Add global astra_set_gpu_index function

32 files changed, 175 insertions, 114 deletions

diff --git a/build/linux/Makefile.in b/build/linux/Makefile.in
index be50cbf..3e8bca1 100644
--- a/build/linux/Makefile.in
+++ b/build/linux/Makefile.in
@@ -57,6 +57,11 @@ MEXLDFLAGS='$$LDFLAGS $(LDFLAGS)'
 MEXLIBS   ="-L.libs -lut"
 MEXSUFFIX = @MEXSUFFIX@
 MEX       = @MEX@
+
+ifeq ($(cuda),yes)
+MEXFLAGS += -DASTRA_CUDA
+endif
+
 endif
 
 LIBDIR=/usr/local/lib
diff --git a/cuda/2d/algo.cu b/cuda/2d/algo.cu
index 5ae5d08..f04607f 100644
--- a/cuda/2d/algo.cu
+++ b/cuda/2d/algo.cu
@@ -99,12 +99,14 @@ void ReconAlgo::reset()
 
 bool ReconAlgo::setGPUIndex(int iGPUIndex)
 {
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	return true;
 }
diff --git a/cuda/2d/astra.cu b/cuda/2d/astra.cu
index 5d53da0..2240629 100644
--- a/cuda/2d/astra.cu
+++ b/cuda/2d/astra.cu
@@ -195,13 +195,15 @@ bool AstraFBP::init(int iGPUIndex)
 		return false;
 	}
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-	{
-		return false;
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+		{
+			return false;
+		}
 	}
 
 	bool ok = allocateVolume(pData->D_volumeData, pData->dims.iVolWidth+2, pData->dims.iVolHeight+2, pData->volumePitch);
@@ -563,13 +565,14 @@ bool astraCudaFP(const float* pfVolume, float* pfSinogram,
 	dims.iVolWidth = iVolWidth;
 	dims.iVolHeight = iVolHeight;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	bool ok;
 
@@ -649,13 +652,14 @@ bool astraCudaFanFP(const float* pfVolume, float* pfSinogram,
 	dims.iVolWidth = iVolWidth;
 	dims.iVolHeight = iVolHeight;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	bool ok;
 
@@ -759,13 +763,14 @@ bool astraCudaFanFP(const float* pfVolume, float* pfSinogram,
 	dims.iVolWidth = iVolWidth;
 	dims.iVolHeight = iVolHeight;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	bool ok;
 
diff --git a/cuda/2d/darthelper.cu b/cuda/2d/darthelper.cu
index db0036e..28b77cc 100644
--- a/cuda/2d/darthelper.cu
+++ b/cuda/2d/darthelper.cu
@@ -344,12 +344,14 @@ void dartSmoothing(float* out, const float* in, float b, unsigned int radius, un
 
 bool setGPUIndex(int iGPUIndex)
 {
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	return true;
 }
diff --git a/cuda/2d/darthelper.h b/cuda/2d/darthelper.h
index e05f01e..3694b96 100644
--- a/cuda/2d/darthelper.h
+++ b/cuda/2d/darthelper.h
@@ -29,8 +29,6 @@ $Id$
 #ifndef _CUDA_ARITH2_H
 #define _CUDA_ARITH2_H
 
-#include <cuda.h>
-
 namespace astraCUDA {
 
 	void roiSelect(float* out, float radius, unsigned int width, unsigned int height);
diff --git a/cuda/3d/astra3d.cu b/cuda/3d/astra3d.cu
index fd4b370..4447775 100644
--- a/cuda/3d/astra3d.cu
+++ b/cuda/3d/astra3d.cu
@@ -381,12 +381,14 @@ bool AstraSIRT3d::enableSinogramMask()
 	
 bool AstraSIRT3d::setGPUIndex(int index)
 {
-	cudaSetDevice(index);
-	cudaError_t err = cudaGetLastError();
+	if (index != -1) {
+		cudaSetDevice(index);
+		cudaError_t err = cudaGetLastError();
 
