Add outputScale argument to 3D CUDA BP

1 files changed, 15 insertions, 9 deletions

diff --git a/cuda/3d/cone_bp.cu b/cuda/3d/cone_bp.cu
index 5648d6f..5e67980 100644
--- a/cuda/3d/cone_bp.cu
+++ b/cuda/3d/cone_bp.cu
@@ -78,7 +78,8 @@ bool bindProjDataTexture(const cudaArray* array)
 
 //__launch_bounds__(32*16, 4)
 __global__ void dev_cone_BP(void* D_volData, unsigned int volPitch, int startAngle,
-                            int angleOffset, const astraCUDA3d::SDimensions3D dims)
+                            int angleOffset, const astraCUDA3d::SDimensions3D dims,
+                            float fOutputScale)
 {
 	float* volData = (float*)D_volData;
 
@@ -147,13 +148,13 @@ __global__ void dev_cone_BP(void* D_volData, unsigned int volPitch, int startAng
 		endZ = dims.iVolZ - startZ;
 
 	for(int i=0; i < endZ; i++)
-		volData[((startZ+i)*dims.iVolY+Y)*volPitch+X] += Z[i];
+		volData[((startZ+i)*dims.iVolY+Y)*volPitch+X] += Z[i] * fOutputScale;
 } //End kernel
 
 
 
 // supersampling version
-__global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims)
+__global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, float fOutputScale)
 {
 	float* volData = (float*)D_volData;
 
@@ -189,6 +190,9 @@ __global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int start
 	float fZ = startZ - 0.5f*dims.iVolZ + 0.5f - 0.5f + 0.5f/dims.iRaysPerVoxelDim;
 	const float fSubStep = 1.0f/dims.iRaysPerVoxelDim;
 
+	fOutputScale /= (dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim);
+
+
 	for (int Z = startZ; Z < endZ; ++Z, fZ += 1.0f)
 	{
 
@@ -236,14 +240,15 @@ __global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int start
 
 		}
 
-		volData[(Z*dims.iVolY+Y)*volPitch+X] += fVal / (dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim);
+		volData[(Z*dims.iVolY+Y)*volPitch+X] += fVal * fOutputScale;
 	}
 }
 
 
 bool ConeBP_Array(cudaPitchedPtr D_volumeData,
                   cudaArray *D_projArray,
-                  const SDimensions3D& dims, const SConeProjection* angles)
+                  const SDimensions3D& dims, const SConeProjection* angles,
+                  float fOutputScale)
 {
 	bindProjDataTexture(D_projArray);
 
@@ -291,9 +296,9 @@ bool ConeBP_Array(cudaPitchedPtr D_volumeData,
 		for (unsigned int i = 0; i < angleCount; i += g_anglesPerBlock) {
 		// printf("Calling BP: %d, %dx%d, %dx%d to %p\n", i, dimBlock.x, dimBlock.y, dimGrid.x, dimGrid.y, (void*)D_volumeData.ptr); 
 			if (dims.iRaysPerVoxelDim == 1)
-				dev_cone_BP<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims);
+				dev_cone_BP<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
 			else
-				dev_cone_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims);
+				dev_cone_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
 		}
 
 		cudaTextForceKernelsCompletion();
@@ -309,14 +314,15 @@ bool ConeBP_Array(cudaPitchedPtr D_volumeData,
 
 bool ConeBP(cudaPitchedPtr D_volumeData,
             cudaPitchedPtr D_projData,
-            const SDimensions3D& dims, const SConeProjection* angles)
+            const SDimensions3D& dims, const SConeProjection* angles,
+            float fOutputScale)
 {
 	// transfer projections to array
 
 	cudaArray* cuArray = allocateProjectionArray(dims);
 	transferProjectionsToArray(D_projData, cuArray, dims);
 
-	bool ret = ConeBP_Array(D_volumeData, cuArray, dims, angles);
+	bool ret = ConeBP_Array(D_volumeData, cuArray, dims, angles, fOutputScale);
 
 	cudaFreeArray(cuArray);