans Serif" 9.;
> #P flonum 736 209 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;
> #P window linecount 1;
> #P comment 687 211 48 196617 green:y;
> #P comment 597 211 34 196617 red: x;
> #P newex 613 382 79 196617 pak 0. 0. 0. 0.;
> #P window setfont Arial 9.;
> #P newex 613 405 96 131137545 jit.matrix 4 char 10 10;
> #P window setfont "Sans Serif" 9.;
> #P comment 685 348 34 196617 blue;
> #P hidden message 363 112 14 196617 1;
> #P flonum 684 363 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;
> #P flonum 591 362 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;
> #P window setfont Arial 9.;
> #P hidden newex 395 57 48 131137545 loadbang;
> #P toggle 502 164 15 0;
> #P toggle 458 164 15 0;
> #P newex 413 189 100 131137545 pak boundmode 0. 0.;
> #P toggle 401 163 15 0;
> #P toggle 369 163 15 0;
> #P newex 337 189 75 131137545 pak mode 0. 0.;
> #P flonum 315 163 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;
> #P flonum 280 163 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;
> #P newex 257 189 69 131137545 pak amt 0. 0.;
> #P newex 257 217 76 131137545 prepend param;
> #P newex 264 405 32 131137545 print;
> #P message 168 202 67 131137545 getparamlist;
> #P newex 21 145 70 131137545 jit.window fa;
> #P newex 158 425 182 131137545 jit.gl.videoplane fa @scale 1.333 1. 1.;
> #P newex 21 123 121 131137545 jit.gl.render fa @ortho 2;
> #P newex 21 79 58 131137545 t b b erase;
> #P newex 158 382 149 131137545 jit.gl.slab fa @file td.repos.jxs;
> #P message 274 75 27 131137545 stop;
> #P flonum 302 55 35 9 0 0 0 2001 0 0 0 221 221 221 222 222 222 0 0 0;
> #P message 302 75 42 131137545 rate $1;
> #P message 199 76 74 131137545 read dozer.mov;
> #P flonum 62 37 35 9 0.5 0 1 2001 0 0 0 221 221 221 222 222 222 0 0 0;
> #P toggle 21 37 15 0;
> #P newex 21 58 51 131137545 qmetro 2;
> #P newex 158 127 144 131137545 jit.qt.movie 320 240;
> #P window setfont "Sans Serif" 9.;
> #P comment 589 345 34 196617 alpha;
> #P flonum 559 209 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222 0 0 0;
> #P hidden flonum 559 209 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222
> 0 0 0;
> #P noclick;
> #P hidden flonum 736 209 35 9 0. 1. 3 3 0 0 0 221 221 221 222 222 222
> 0 0 0;
> #P noclick;
> #P connect 31 0 35 3;
> #P connect 30 0 35 0;
> #P connect 34 0 12 1;
> #P fasten 38 0 35 2 664 287;
> #P fasten 1 0 35 2 600 253 706 253 664 287;
> #P fasten 2 0 35 1 641 286;
> #P connect 35 0 34 0;
> #P hidden connect 32 0 21 0;
> #P hidden connect 32 0 22 0;
> #P hidden connect 32 0 24 0;
> #P hidden connect 32 0 25 0;
> #P hidden connect 29 0 32 0;
> #P connect 28 0 26 2;
> #P connect 27 0 26 1;
> #P connect 25 0 23 2;
> #P connect 24 0 23 1;
> #P connect 22 0 20 2;
> #P connect 10 0 9 0;
> #P connect 21 0 20 1;
> #P fasten 12 1 18 0 302 404 269 404;
> #P fasten 26 0 19 0 418 211 262 211;
> #P fasten 23 0 19 0 342 211 262 211;
> #P connect 20 0 19 0;
> #P connect 12 0 15 0;
> #P fasten 19 0 12 0 262 240 163 240;
> #P connect 4 0 12 0;
> #P fasten 17 0 12 0 173 223 163 223;
> #P fasten 13 1 4 0 50 117 163 117;
> #P fasten 9 0 4 0 307 117 163 117;
> #P fasten 11 0 4 0 279 117 163 117;
> #P fasten 8 0 4 0 204 117 163 117;
> #P connect 7 0 5 1;
> #P fasten 13 2 14 0 74 105 26 105;
> #P connect 13 0 14 0;
> #P connect 5 0 13 0;
> #P connect 6 0 5 0;
> #P fasten 0 0 35 1 705 253 661 253 699 149 694 149 656 253 599 253
> 641 287;
> #P window clipboard copycount 39;
>
>
>


I should also note that with the given changes, this shader will take

the first and fourth plane of an input texture (as opposed to matrix

input). This might not be preferred in some cases.