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	return true;
 }
@@ -858,12 +860,14 @@ bool AstraCGLS3d::enableSinogramMask()
 	
 bool AstraCGLS3d::setGPUIndex(int index)
 {
-	cudaSetDevice(index);
-	cudaError_t err = cudaGetLastError();
+	if (index != -1) {
+		cudaSetDevice(index);
+		cudaError_t err = cudaGetLastError();
 
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	return true;
 }
@@ -1156,13 +1160,14 @@ bool astraCudaConeFP(const float* pfVolume, float* pfProjections,
 	if (iDetectorSuperSampling == 0)
 		return false;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	cudaPitchedPtr D_volumeData = allocateVolumeData(dims);
 	bool ok = D_volumeData.ptr;
@@ -1264,13 +1269,14 @@ bool astraCudaPar3DFP(const float* pfVolume, float* pfProjections,
 	if (iDetectorSuperSampling == 0)
 		return false;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 
 	cudaPitchedPtr D_volumeData = allocateVolumeData(dims);
@@ -1382,13 +1388,14 @@ bool astraCudaConeBP(float* pfVolume, const float* pfProjections,
 	if (iProjAngles == 0 || iProjU == 0 || iProjV == 0 || pfAngles == 0)
 		return false;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 	cudaPitchedPtr D_volumeData = allocateVolumeData(dims);
 	bool ok = D_volumeData.ptr;
@@ -1487,13 +1494,14 @@ bool astraCudaPar3DBP(float* pfVolume, const float* pfProjections,
 	if (iProjAngles == 0 || iProjU == 0 || iProjV == 0 || pfAngles == 0)
 		return false;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 
 	cudaPitchedPtr D_volumeData = allocateVolumeData(dims);
@@ -1568,13 +1576,14 @@ bool astraCudaFDK(float* pfVolume, const float* pfProjections,
 	if (iVoxelSuperSampling == 0)
 		return false;
 
-	cudaSetDevice(iGPUIndex);
-	cudaError_t err = cudaGetLastError();
-
-	// Ignore errors caused by calling cudaSetDevice multiple times
-	if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-		return false;
+	if (iGPUIndex != -1) {
+		cudaSetDevice(iGPUIndex);
+		cudaError_t err = cudaGetLastError();
 
+		// Ignore errors caused by calling cudaSetDevice multiple times
+		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+			return false;
+	}
 
 
 	cudaPitchedPtr D_volumeData = allocateVolumeData(dims);
diff --git a/cuda/3d/darthelper3d.cu b/cuda/3d/darthelper3d.cu
index 68330a1..6e60ec2 100644
--- a/cuda/3d/darthelper3d.cu
+++ b/cuda/3d/darthelper3d.cu
@@ -215,12 +215,14 @@ namespace astraCUDA3d {
 
 	bool setGPUIndex(int iGPUIndex)
 	{
-		cudaSetDevice(iGPUIndex);
-		cudaError_t err = cudaGetLastError();
+		if (iGPUIndex != -1) {
+			cudaSetDevice(iGPUIndex);
+			cudaError_t err = cudaGetLastError();
 
-		// Ignore errors caused by calling cudaSetDevice multiple times
-		if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
-			return false;
+			// Ignore errors caused by calling cudaSetDevice multiple times
+			if (err != cudaSuccess && err != cudaErrorSetOnActiveProcess)
+				return false;
+		}
 
 		return true;
 	}
diff --git a/include/astra/CudaBackProjectionAlgorithm.h b/include/astra/CudaBackProjectionAlgorithm.h
index 965c734..09a2a65 100644
--- a/include/astra/CudaBackProjectionAlgorithm.h
+++ b/include/astra/CudaBackProjectionAlgorithm.h
@@ -91,7 +91,7 @@ public:
 	bool initialize(CProjector2D* _pProjector,
 	                CFloat32ProjectionData2D* _pSinogram, 
 					CFloat32VolumeData2D* _pReconstruction,
-					int _iGPUindex = 0, int _iPixelSuperSampling = 1);
+					int _iGPUindex = -1, int _iPixelSuperSampling = 1);
 
