Merge branch 'texture'

This replaces the deprecated CUDA texture reference API by texture objects.

5 files changed, 120 insertions, 151 deletions

diff --git a/cuda/3d/cone_bp.cu b/cuda/3d/cone_bp.cu
index e265304..41d781c 100644
--- a/cuda/3d/cone_bp.cu
+++ b/cuda/3d/cone_bp.cu
@@ -35,10 +35,6 @@ along with the ASTRA Toolbox. If not, see <http://www.gnu.org/licenses/>.
 
 #include <cuda.h>
 
-typedef texture<float, 3, cudaReadModeElementType> texture3D;
-
-static texture3D gT_coneProjTexture;
-
 namespace astraCUDA3d {
 
 static const unsigned int g_volBlockZ = 6;
@@ -57,28 +53,12 @@ struct DevConeParams {
 
 __constant__ DevConeParams gC_C[g_MaxAngles];
 
-bool bindProjDataTexture(const cudaArray* array)
-{
-	cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<float>();
-
-	gT_coneProjTexture.addressMode[0] = cudaAddressModeBorder;
-	gT_coneProjTexture.addressMode[1] = cudaAddressModeBorder;
-	gT_coneProjTexture.addressMode[2] = cudaAddressModeBorder;
-	gT_coneProjTexture.filterMode = cudaFilterModeLinear;
-	gT_coneProjTexture.normalized = false;
-
-	cudaBindTextureToArray(gT_coneProjTexture, array, channelDesc);
-
-	// TODO: error value?
-
-	return true;
-}
-
-
 //__launch_bounds__(32*16, 4)
 template<bool FDKWEIGHT, unsigned int ZSIZE>
-__global__ void dev_cone_BP(void* D_volData, unsigned int volPitch, int startAngle,
-                            int angleOffset, const astraCUDA3d::SDimensions3D dims,
+__global__ void dev_cone_BP(void* D_volData, unsigned int volPitch,
+                            cudaTextureObject_t tex,
+                            int startAngle, int angleOffset,
+                            const astraCUDA3d::SDimensions3D dims,
                             float fOutputScale)
 {
 	float* volData = (float*)D_volData;
@@ -133,7 +113,7 @@ __global__ void dev_cone_BP(void* D_volData, unsigned int volPitch, int startAng
 				fr = __fdividef(1.0f, fDen);
 				fU = fUNum * fr;
 				fV = fVNum * fr;
-				float fVal = tex3D(gT_coneProjTexture, fU, fAngle, fV);
+				float fVal = tex3D<float>(tex, fU, fAngle, fV);
 				Z[idx] += fr*fr*fVal;
 
 				fUNum += fCu.z;
@@ -154,7 +134,7 @@ __global__ void dev_cone_BP(void* D_volData, unsigned int volPitch, int startAng
 
 
 // supersampling version
-__global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, int iRaysPerVoxelDim, float fOutputScale)
+__global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, cudaTextureObject_t tex, int startAngle, int angleOffset, const SDimensions3D dims, int iRaysPerVoxelDim, float fOutputScale)
 {
 	float* volData = (float*)D_volData;
 
@@ -220,7 +200,7 @@ __global__ void dev_cone_BP_SS(void* D_volData, unsigned int volPitch, int start
 				const float fU = fUNum * fr;
 				const float fV = fVNum * fr;
 
-				fVal += tex3D(gT_coneProjTexture, fU, fAngle, fV) * fr * fr;
+				fVal += tex3D<float>(tex, fU, fAngle, fV) * fr * fr;
 
 				fZs += fSubStep;
 			}
@@ -313,7 +293,9 @@ bool ConeBP_Array(cudaPitchedPtr D_volumeData,
                   const SDimensions3D& dims, const SConeProjection* angles,
                   const SProjectorParams3D& params)
 {
-	bindProjDataTexture(D_projArray);
+	cudaTextureObject_t D_texObj;
+	if (!createTextureObject3D(D_projArray, D_texObj))
+		return false;
 
 	float fOutputScale;
 	if (params.bFDKWeighting) {
@@ -323,14 +305,16 @@ bool ConeBP_Array(cudaPitchedPtr D_volumeData,
 		fOutputScale = params.fOutputScale * (params.fVolScaleX * params.fVolScaleY * params.fVolScaleZ);
 	}
 
