Can anyone say why this basic PB crashes AE?

Question

Hi,
I wrote a basic light wrap filter - it works in PB but crashes AE immidiately when I apply the filter to a layer.
<languageVersion : 1.0;>
kernel efx_lightwrap
<   namespace : "efx LightWrap";
    vendor : "EFX Filip Kaczorek";
    version : 1;
    description : "LightWrap";
    displayname : "EFX LightWrap";
    category : "EFX";
>
    dependent float maxd;
    dependent float fgam;
    input image4 src;
    output pixel4 dst;
   parameter float thr
        <
            minValue: 0.0;
            maxValue: 0.99;
            defaultValue: 0.99;
            aeDisplayName: "alpha threshold";
        >;
   parameter int rad
        <
            minValue: 0;
            maxValue: 20;
            defaultValue: 1;
            aeDisplayName: "radius";
        >;
   parameter float gam
        <
            minValue: 0.0;
            maxValue: 1.0;
            defaultValue: 1.0;
            aeDisplayName: "falloff shape";
        >;
   parameter float ammount
        <
            minValue: 0.0;
            maxValue: 1.0;
            defaultValue: 1.0;
            aeDisplayName: "ammount";
        >;
   parameter pixel3 lightcol
    <
        minValue: pixel3(0.0,0.0,0.0);
        maxValue: pixel3(1.0,1.0,1.0);
        defaultValue: pixel3(0.0,0.0,1.0);
        description: "Light color.";
        parameterType: "colorRGB";
        aeDisplayName: "Light color";
        aeUIControl: "aeColor";
    >;
  void evaluateDependents()
       maxd=distance( float2(0.0,0.0),float2(float(rad),float(rad)) );
        fgam=8.0*gam-7.0;
   pixel4 check(pixel4 this_p){
        float dist=9999.0;
        this_p.a=0.0;
        pixel4 p=pixel4(0.0);
        float td=0.0;
        int i=0;
        int j=0;
        for(i = -rad; i <= rad; i++)
                  for(j = -rad; j <= rad; j++)
                      p=sampleNearest(src,outCoord() + float2(float(i),float(j)));
                      if(p.a<thr){
                             td=distance( float2(0.0,0.0),float2(float(i),float(j)) );
                             if(dist>td){dist=td;}
        if(dist<9999.0){
            this_p.a=1.0-(dist/maxd);
            this_p.a=pow(this_p.a,(1.0-fgam));
            this_p.rgb=lightcol;
        return this_p;
    void
    evaluatePixel()
        pixel4 s_pix=sampleNearest(src,outCoord());
        pixel4 t_pix=s_pix;
        pixel4 lw_pix=s_pix;
        if(s_pix.a>thr){
               lw_pix=check(lw_pix);  
        if(lw_pix.a>0.0){
             t_pix.rgb=s_pix.rgb+(lw_pix.rgb*pixel3(lw_pix.a*ammount));
        dst = t_pix;
After some experimenting I think it has something to do with sampling the input image inside the check() function.
If I replace the sampled values with constant ones - it doesn't crash.
If I move all the calculations up to the main evaluatePixel() function - it doesn't crash.
I'd call it a day with that conclusion if it wasn't for the fact that this is basically my template PB and I have many PB filters built exactly the same way that don't crash.
Including a Gaussian Filter that does way more sampling within its sub-function.
So any ideas?

Answer 1

In Pixel Bender it runs in both modes.
It crashes on 3 different machines, although all HP
Two different workstations and a laptop (so xeons and i7, FireGLv5600, Quadro 4800, GF230 mobile).
All Win7 x64 and AE CS5.
Like I said, it crashes AE even if I strip the code to only one line in the function: p=sampleNearest...
This seems to be the only thing that gives problems.
Weird thing, THIS IS almost my template filter - I start all my filters looking like that and that one is THE ONLY ONE crashing???
The solution i found was to move the code into the main body - not use a subfunction. Now it works great, though the esthetics of the code suffer