 	/** Get a description of the class.
 	 *
diff --git a/include/astra/CudaCglsAlgorithm.h b/include/astra/CudaCglsAlgorithm.h
index 7734d6e..a529bfd 100644
--- a/include/astra/CudaCglsAlgorithm.h
+++ b/include/astra/CudaCglsAlgorithm.h
@@ -101,7 +101,7 @@ public:
 	bool initialize(CProjector2D* _pProjector, 
 					CFloat32ProjectionData2D* _pSinogram, 
 					CFloat32VolumeData2D* _pReconstruction,
-					int _iGPUindex = 0, int _iDetectorSuperSampling = 1,
+					int _iGPUindex = -1, int _iDetectorSuperSampling = 1,
 					int _iPixelSuperSampling = 1);
 
 	/** Get a description of the class.
diff --git a/include/astra/CudaEMAlgorithm.h b/include/astra/CudaEMAlgorithm.h
index ff22dd4..ca9c9f6 100644
--- a/include/astra/CudaEMAlgorithm.h
+++ b/include/astra/CudaEMAlgorithm.h
@@ -72,7 +72,7 @@ public:
 	bool initialize(CProjector2D* _pProjector,
 	                CFloat32ProjectionData2D* _pSinogram, 
 					CFloat32VolumeData2D* _pReconstruction,
-	                int _iGPUindex = 0, int _iDetectorSuperSampling = 1,
+	                int _iGPUindex = -1, int _iDetectorSuperSampling = 1,
 	                int _iPixelSuperSampling = 1);
 
 	/** Get a description of the class.
diff --git a/include/astra/CudaFilteredBackProjectionAlgorithm.h b/include/astra/CudaFilteredBackProjectionAlgorithm.h
index 4b7b904..c496c8f 100644
--- a/include/astra/CudaFilteredBackProjectionAlgorithm.h
+++ b/include/astra/CudaFilteredBackProjectionAlgorithm.h
@@ -61,7 +61,7 @@ public:
 	virtual ~CCudaFilteredBackProjectionAlgorithm();
 
 	virtual bool initialize(const Config& _cfg);
-	bool initialize(CFloat32ProjectionData2D * _pSinogram, CFloat32VolumeData2D * _pReconstruction, E_FBPFILTER _eFilter, const float * _pfFilter = NULL, int _iFilterWidth = 0, int _iGPUIndex = 0, float _fFilterParameter = -1.0f);
+	bool initialize(CFloat32ProjectionData2D * _pSinogram, CFloat32VolumeData2D * _pReconstruction, E_FBPFILTER _eFilter, const float * _pfFilter = NULL, int _iFilterWidth = 0, int _iGPUIndex = -1, float _fFilterParameter = -1.0f);
 
 	virtual void run(int _iNrIterations = 0);
 
diff --git a/include/astra/CudaForwardProjectionAlgorithm.h b/include/astra/CudaForwardProjectionAlgorithm.h
index 53b6c8e..6a6ba24 100644
--- a/include/astra/CudaForwardProjectionAlgorithm.h
+++ b/include/astra/CudaForwardProjectionAlgorithm.h
@@ -103,7 +103,7 @@ public:
 					CVolumeGeometry2D* _pVolumeGeometry, 
 					CFloat32VolumeData2D* _pVolume, 
 					CFloat32ProjectionData2D* _pSinogram,
-					int _iGPUindex = 0, int _iDetectorSuperSampling = 1);
+					int _iGPUindex = -1, int _iDetectorSuperSampling = 1);
 