+	bool ok = true;
+
 	for (unsigned int th = 0; th < dims.iProjAngles; th += g_MaxAngles) {
 		unsigned int angleCount = g_MaxAngles;
 		if (th + angleCount > dims.iProjAngles)
 			angleCount = dims.iProjAngles - th;
 
-		bool ok = transferConstants(angles, angleCount, params);
+		ok = transferConstants(angles, angleCount, params);
 		if (!ok)
-			return false;
+			break;
 
 		dim3 dimBlock(g_volBlockX, g_volBlockY);
 
@@ -343,31 +327,33 @@ bool ConeBP_Array(cudaPitchedPtr D_volumeData,
 		// 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 (params.bFDKWeighting) {
 				if (dims.iVolZ == 1) {
-					dev_cone_BP<true, 1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_cone_BP<true, 1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				} else {
-					dev_cone_BP<true, g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_cone_BP<true, g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				}
 			} else if (params.iRaysPerVoxelDim == 1) {
 				if (dims.iVolZ == 1) {
-					dev_cone_BP<false, 1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_cone_BP<false, 1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				} else {
-					dev_cone_BP<false, g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_cone_BP<false, g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				}
 			} else
-				dev_cone_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, params.iRaysPerVoxelDim, fOutputScale);
+				dev_cone_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, params.iRaysPerVoxelDim, fOutputScale);
 		}
 
 		// TODO: Consider not synchronizing here, if possible.
-		if (!checkCuda(cudaThreadSynchronize(), "cone_bp"))
-			return false;
+		ok = checkCuda(cudaThreadSynchronize(), "cone_bp");
+		if (!ok)
+			break;
 
 		angles = angles + angleCount;
 		// printf("%f\n", toc(t));
 
 	}
 
+	cudaDestroyTextureObject(D_texObj);
 
-	return true;
+	return ok;
 }
 
 bool ConeBP(cudaPitchedPtr D_volumeData,
diff --git a/cuda/3d/cone_fp.cu b/cuda/3d/cone_fp.cu
index 2c3d1f6..2ef58ee 100644
--- a/cuda/3d/cone_fp.cu
+++ b/cuda/3d/cone_fp.cu
@@ -35,10 +35,6 @@ along with the ASTRA Toolbox. If not, see <http://www.gnu.org/licenses/>.
 
 #include <cuda.h>
 
-typedef texture<float, 3, cudaReadModeElementType> texture3D;
-
-static texture3D gT_coneVolumeTexture;
-
 namespace astraCUDA3d {
 
 static const unsigned int g_anglesPerBlock = 4;
@@ -63,24 +59,6 @@ __constant__ float gC_DetVY[g_MaxAngles];
 __constant__ float gC_DetVZ[g_MaxAngles];
 
 
-bool bindVolumeDataTexture(const cudaArray* array)
-{
-	cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<float>();
-
-	gT_coneVolumeTexture.addressMode[0] = cudaAddressModeBorder;
-	gT_coneVolumeTexture.addressMode[1] = cudaAddressModeBorder;
-	gT_coneVolumeTexture.addressMode[2] = cudaAddressModeBorder;
-	gT_coneVolumeTexture.filterMode = cudaFilterModeLinear;
-	gT_coneVolumeTexture.normalized = false;
-
-	cudaBindTextureToArray(gT_coneVolumeTexture, array, channelDesc);
-
-	// TODO: error value?
-
-	return true;
-}
-
-
 // x=0, y=1, z=2
 struct DIR_X {
 	__device__ float nSlices(const SDimensions3D& dims) const { return dims.iVolX; }
@@ -89,7 +67,7 @@ struct DIR_X {
 	__device__ float c0(float x, float y, float z) const { return x; }
 	__device__ float c1(float x, float y, float z) const { return y; }
 	__device__ float c2(float x, float y, float z) const { return z; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_coneVolumeTexture, f0, f1, f2); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f0, f1, f2); }
 	__device__ float x(float f0, float f1, float f2) const { return f0; }
 	__device__ float y(float f0, float f1, float f2) const { return f1; }
 	__device__ float z(float f0, float f1, float f2) const { return f2; }
@@ -103,7 +81,7 @@ struct DIR_Y {
 	__device__ float c0(float x, float y, float z) const { return y; }
 	__device__ float c1(float x, float y, float z) const { return x; }
 	__device__ float c2(float x, float y, float z) const { return z; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_coneVolumeTexture, f1, f0, f2); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f1, f0, f2); }
 	__device__ float x(float f0, float f1, float f2) const { return f1; }
 	__device__ float y(float f0, float f1, float f2) const { return f0; }
 	__device__ float z(float f0, float f1, float f2) const { return f2; }
@@ -117,7 +95,7 @@ struct DIR_Z {
 	__device__ float c0(float x, float y, float z) const { return z; }
 	__device__ float c1(float x, float y, float z) const { return x; }
 	__device__ float c2(float x, float y, float z) const { return y; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_coneVolumeTexture, f1, f2, f0); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f1, f2, f0); }
 	__device__ float x(float f0, float f1, float f2) const { return f1; }
 	__device__ float y(float f0, float f1, float f2) const { return f2; }
 	__device__ float z(float f0, float f1, float f2) const { return f0; }
@@ -144,6 +122,7 @@ struct SCALE_NONCUBE {
 
