Text archives Help

---

Thiago Ize wrote:

Lower down there is this:
 if(last != -1){
   shadowRays.resize ( first, last + 1);
which I think should handle the problem since shadowRays will start with a valid ray or (partially) valid sse group. Speaking of which, I think

This makes sense.

it might be more efficient the way it was before where first pointed to the start of the 4-wide simd group, even though the first ray (or 3) of those were invalid. The reason for that is that operating on sse registers of 4 rays with just one valid ray should in principle be just as fast as operating on a single valid scalar ray. If that simd group contains more than one valid ray then it's even more likely to be faster.

I think you missed the point.  first is supposed to point to the first valid 
ray.  Take this as an example:

x = outside of [begin,end)
1 = invalid ray
0 = valid ray

0123-4567-8901
xx11-1010-0xxx

The first block will be done in a non sse loop, and first will not be set.
The second block will be an sse loop.  In your original code first will be 
set to 4.
The third block will be non sse loop.  Here we will copy the data from 
rays[4] instead of the first valid ray (rays[5]).

James

Thiago

James Bigler wrote:

Please help clarify something for me with the recent changes to the shadow algorithm.

      for(;i<b;i++){
        Vector dir = shadowRays.getDirection(i);
        if(Dot(dir, sourceRays.getFFNormal(i)) > 0) {

// Block 1: When we have a good ray.  Pretty straight forward.

          shadowRays.setOrigin(i, sourceRays.getHitPosition(i));
          shadowRays.setTime(i, sourceRays.getTime(i));
          // See comment above.
          shadowRays.setHitMaterial(i, NULL);
          last = i;
          if (first < 0)
            first = i;
        }
        else if (first >= 0) {

// Block 2: We have a bad ray and we have found a good ray in the packet somewhere previously.

          //if we've already found a valid ray, use that to copy valid
          //data into the invalid ray.
          shadowRays.setOrigin(i, shadowRays.getOrigin(first));
          shadowRays.setDirection(i, shadowRays.getDirection(first));
          shadowRays.setTime(i, shadowRays.getTime(first));

          shadowRays.maskRay(i); //this sets minT -MAXT and hitMatl to -1
          //We set minT to 0 instead of -MAXT just in case an algorithm
          //uses the ray hit position to compute things. In which case
          //having the hit position at the origin would probably break
          //things less than having it be infinitely far away in the
          //negative direction.
shadowRays.overrideMinT(i, 0);//TOOD: verify 0 is ok (should we use -eps?).
        } else {

// Block 3: We have a bad ray, and we have *not* found a good ray previously.

          shadowRays.maskRay(i);
        }
      }

So we could end up in a situation where some of the bad rays have good data (block 2) and some have nothing (block 3). Block 3 rays would be the first block of bad rays. Is this a problem?

James

---

• [Manta] Shadow algorithm question, James Bigler, 02/07/2008
  • Re: [Manta] Shadow algorithm question, Thiago Ize, 02/07/2008
    • Re: [Manta] Shadow algorithm question, James Bigler, 02/07/2008

---

Archive powered by MHonArc 2.6.16.