Hack faster colour distance

applications/reconstruct/src/depth_camera.cu

@@ -38,14 +38,19 @@ extern __device__ float spatialWeighting(float r);
  * Real-time foreground-background segmentation using codebook model.
  * Real-Time Imaging. https://doi.org/10.1016/j.rti.2004.12.004
  */
- __device__ float colordiffFloat(const uchar4 &pa, const uchar4 &pb) {
-	float x_2 = pb.x * pb.x + pb.y * pb.y + pb.z * pb.z;
-	float v_2 = pa.x * pa.x + pa.y * pa.y + pa.z * pa.z;
-	float xv_2 = pow(pb.x * pa.x + pb.y * pa.y + pb.z * pa.z, 2);
-	float p_2 = xv_2 / v_2;
+ __device__ inline float colordiffFloat(const uchar4 &pa, const uchar4 &pb) {
+	const float x_2 = pb.x * pb.x + pb.y * pb.y + pb.z * pb.z;
+	const float v_2 = pa.x * pa.x + pa.y * pa.y + pa.z * pa.z;
+	const float xv_2 = pow(pb.x * pa.x + pb.y * pa.y + pb.z * pa.z, 2);
+	const float p_2 = xv_2 / v_2;
 	return sqrt(x_2 - p_2);
 }
 
+__device__ float colordiffFloat2(const uchar4 &pa, const uchar4 &pb) {
+	float3 delta = make_float3((float)pa.x - (float)pb.x, (float)pa.y - (float)pb.y, (float)pa.z - (float)pb.z);
+	return length(delta);
+}
+
 /*
  * Colour weighting as suggested in:
  * C. Kuster et al. Spatio-Temporal Geometry Fusion for Multiple Hybrid Cameras using Moving Least Squares Surfaces. 2014.
@@ -98,7 +103,7 @@ __global__ void mls_smooth_kernel(ftl::cuda::TextureObject<float> output, HashDa
 							uchar4 c2 = tex2D<uchar4>(camera.colour, screenPos.x+u, screenPos.y+v);
 							//float4 normal = tex2D<float4>(camera.normal, screenPos.x+u, screenPos.y+v);
 							const float3 worldPos = camera.pose * camera.params.kinectDepthToSkeleton(screenPos.x+u, screenPos.y+v, depth);
-							const float weight = spatialWeighting(length(mPos - worldPos))*colourWeighting(colordiffFloat(c1,c2));
+							const float weight = spatialWeighting(length(mPos - worldPos))*colourWeighting(colordiffFloat2(c1,c2));
 
 							wpos += weight*worldPos;
 							//wnorm += weight*make_float3(normal);