 template<class COORD, class SCALE>
 __global__ void cone_FP_t(float* D_projData, unsigned int projPitch,
+                          cudaTextureObject_t tex,
                           unsigned int startSlice,
                           unsigned int startAngle, unsigned int endAngle,
                           const SDimensions3D dims,
@@ -208,7 +187,7 @@ __global__ void cone_FP_t(float* D_projData, unsigned int projPitch,
 
 		for (int s = startSlice; s < endSlice; ++s)
 		{
-			fVal += c.tex(f0, f1, f2);
+			fVal += c.tex(tex, f0, f1, f2);
 			f0 += 1.0f;
 			f1 += a1;
 			f2 += a2;
@@ -222,6 +201,7 @@ __global__ void cone_FP_t(float* D_projData, unsigned int projPitch,
 
 template<class COORD>
 __global__ void cone_FP_SS_t(float* D_projData, unsigned int projPitch,
+                             cudaTextureObject_t tex,
                              unsigned int startSlice,
                              unsigned int startAngle, unsigned int endAngle,
                              const SDimensions3D dims, int iRaysPerDetDim,
@@ -295,7 +275,7 @@ __global__ void cone_FP_SS_t(float* D_projData, unsigned int projPitch,
 
 		for (int s = startSlice; s < endSlice; ++s)
 		{
-			fVal += c.tex(f0, f1, f2);
+			fVal += c.tex(tex, f0, f1, f2);
 			f0 += 1.0f;
 			f1 += a1;
 			f2 += a2;
@@ -313,7 +293,9 @@ __global__ void cone_FP_SS_t(float* D_projData, unsigned int projPitch,
 
 
 bool ConeFP_Array_internal(cudaPitchedPtr D_projData,
-                  const SDimensions3D& dims, unsigned int angleCount, const SConeProjection* angles,
+                  cudaTextureObject_t D_texObj,
+                  const SDimensions3D& dims,
+                  unsigned int angleCount, const SConeProjection* angles,
                   const SProjectorParams3D& params)
 {
 	// transfer angles to constant memory
@@ -419,29 +401,29 @@ bool ConeFP_Array_internal(cudaPitchedPtr D_projData,
 					for (unsigned int i = 0; i < dims.iVolX; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 							if (cube)
-								cone_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+								cone_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, scube);
 							else
-								cone_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeX);
+								cone_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, snoncubeX);
 						else
-							cone_FP_SS_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeX);
+							cone_FP_SS_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeX);
 				} else if (blockDirection == 1) {
 					for (unsigned int i = 0; i < dims.iVolY; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 							if (cube)
-								cone_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+								cone_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, scube);
 							else
-								cone_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeY);
+								cone_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, snoncubeY);
 						else
-							cone_FP_SS_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeY);
+							cone_FP_SS_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeY);
 				} else if (blockDirection == 2) {
 					for (unsigned int i = 0; i < dims.iVolZ; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 							if (cube)
-								cone_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+								cone_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, scube);
 							else
-								cone_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeZ);
+								cone_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, snoncubeZ);
 						else
-							cone_FP_SS_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeZ);
+							cone_FP_SS_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeZ);
 				}
 