 	/** Get all information parameters
 	 *
diff --git a/include/astra/CudaForwardProjectionAlgorithm3D.h b/include/astra/CudaForwardProjectionAlgorithm3D.h
index 72c6a00..3312eec 100644
--- a/include/astra/CudaForwardProjectionAlgorithm3D.h
+++ b/include/astra/CudaForwardProjectionAlgorithm3D.h
@@ -74,7 +74,7 @@ public:
 	bool initialize(CProjector3D* _pProjector, 
 					CFloat32ProjectionData3DMemory* _pSinogram, 
 					CFloat32VolumeData3DMemory* _pReconstruction,
-					int _iGPUindex = 0, int _iDetectorSuperSampling = 1);
+					int _iGPUindex = -1, int _iDetectorSuperSampling = 1);
 
 
 	/** Get all information parameters
diff --git a/include/astra/CudaReconstructionAlgorithm2D.h b/include/astra/CudaReconstructionAlgorithm2D.h
index 88fc344..68f4b5d 100644
--- a/include/astra/CudaReconstructionAlgorithm2D.h
+++ b/include/astra/CudaReconstructionAlgorithm2D.h
@@ -90,7 +90,7 @@ public:
 	virtual bool initialize(CProjector2D* _pProjector, 
 	                        CFloat32ProjectionData2D* _pSinogram, 
 	                        CFloat32VolumeData2D* _pReconstruction,
-	                        int _iGPUindex = 0, int _iDetectorSuperSampling = 1,
+	                        int _iGPUindex = -1, int _iDetectorSuperSampling = 1,
 	                        int _iPixelSuperSampling = 1);
 
 
diff --git a/include/astra/CudaSartAlgorithm.h b/include/astra/CudaSartAlgorithm.h
index 319e1e9..49845f2 100644
--- a/include/astra/CudaSartAlgorithm.h
+++ b/include/astra/CudaSartAlgorithm.h
@@ -93,7 +93,7 @@ public:
 	bool initialize(CProjector2D* _pProjector,
 	                CFloat32ProjectionData2D* _pSinogram, 
 					CFloat32VolumeData2D* _pReconstruction,
-					int _iGPUindex = 0, int _iDetectorSuperSampling = 1);
+					int _iGPUindex = -1, int _iDetectorSuperSampling = 1);
 
 	/** Get a description of the class.
 	 *
diff --git a/include/astra/CudaSirtAlgorithm.h b/include/astra/CudaSirtAlgorithm.h
index 7f3d67b..34b5a66 100644
--- a/include/astra/CudaSirtAlgorithm.h
+++ b/include/astra/CudaSirtAlgorithm.h
@@ -107,7 +107,7 @@ public:
 	bool initialize(CProjector2D* _pProjector,
 	                CFloat32ProjectionData2D* _pSinogram, 
 					CFloat32VolumeData2D* _pReconstruction,
-					int _iGPUindex = 0, int _iDetectorSuperSampling = 1,
+					int _iGPUindex = -1, int _iDetectorSuperSampling = 1,
 					int _iPixelSuperSampling = 1);
 
 	/** Get a description of the class.
diff --git a/matlab/mex/astra_mex_c.cpp b/matlab/mex/astra_mex_c.cpp
index 0068664..816f1f3 100644
--- a/matlab/mex/astra_mex_c.cpp
+++ b/matlab/mex/astra_mex_c.cpp
@@ -36,6 +36,8 @@ $Id$
 
 #include "astra/Globals.h"
 
+#include "../cuda/2d/darthelper.h"
+
 using namespace std;
 using namespace astra;
 
@@ -72,6 +74,22 @@ void astra_mex_use_cuda(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs
 }
 