I should also note that with the given changes, this shader will take
the first and fourth plane of an input texture (as opposed to matrix
input).  This might not be preferred in some cases.

for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map. Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere". 

That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.

for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map.  Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere".  
That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.


Toy could use a compression technique and unpack the [0-255] data

which is [0, 1] in the shader and instead make it [-1, 1] or some

> for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map. Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere".

> That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.

> 3/4 oz mix of (meyer) lemon and (yellow) grapefruit juice

> Shake with ice and serve in a Pernod-rinsed cocktail glass

Toy could use a compression technique and unpack the [0-255] data
which is [0, 1] in the shader and instead make it [-1, 1] or some
other range.

On 3/30/07, pnyboer  wrote:
>
> for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map.  Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere".
> That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.
> --
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> The Lifespan
> 2 oz brandy
> 1/2 oz orgeat
> 1/2 oz maraschino
> 3/4 oz mix of (meyer) lemon and (yellow) grapefruit juice
> Shake with ice and serve in a Pernod-rinsed cocktail glass
>


One solution to this might be to use the "relative" mode, @param mode 1.

And use the param amt to scale the distortion. Another solution would

be to pass in your matrix as float32 (0.-1.), in which case you won't

get any linear interpolation. This is a limitation of most hardware

One solution to this might be to use the "relative" mode, @param mode 1.
 And use the param amt to scale the distortion.  Another solution would
be to pass in your matrix as float32 (0.-1.), in which case you won't
get any linear interpolation.  This is a limitation of most hardware
when using floating-point textures.

Sorry, misunderstood your previous post. The real answer is to use

float textures although these have issues as well.

> Toy could use a compression technique and unpack the [0-255] data

> which is [0, 1] in the shader and instead make it [-1, 1] or some

> > for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map. Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere".

> > That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.

> > 3/4 oz mix of (meyer) lemon and (yellow) grapefruit juice

> > Shake with ice and serve in a Pernod-rinsed cocktail glass

Sorry, misunderstood your previous post.  The real answer is to use
float textures although these have issues as well.

On 3/30/07, Wesley Smith  wrote:
> Toy could use a compression technique and unpack the [0-255] data
> which is [0, 1] in the shader and instead make it [-1, 1] or some
> other range.
>
> wes
>
> On 3/30/07, pnyboer  wrote:
> >
> > for the sake of the archives, it is also worth noting that becuase the input texture is char 0-255 data, it doesn't seem to be an exact analog of jit.repos, which takes long data as the reposition map.  Absolute maps for anything images with resolution higher than 255x255 won't be able to remap "any pixel anywhere".
> > That's not to say this shader is not totally bitchin', but it does illuminate a limitation, I suppose.
> > --
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > The Lifespan
> > 2 oz brandy
> > 1/2 oz orgeat
> > 1/2 oz maraschino
> > 3/4 oz mix of (meyer) lemon and (yellow) grapefruit juice
> > Shake with ice and serve in a Pernod-rinsed cocktail glass
> >
>


the reason I found this out was the unusual results I was getting trying to use the cartopol example in the jit.repos.help file. I finally realized it was becuause of the need for long types to describe the pixel coordinates...

the reason I found this out was the unusual results I was getting trying to use the cartopol example in the jit.repos.help file.  I finally realized it was becuause of the need for long types to describe the pixel coordinates...


but it is good to know that the floating point textures will be left un-interpolated - I was wondering how to get rid of that!

It probably wouldn't be too hard to put together an upsampling shader

that does 4-point linear interpolation using the fract(xy) and 3 mix()

> but it is good to know that the floating point textures will be left un-interpolated - I was wondering how to get rid of that!

It probably wouldn't be too hard to put together an upsampling shader
that does 4-point linear interpolation using the fract(xy) and 3 mix()
operations.

pnyboer wrote:
> but it is good to know that the floating point textures will be left un-interpolated - I was wondering how to get rid of that!
> --
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> The Lifespan
> 2 oz brandy
> 1/2 oz orgeat
> 1/2 oz maraschino
> 3/4 oz mix of (meyer) lemon and (yellow) grapefruit juice
> Shake with ice and serve in a Pernod-rinsed cocktail glass