 			}
@@ -470,10 +452,14 @@ bool ConeFP(cudaPitchedPtr D_volumeData,
             const SProjectorParams3D& params)
 {
 	// transfer volume to array
-
 	cudaArray* cuArray = allocateVolumeArray(dims);
 	transferVolumeToArray(D_volumeData, cuArray, dims);
-	bindVolumeDataTexture(cuArray);
+
+	cudaTextureObject_t D_texObj;
+	if (!createTextureObject3D(cuArray, D_texObj)) {
+		cudaFreeArray(cuArray);
+		return false;
+	}
 
 	bool ret;
 
@@ -485,7 +471,7 @@ bool ConeFP(cudaPitchedPtr D_volumeData,
 		cudaPitchedPtr D_subprojData = D_projData;
 		D_subprojData.ptr = (char*)D_projData.ptr + iAngle * D_projData.pitch;
 
-		ret = ConeFP_Array_internal(D_subprojData,
+		ret = ConeFP_Array_internal(D_subprojData, D_texObj,
 		                            dims, iEndAngle - iAngle, angles + iAngle,
 		                            params);
 		if (!ret)
diff --git a/cuda/3d/par3d_bp.cu b/cuda/3d/par3d_bp.cu
index 1dc75ce..27d95fe 100644
--- a/cuda/3d/par3d_bp.cu
+++ b/cuda/3d/par3d_bp.cu
@@ -35,10 +35,6 @@ along with the ASTRA Toolbox. If not, see <http://www.gnu.org/licenses/>.
 
 #include <cuda.h>
 
-typedef texture<float, 3, cudaReadModeElementType> texture3D;
-
-static texture3D gT_par3DProjTexture;
-
 namespace astraCUDA3d {
 
 static const unsigned int g_volBlockZ = 6;
@@ -58,26 +54,8 @@ __constant__ DevPar3DParams gC_C[g_MaxAngles];
 __constant__ float gC_scale[g_MaxAngles];
 
 
-static bool bindProjDataTexture(const cudaArray* array)
-{
-	cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<float>();
-
-	gT_par3DProjTexture.addressMode[0] = cudaAddressModeBorder;
-	gT_par3DProjTexture.addressMode[1] = cudaAddressModeBorder;
-	gT_par3DProjTexture.addressMode[2] = cudaAddressModeBorder;
-	gT_par3DProjTexture.filterMode = cudaFilterModeLinear;
-	gT_par3DProjTexture.normalized = false;
-
-	cudaBindTextureToArray(gT_par3DProjTexture, array, channelDesc);
-
-	// TODO: error value?
-
-	return true;
-}
-
-
 template<unsigned int ZSIZE>
-__global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, float fOutputScale)
+__global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, cudaTextureObject_t tex, int startAngle, int angleOffset, const SDimensions3D dims, float fOutputScale)
 {
 	float* volData = (float*)D_volData;
 
@@ -125,7 +103,7 @@ __global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAn
 
 			for (int idx = 0; idx < ZSIZE; ++idx) {
 
-				float fVal = tex3D(gT_par3DProjTexture, fU, fAngle, fV);
+				float fVal = tex3D<float>(tex, fU, fAngle, fV);
 				Z[idx] += fVal * fS;
 
 				fU += fCu.z;
@@ -144,7 +122,7 @@ __global__ void dev_par3D_BP(void* D_volData, unsigned int volPitch, int startAn
 }
 
 // supersampling version
-__global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int startAngle, int angleOffset, const SDimensions3D dims, int iRaysPerVoxelDim, float fOutputScale)
+__global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, cudaTextureObject_t tex, int startAngle, int angleOffset, const SDimensions3D dims, int iRaysPerVoxelDim, float fOutputScale)
 {
 	float* volData = (float*)D_volData;
 
@@ -206,7 +184,7 @@ __global__ void dev_par3D_BP_SS(void* D_volData, unsigned int volPitch, int star
 				const float fU = fCu.w + fXs * fCu.x + fYs * fCu.y + fZs * fCu.z;
 				const float fV = fCv.w + fXs * fCv.x + fYs * fCv.y + fZs * fCv.z;
 