 //-----------------------------------------------------------------------------------------
+/** set_gpu_index = astra_mex('set_gpu_index');
+ * 
+ * Set active GPU
+ */
+void astra_mex_set_gpu_index(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs[])
+{ 
+#ifdef ASTRA_CUDA
+	if (nrhs >= 2) {
+		bool ret = astraCUDA::setGPUIndex((int)mxGetScalar(prhs[1]));
+		if (!ret)
+			mexPrintf("Failed to set GPU %d\n", (int)mxGetScalar(prhs[1]));
+	}
+#endif
+}
+
+//-----------------------------------------------------------------------------------------
 /** version_number = astra_mex('version');
  * 
  * Fetch the version number of the toolbox.
@@ -117,6 +135,8 @@ void mexFunction(int nlhs, mxArray* plhs[],
 		astra_mex_use_cuda(nlhs, plhs, nrhs, prhs); 
 	} else if (sMode ==  std::string("credits")) {	
 		astra_mex_credits(nlhs, plhs, nrhs, prhs); 
+	} else if (sMode == std::string("set_gpu_index")) {
+		astra_mex_set_gpu_index(nlhs, plhs, nrhs, prhs);
 	} else {
 		printHelp();
 	}
diff --git a/matlab/tools/astra_set_gpu_index.m b/matlab/tools/astra_set_gpu_index.m
new file mode 100644
index 0000000..192ba00
--- /dev/null
+++ b/matlab/tools/astra_set_gpu_index.m
@@ -0,0 +1,18 @@
+function astra_set_gpu_index(index)
+
+%--------------------------------------------------------------------------
+% Set the index of the GPU to use
+%--------------------------------------------------------------------------
+%------------------------------------------------------------------------
+% This file is part of the
+% All Scale Tomographic Reconstruction Antwerp Toolbox ("ASTRA-Toolbox")
+%
+% Copyright: iMinds-Vision Lab, University of Antwerp
+% License: Open Source under GPLv3
+% Contact: mailto:astra@ua.ac.be
+% Website: http://astra.ua.ac.be
+%------------------------------------------------------------------------
+% $Id$
+
+astra_mex('set_gpu_index', index);
+
diff --git a/src/CudaBackProjectionAlgorithm3D.cpp b/src/CudaBackProjectionAlgorithm3D.cpp
index b60adf1..b096756 100644
--- a/src/CudaBackProjectionAlgorithm3D.cpp
+++ b/src/CudaBackProjectionAlgorithm3D.cpp
@@ -51,7 +51,7 @@ std::string CCudaBackProjectionAlgorithm3D::type = "BP3D_CUDA";
 CCudaBackProjectionAlgorithm3D::CCudaBackProjectionAlgorithm3D() 
 {
 	m_bIsInitialized = false;
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iVoxelSuperSampling = 1;
 }
 
@@ -101,7 +101,7 @@ bool CCudaBackProjectionAlgorithm3D::initialize(const Config& _cfg)
 		return false;
 	}
 
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 	m_iVoxelSuperSampling = (int)_cfg.self->getOptionNumerical("VoxelSuperSampling", 1);
 	CC.markOptionParsed("VoxelSuperSampling");
diff --git a/src/CudaCglsAlgorithm3D.cpp b/src/CudaCglsAlgorithm3D.cpp
index 07569a2..37d3592 100644
--- a/src/CudaCglsAlgorithm3D.cpp
+++ b/src/CudaCglsAlgorithm3D.cpp
@@ -51,7 +51,7 @@ CCudaCglsAlgorithm3D::CCudaCglsAlgorithm3D()
 {
 	m_bIsInitialized = false;
 	m_pCgls = 0;
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iVoxelSuperSampling = 1;
 	m_iDetectorSuperSampling = 1;
 }
@@ -106,7 +106,7 @@ bool CCudaCglsAlgorithm3D::initialize(const Config& _cfg)
 		return false;
 	}
 
