From a70ad8df8fc2a3da63fc91dd18bbfd55be7a89dd Mon Sep 17 00:00:00 2001 From: Willem Jan Palenstijn Date: Thu, 12 Mar 2015 12:30:47 +0100 Subject: Add outputScale argument to 3D CUDA BP --- cuda/3d/par3d_bp.cu | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-) (limited to 'cuda/3d/par3d_bp.cu') diff --git a/cuda/3d/par3d_bp.cu b/cuda/3d/par3d_bp.cu index 0c33280..1217949 100644 --- a/cuda/3d/par3d_bp.cu +++ b/cuda/3d/par3d_bp.cu @@ -77,7 +77,7 @@ static bool bindProjDataTexture(const cudaArray* array) } -__global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims) +__global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, float fOutputScale) { float* volData = (float*)D_volData; @@ -139,11 +139,11 @@ __global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAn 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; } // supersampling version -__global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims) +__global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, float fOutputScale) { float* volData = (float*)D_volData; @@ -180,6 +180,9 @@ __global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int star const float fSubStep = 1.0f/dims.iRaysPerVoxelDim; + fOutputScale /= (dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim); + + for (int Z = startZ; Z < endZ; ++Z, fZ += 1.0f) { @@ -217,14 +220,15 @@ __global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int star } - volData[(Z*dims.iVolY+Y)*volPitch+X] += fVal / (dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim*dims.iRaysPerVoxelDim); + volData[(Z*dims.iVolY+Y)*volPitch+X] += fVal * fOutputScale; } } bool Par3DBP_Array(cudaPitchedPtr D_volumeData, cudaArray *D_projArray, - const SDimensions3D& dims, const SPar3DProjection* angles) + const SDimensions3D& dims, const SPar3DProjection* angles, + float fOutputScale) { bindProjDataTexture(D_projArray); @@ -271,9 +275,9 @@ bool Par3DBP_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_par3D_BP<<>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims); + dev_par3D_BP<<>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale); else - dev_par3D_BP_SS<<>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims); + dev_par3D_BP_SS<<>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale); } cudaTextForceKernelsCompletion(); @@ -288,14 +292,15 @@ bool Par3DBP_Array(cudaPitchedPtr D_volumeData, bool Par3DBP(cudaPitchedPtr D_volumeData, cudaPitchedPtr D_projData, - const SDimensions3D& dims, const SPar3DProjection* angles) + const SDimensions3D& dims, const SPar3DProjection* angles, + float fOutputScale) { // transfer projections to array cudaArray* cuArray = allocateProjectionArray(dims); transferProjectionsToArray(D_projData, cuArray, dims); - bool ret = Par3DBP_Array(D_volumeData, cuArray, dims, angles); + bool ret = Par3DBP_Array(D_volumeData, cuArray, dims, angles, fOutputScale); cudaFreeArray(cuArray); -- cgit v1.2.3