-				fVal += tex3D(gT_par3DProjTexture, fU, fAngle, fV) * fS;
+				fVal += tex3D<float>(tex, fU, fAngle, fV) * fS;
 				fZs += fSubStep;
 			}
 			fYs += fSubStep;
@@ -259,18 +237,22 @@ bool Par3DBP_Array(cudaPitchedPtr D_volumeData,
                    const SDimensions3D& dims, const SPar3DProjection* angles,
                    const SProjectorParams3D& params)
 {
-	bindProjDataTexture(D_projArray);
+	cudaTextureObject_t D_texObj;
+	if (!createTextureObject3D(D_projArray, D_texObj))
+		return false;
 
 	float fOutputScale = params.fOutputScale * params.fVolScaleX * params.fVolScaleY * params.fVolScaleZ;
 
+	bool ok = true;
+
 	for (unsigned int th = 0; th < dims.iProjAngles; th += g_MaxAngles) {
 		unsigned int angleCount = g_MaxAngles;
 		if (th + angleCount > dims.iProjAngles)
 			angleCount = dims.iProjAngles - th;
 
-		bool ok = transferConstants(angles, angleCount, params);
+		ok = transferConstants(angles, angleCount, params);
 		if (!ok)
-			return false;
+			break;
 
 		dim3 dimBlock(g_volBlockX, g_volBlockY);
 
@@ -283,23 +265,26 @@ bool Par3DBP_Array(cudaPitchedPtr D_volumeData,
 			// 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 (params.iRaysPerVoxelDim == 1) {
 				if (dims.iVolZ == 1) {
-					dev_par3D_BP<1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_par3D_BP<1><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				} else {
-					dev_par3D_BP<g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, fOutputScale);
+					dev_par3D_BP<g_volBlockZ><<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, fOutputScale);
 				}
 			} else
-				dev_par3D_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), i, th, dims, params.iRaysPerVoxelDim, fOutputScale);
+				dev_par3D_BP_SS<<<dimGrid, dimBlock>>>(D_volumeData.ptr, D_volumeData.pitch/sizeof(float), D_texObj, i, th, dims, params.iRaysPerVoxelDim, fOutputScale);
 		}
 
 		// TODO: Consider not synchronizing here, if possible.
-		if (!checkCuda(cudaThreadSynchronize(), "cone_bp"))
-			return false;
+		ok = checkCuda(cudaThreadSynchronize(), "cone_bp");
+		if (!ok)
+			break;
 
 		angles = angles + angleCount;
 		// printf("%f\n", toc(t));
 
 	}
 
+	cudaDestroyTextureObject(D_texObj);
+
 	return true;
 }
 
diff --git a/cuda/3d/par3d_fp.cu b/cuda/3d/par3d_fp.cu
index 5daddc1..b2178ec 100644
--- a/cuda/3d/par3d_fp.cu
+++ b/cuda/3d/par3d_fp.cu
@@ -35,10 +35,6 @@ along with the ASTRA Toolbox. If not, see <http://www.gnu.org/licenses/>.
 
 #include <cuda.h>
 
-typedef texture<float, 3, cudaReadModeElementType> texture3D;
-
-static texture3D gT_par3DVolumeTexture;
-
 namespace astraCUDA3d {
 
 static const unsigned int g_anglesPerBlock = 4;
@@ -63,24 +59,6 @@ __constant__ float gC_DetVY[g_MaxAngles];
 __constant__ float gC_DetVZ[g_MaxAngles];
 