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 	m_iDetectorSuperSampling = (int)_cfg.self->getOptionNumerical("DetectorSuperSampling", 1);
 	CC.markOptionParsed("DetectorSuperSampling");
diff --git a/src/CudaDartMaskAlgorithm.cpp b/src/CudaDartMaskAlgorithm.cpp
index 9c9b83f..911e448 100644
--- a/src/CudaDartMaskAlgorithm.cpp
+++ b/src/CudaDartMaskAlgorithm.cpp
@@ -81,7 +81,7 @@ bool CCudaDartMaskAlgorithm::initialize(const Config& _cfg)
 	CC.markNodeParsed("MaskDataId");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaDartMaskAlgorithm3D.cpp b/src/CudaDartMaskAlgorithm3D.cpp
index 7965587..f5d5d82 100644
--- a/src/CudaDartMaskAlgorithm3D.cpp
+++ b/src/CudaDartMaskAlgorithm3D.cpp
@@ -81,7 +81,7 @@ bool CCudaDartMaskAlgorithm3D::initialize(const Config& _cfg)
 	CC.markNodeParsed("MaskDataId");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaDartSmoothingAlgorithm.cpp b/src/CudaDartSmoothingAlgorithm.cpp
index 91cde6d..e77606b 100644
--- a/src/CudaDartSmoothingAlgorithm.cpp
+++ b/src/CudaDartSmoothingAlgorithm.cpp
@@ -81,7 +81,7 @@ bool CCudaDartSmoothingAlgorithm::initialize(const Config& _cfg)
 	CC.markNodeParsed("OutDataId");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaDartSmoothingAlgorithm3D.cpp b/src/CudaDartSmoothingAlgorithm3D.cpp
index 50ef847..e44970f 100644
--- a/src/CudaDartSmoothingAlgorithm3D.cpp
+++ b/src/CudaDartSmoothingAlgorithm3D.cpp
@@ -81,7 +81,7 @@ bool CCudaDartSmoothingAlgorithm3D::initialize(const Config& _cfg)
 	CC.markNodeParsed("OutDataId");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaDataOperationAlgorithm.cpp b/src/CudaDataOperationAlgorithm.cpp
index ed2ac94..f27cab8 100644
--- a/src/CudaDataOperationAlgorithm.cpp
+++ b/src/CudaDataOperationAlgorithm.cpp
@@ -94,7 +94,7 @@ bool CCudaDataOperationAlgorithm::initialize(const Config& _cfg)
 	CC.markNodeParsed("Scalar");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaFDKAlgorithm3D.cpp b/src/CudaFDKAlgorithm3D.cpp
index 8f9e7b8..6b65641 100644
--- a/src/CudaFDKAlgorithm3D.cpp
+++ b/src/CudaFDKAlgorithm3D.cpp
@@ -48,7 +48,7 @@ std::string CCudaFDKAlgorithm3D::type = "FDK_CUDA";
 CCudaFDKAlgorithm3D::CCudaFDKAlgorithm3D() 
 {
 	m_bIsInitialized = false;
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iVoxelSuperSampling = 1;
 }
 
@@ -100,7 +100,7 @@ bool CCudaFDKAlgorithm3D::initialize(const Config& _cfg)
 		return false;
 	}
 
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 	m_iVoxelSuperSampling = (int)_cfg.self->getOptionNumerical("VoxelSuperSampling", 1);
 	CC.markOptionParsed("VoxelSuperSampling");
diff --git a/src/CudaFilteredBackProjectionAlgorithm.cpp b/src/CudaFilteredBackProjectionAlgorithm.cpp
index 75a1534..3656f3f 100644
--- a/src/CudaFilteredBackProjectionAlgorithm.cpp
+++ b/src/CudaFilteredBackProjectionAlgorithm.cpp
@@ -153,7 +153,7 @@ bool CCudaFilteredBackProjectionAlgorithm::initialize(const Config& _cfg)
 	ASTRA_DELETE(node);
 
 	// GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 
 	// Pixel supersampling factor
@@ -286,7 +286,7 @@ bool CCudaFilteredBackProjectionAlgorithm::check()
 	ASTRA_CONFIG_CHECK(m_pReconstruction->isInitialized(), "FBP_CUDA", "Reconstruction Data Object Not Initialized.");
 
