cessen/psychopath
1
0
Fork 0
Code Issues 1 Pull Requests Projects Releases Wiki Activity

Rearrnging the operations in ray/bbox test for slight perf improvement.

Browse Source 
The difference is extremely small, but consistent.  I'm guessing
due to better pipelining of instructions.
This commit is contained in:
Nathan Vegdahl 2017-07-06 22:10:37 -07:00
parent 09daf617ef
commit 649a6a0869
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/bbox.rs
View File

@ -44,15 +44,15 @@ impl BBox {
let t2 = (self.max.co - ray.orig.co) * ray.dir_inv.co;
// Find the far and near intersection
let mut near_t = t1.v_min(t2);
let mut far_t = t1.v_max(t2);
near_t.set_3(0.0);
let mut near_t = t1.v_min(t2);
far_t.set_3(std::f32::INFINITY);
let hitt0 = near_t.h_max();
let hitt1 = far_t.h_min() * BBOX_MAXT_ADJUST;
near_t.set_3(0.0);
let far_hit_t = fast_minf32(far_t.h_min() * BBOX_MAXT_ADJUST, ray.max_t);
let near_hit_t = near_t.h_max();
// Did we hit?
return hitt0 <= fast_minf32(hitt1, ray.max_t);
return near_hit_t <= far_hit_t;
}
// Creates a new BBox transformed into a different space.