 
-static bool bindVolumeDataTexture(const cudaArray* array)
-{
-	cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<float>();
-
-	gT_par3DVolumeTexture.addressMode[0] = cudaAddressModeBorder;
-	gT_par3DVolumeTexture.addressMode[1] = cudaAddressModeBorder;
-	gT_par3DVolumeTexture.addressMode[2] = cudaAddressModeBorder;
-	gT_par3DVolumeTexture.filterMode = cudaFilterModeLinear;
-	gT_par3DVolumeTexture.normalized = false;
-
-	cudaBindTextureToArray(gT_par3DVolumeTexture, array, channelDesc);
-
-	// TODO: error value?
-
-	return true;
-}
-
-
 // x=0, y=1, z=2
 struct DIR_X {
 	__device__ float nSlices(const SDimensions3D& dims) const { return dims.iVolX; }
@@ -89,7 +67,7 @@ struct DIR_X {
 	__device__ float c0(float x, float y, float z) const { return x; }
 	__device__ float c1(float x, float y, float z) const { return y; }
 	__device__ float c2(float x, float y, float z) const { return z; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_par3DVolumeTexture, f0, f1, f2); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f0, f1, f2); }
 	__device__ float x(float f0, float f1, float f2) const { return f0; }
 	__device__ float y(float f0, float f1, float f2) const { return f1; }
 	__device__ float z(float f0, float f1, float f2) const { return f2; }
@@ -103,7 +81,7 @@ struct DIR_Y {
 	__device__ float c0(float x, float y, float z) const { return y; }
 	__device__ float c1(float x, float y, float z) const { return x; }
 	__device__ float c2(float x, float y, float z) const { return z; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_par3DVolumeTexture, f1, f0, f2); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f1, f0, f2); }
 	__device__ float x(float f0, float f1, float f2) const { return f1; }
 	__device__ float y(float f0, float f1, float f2) const { return f0; }
 	__device__ float z(float f0, float f1, float f2) const { return f2; }
@@ -117,7 +95,7 @@ struct DIR_Z {
 	__device__ float c0(float x, float y, float z) const { return z; }
 	__device__ float c1(float x, float y, float z) const { return x; }
 	__device__ float c2(float x, float y, float z) const { return y; }
-	__device__ float tex(float f0, float f1, float f2) const { return tex3D(gT_par3DVolumeTexture, f1, f2, f0); }
+	__device__ float tex(cudaTextureObject_t tex, float f0, float f1, float f2) const { return tex3D<float>(tex, f1, f2, f0); }
 	__device__ float x(float f0, float f1, float f2) const { return f1; }
 	__device__ float y(float f0, float f1, float f2) const { return f2; }
 	__device__ float z(float f0, float f1, float f2) const { return f0; }
@@ -145,6 +123,7 @@ struct SCALE_NONCUBE {
 
 template<class COORD, class SCALE>
 __global__ void par3D_FP_t(float* D_projData, unsigned int projPitch,
+                           cudaTextureObject_t tex,
                            unsigned int startSlice,
                            unsigned int startAngle, unsigned int endAngle,
                            const SDimensions3D dims,
@@ -210,7 +189,7 @@ __global__ void par3D_FP_t(float* D_projData, unsigned int projPitch,
 
 		for (int s = startSlice; s < endSlice; ++s)
 		{
-			fVal += c.tex(f0, f1, f2);
+			fVal += c.tex(tex, f0, f1, f2);
 			f0 += 1.0f;
 			f1 += a1;
 			f2 += a2;
@@ -225,6 +204,7 @@ __global__ void par3D_FP_t(float* D_projData, unsigned int projPitch,
 // Supersampling version
 template<class COORD>
 __global__ void par3D_FP_SS_t(float* D_projData, unsigned int projPitch,
+                              cudaTextureObject_t tex,
                               unsigned int startSlice,
                               unsigned int startAngle, unsigned int endAngle,
                               const SDimensions3D dims, int iRaysPerDetDim,
@@ -301,7 +281,7 @@ __global__ void par3D_FP_SS_t(float* D_projData, unsigned int projPitch,
 
 		for (int s = startSlice; s < endSlice; ++s)
 		{
-			fVal += c.tex(f0, f1, f2);
+			fVal += c.tex(tex, f0, f1, f2);
 			f0 += 1.0f;
 			f1 += a1;
 			f2 += a2;
@@ -415,7 +395,9 @@ __global__ void par3D_FP_SumSqW_t(float* D_projData, unsigned int projPitch,
 