 	// check gpu index
-	ASTRA_CONFIG_CHECK(m_iGPUIndex >= 0, "FBP_CUDA", "GPUIndex must be a non-negative integer.");
+	ASTRA_CONFIG_CHECK(m_iGPUIndex >= -1, "FBP_CUDA", "GPUIndex must be a non-negative integer.");
 	// check pixel supersampling
 	ASTRA_CONFIG_CHECK(m_iPixelSuperSampling >= 0, "FBP_CUDA", "PixelSuperSampling must be a non-negative integer.");
 
diff --git a/src/CudaForwardProjectionAlgorithm.cpp b/src/CudaForwardProjectionAlgorithm.cpp
index 965c4af..bd9cffb 100644
--- a/src/CudaForwardProjectionAlgorithm.cpp
+++ b/src/CudaForwardProjectionAlgorithm.cpp
@@ -87,7 +87,7 @@ bool CCudaForwardProjectionAlgorithm::initialize(const Config& _cfg)
 	CC.markNodeParsed("VolumeDataId");
 
 	// GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
@@ -153,7 +153,7 @@ bool CCudaForwardProjectionAlgorithm::check()
 	//ASTRA_CONFIG_CHECK((iImageSideBlocks * G_BLOCKIMAGESIZE) == m_pVolume->getHeight(), "FP_CUDA", "Volume Height must be a multiple of G_BLOCKIMAGESIZE");
 	//ASTRA_CONFIG_CHECK(m_pProjectionGeometry->getDetectorCount() == (m_pVolume->getWidth() * 3 / 2), "SIRT_CUDA", "Number of detectors must be 1.5 times the width of the image");
 
-	ASTRA_CONFIG_CHECK(m_iGPUIndex >= 0, "FP_CUDA", "GPUIndex must be a non-negative integer.");
+	ASTRA_CONFIG_CHECK(m_iGPUIndex >= -1, "FP_CUDA", "GPUIndex must be a non-negative integer.");
 
 	// success
 	m_bIsInitialized = true;
diff --git a/src/CudaForwardProjectionAlgorithm3D.cpp b/src/CudaForwardProjectionAlgorithm3D.cpp
index e9289f1..f64620f 100644
--- a/src/CudaForwardProjectionAlgorithm3D.cpp
+++ b/src/CudaForwardProjectionAlgorithm3D.cpp
@@ -54,7 +54,7 @@ std::string CCudaForwardProjectionAlgorithm3D::type = "FP3D_CUDA";
 CCudaForwardProjectionAlgorithm3D::CCudaForwardProjectionAlgorithm3D() 
 {
 	m_bIsInitialized = false;
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iDetectorSuperSampling = 1;
 	m_pProjector = 0;
 	m_pProjections = 0;
@@ -107,7 +107,7 @@ bool CCudaForwardProjectionAlgorithm3D::initialize(const Config& _cfg)
 	CC.markNodeParsed("ProjectorId");
 
 	// GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 	m_iDetectorSuperSampling = (int)_cfg.self->getOptionNumerical("DetectorSuperSampling", 1);
 	CC.markOptionParsed("DetectorSuperSampling");
@@ -160,7 +160,7 @@ bool CCudaForwardProjectionAlgorithm3D::check()
 	ASTRA_CONFIG_CHECK(m_pVolume->isInitialized(), "FP3D_CUDA", "Volume Data Object Not Initialized.");
 
 	ASTRA_CONFIG_CHECK(m_iDetectorSuperSampling >= 1, "FP3D_CUDA", "DetectorSuperSampling must be a positive integer.");
-	ASTRA_CONFIG_CHECK(m_iGPUIndex >= 0, "FP3D_CUDA", "GPUIndex must be a non-negative integer.");
+	ASTRA_CONFIG_CHECK(m_iGPUIndex >= -1, "FP3D_CUDA", "GPUIndex must be a non-negative integer.");
 