 
 bool Par3DFP_Array_internal(cudaPitchedPtr D_projData,
-                   const SDimensions3D& dims, unsigned int angleCount, const SPar3DProjection* angles,
+                   cudaTextureObject_t D_texObj,
+                   const SDimensions3D& dims,
+                   unsigned int angleCount, const SPar3DProjection* angles,
                    const SProjectorParams3D& params)
 {
 	// transfer angles to constant memory
@@ -520,29 +502,29 @@ bool Par3DFP_Array_internal(cudaPitchedPtr D_projData,
 					for (unsigned int i = 0; i < dims.iVolX; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 								if (cube)
-										par3D_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+										par3D_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), D_texObj, i, blockStart, blockEnd, dims, scube);
 								else
-										par3D_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeX);
+										par3D_FP_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, snoncubeX);
 						else
-							par3D_FP_SS_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeX);
+							par3D_FP_SS_t<DIR_X><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeX);
 				} else if (blockDirection == 1) {
 					for (unsigned int i = 0; i < dims.iVolY; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 								if (cube)
-										par3D_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+										par3D_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, scube);
 								else
-										par3D_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeY);
+										par3D_FP_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, snoncubeY);
 						else
-							par3D_FP_SS_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeY);
+							par3D_FP_SS_t<DIR_Y><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeY);
 				} else if (blockDirection == 2) {
 					for (unsigned int i = 0; i < dims.iVolZ; i += g_blockSlices)
 						if (params.iRaysPerDetDim == 1)
 								if (cube)
-										par3D_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, scube);
+										par3D_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, scube);
 								else
-										par3D_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, snoncubeZ);
+										par3D_FP_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, snoncubeZ);
 						else
-							par3D_FP_SS_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float), i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeZ);
+							par3D_FP_SS_t<DIR_Z><<<dimGrid, dimBlock, 0, stream>>>((float*)D_projData.ptr, D_projData.pitch/sizeof(float),  D_texObj,i, blockStart, blockEnd, dims, params.iRaysPerDetDim, snoncubeZ);
 				}
 
 			}
@@ -569,10 +551,16 @@ bool Par3DFP(cudaPitchedPtr D_volumeData,
              const SDimensions3D& dims, const SPar3DProjection* angles,
              const SProjectorParams3D& params)
 {
+
 	// transfer volume to array
 	cudaArray* cuArray = allocateVolumeArray(dims);
 	transferVolumeToArray(D_volumeData, cuArray, dims);
-	bindVolumeDataTexture(cuArray);
+
+	cudaTextureObject_t D_texObj;
+	if (!createTextureObject3D(cuArray, D_texObj)) {
+		cudaFreeArray(cuArray);
+		return false;
+	}
 
 	bool ret;
 
@@ -584,7 +572,7 @@ bool Par3DFP(cudaPitchedPtr D_volumeData,
 		cudaPitchedPtr D_subprojData = D_projData;
 		D_subprojData.ptr = (char*)D_projData.ptr + iAngle * D_projData.pitch;
 
-		ret = Par3DFP_Array_internal(D_subprojData,
+		ret = Par3DFP_Array_internal(D_subprojData, D_texObj,
 		                             dims, iEndAngle - iAngle, angles + iAngle,
 		                             params);
 		if (!ret)
@@ -593,6 +581,8 @@ bool Par3DFP(cudaPitchedPtr D_volumeData,
 
 	cudaFreeArray(cuArray);
 
+	cudaDestroyTextureObject(D_texObj);
+
 	return ret;
 }
 
diff --git a/cuda/3d/util3d.cu b/cuda/3d/util3d.cu
index 71b5668..3dc915d 100644
--- a/cuda/3d/util3d.cu
+++ b/cuda/3d/util3d.cu
@@ -378,6 +378,28 @@ bool transferHostProjectionsToArray(const float *projData, cudaArray* array, con
 	return checkCuda(cudaMemcpy3D(&p), "transferHostProjectionsToArray 3D");
 }
 
+bool createTextureObject3D(cudaArray* array, cudaTextureObject_t& texObj)
+{
+	cudaChannelFormatDesc channelDesc =
+	    cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
+
+	cudaResourceDesc resDesc;
+	memset(&resDesc, 0, sizeof(resDesc));
+	resDesc.resType = cudaResourceTypeArray;
+	resDesc.res.array.array = array;
+
+	cudaTextureDesc texDesc;
+	memset(&texDesc, 0, sizeof(texDesc));
+	texDesc.addressMode[0] = cudaAddressModeBorder;
+	texDesc.addressMode[1] = cudaAddressModeBorder;
+	texDesc.addressMode[2] = cudaAddressModeBorder;
+	texDesc.filterMode = cudaFilterModeLinear;
+	texDesc.readMode = cudaReadModeElementType;
+	texDesc.normalizedCoords = 0;
+
+	return checkCuda(cudaCreateTextureObject(&texObj, &resDesc, &texDesc, NULL), "createTextureObject3D");
+}
+
 
 
 float dotProduct3D(cudaPitchedPtr data, unsigned int x, unsigned int y,