 	// check compatibility between projector and data classes
 //	ASTRA_CONFIG_CHECK(m_pSinogram->getGeometry()->isEqual(m_pProjector->getProjectionGeometry()), "SIRT_CUDA", "Projection Data not compatible with the specified Projector.");
diff --git a/src/CudaReconstructionAlgorithm2D.cpp b/src/CudaReconstructionAlgorithm2D.cpp
index d567158..385cef1 100644
--- a/src/CudaReconstructionAlgorithm2D.cpp
+++ b/src/CudaReconstructionAlgorithm2D.cpp
@@ -76,7 +76,7 @@ void CCudaReconstructionAlgorithm2D::_clear()
 	m_bAlgoInit = false;
 	CReconstructionAlgorithm2D::_clear();
 
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iDetectorSuperSampling = 1;
 	m_iPixelSuperSampling = 1;
 }
@@ -153,7 +153,7 @@ bool CCudaReconstructionAlgorithm2D::initialize(const Config& _cfg)
 	}
 
 	// GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
@@ -240,7 +240,7 @@ bool CCudaReconstructionAlgorithm2D::_check()
 
 	ASTRA_CONFIG_CHECK(m_iDetectorSuperSampling >= 1, "SIRT_CUDA", "DetectorSuperSampling must be a positive integer.");
 	ASTRA_CONFIG_CHECK(m_iPixelSuperSampling >= 1, "SIRT_CUDA", "PixelSuperSampling must be a positive integer.");
-	ASTRA_CONFIG_CHECK(m_iGPUIndex >= 0, "SIRT_CUDA", "GPUIndex must be a non-negative integer.");
+	ASTRA_CONFIG_CHECK(m_iGPUIndex >= -1, "SIRT_CUDA", "GPUIndex must be a non-negative integer.");
 
 	// check compatibility between projector and data classes
 //	ASTRA_CONFIG_CHECK(m_pSinogram->getGeometry()->isEqual(m_pProjector->getProjectionGeometry()), "SIRT_CUDA", "Projection Data not compatible with the specified Projector.");
diff --git a/src/CudaRoiSelectAlgorithm.cpp b/src/CudaRoiSelectAlgorithm.cpp
index f835c59..bbe2235 100644
--- a/src/CudaRoiSelectAlgorithm.cpp
+++ b/src/CudaRoiSelectAlgorithm.cpp
@@ -74,7 +74,7 @@ bool CCudaRoiSelectAlgorithm::initialize(const Config& _cfg)
 	CC.markNodeParsed("DataId");
 
 	// Option: GPU number
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUIndex", m_iGPUIndex);
 	CC.markOptionParsed("GPUindex");
 	if (!_cfg.self->hasOption("GPUindex"))
diff --git a/src/CudaSirtAlgorithm3D.cpp b/src/CudaSirtAlgorithm3D.cpp
index f23d0f6..179ded8 100644
--- a/src/CudaSirtAlgorithm3D.cpp
+++ b/src/CudaSirtAlgorithm3D.cpp
@@ -52,7 +52,7 @@ CCudaSirtAlgorithm3D::CCudaSirtAlgorithm3D()
 {
 	m_bIsInitialized = false;
 	m_pSirt = 0;
-	m_iGPUIndex = 0;
+	m_iGPUIndex = -1;
 	m_iVoxelSuperSampling = 1;
 	m_iDetectorSuperSampling = 1;
 }
@@ -107,7 +107,7 @@ bool CCudaSirtAlgorithm3D::initialize(const Config& _cfg)
 		return false;
 	}
 
-	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", 0);
+	m_iGPUIndex = (int)_cfg.self->getOptionNumerical("GPUindex", -1);
 	CC.markOptionParsed("GPUindex");
 	m_iDetectorSuperSampling = (int)_cfg.self->getOptionNumerical("DetectorSuperSampling", 1);
 	CC.markOptionParsed("DetectorSuperSampling");