1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
|
/*
//
// Copyright 2010 SRI International
//
// This file is part of the Computational Morphometry Toolkit.
//
// http://www.nitrc.org/projects/cmtk/
//
// The Computational Morphometry Toolkit is free software: you can
// redistribute it and/or modify it under the terms of the GNU General Public
// License as published by the Free Software Foundation, either version 3 of
// the License, or (at your option) any later version.
//
// The Computational Morphometry Toolkit is distributed in the hope that it
// will be useful, but WITHOUT ANY WARRANTY; without even the implied
// warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License along
// with the Computational Morphometry Toolkit. If not, see
// <http://www.gnu.org/licenses/>.
//
// $Revision: 3168 $
//
// $LastChangedDate: 2011-04-22 12:51:51 -0700 (Fri, 22 Apr 2011) $
//
// $LastChangedBy: torstenrohlfing $
//
*/
#ifdef _WIN32
// This fixes a strange compile error using VisualStudio 2010 Express.
// See http://forums.nvidia.com/index.php?showtopic=67822
#define WIN32_LEAN_AND_MEAN
#endif
#include "cmtkDeviceImageConvolution_kernels.h"
#include "System/cmtkMemory.h"
#include "GPU/cmtkCUDA.h"
#include "GPU/cmtkDeviceMemory.h"
#include <cuda_runtime_api.h>
/// Texture reference to volume data.
texture<float, 3, cudaReadModeElementType> texRef;
__constant__ float deviceKernel[128];
__global__
void
cmtkDeviceImageConvolutionKernelX( float* dest, int dims0, int dims1, int dims2, int kernelLength, int kernelCenter )
{
const int offs = threadIdx.x + blockDim.x * blockIdx.x;
const int x = offs % dims0;
const int y = (offs / dims0) % dims1;
const int z = offs / (dims0 * dims1);
float sum = 0, total = 0;
for ( int i = 0; i < kernelLength; ++i )
{
const int xx = x + i - kernelCenter;
const float w = ( (xx>=0) && (xx<dims0) ) ? deviceKernel[i] : 0;
sum += w * tex3D( texRef, xx, y, z );
total += w;
}
dest[offs] = sum / total;
}
__global__
void
cmtkDeviceImageConvolutionKernelY( float* dest, int dims0, int dims1, int dims2, int kernelLength, int kernelCenter )
{
const int offs = threadIdx.x + blockDim.x * blockIdx.x;
const int x = offs % dims0;
const int y = (offs / dims0) % dims1;
const int z = offs / (dims0 * dims1);
float sum = 0, total = 0;
for ( int i = 0; i < kernelLength; ++i )
{
const int yy = y + i - kernelCenter;
const float w = ( (yy>=0) && (yy<dims1) ) ? deviceKernel[i] : 0;
sum += w * tex3D( texRef, x, yy, z );
total += w;
}
dest[offs] = sum / total;
}
__global__
void
cmtkDeviceImageConvolutionKernelZ( float* dest, int dims0, int dims1, int dims2, int kernelLength, int kernelCenter )
{
const int offs = threadIdx.x + blockDim.x * blockIdx.x;
const int x = offs % dims0;
const int y = (offs / dims0) % dims1;
const int z = offs / (dims0 * dims1);
float sum = 0, total = 0;
for ( int i = 0; i < kernelLength; ++i )
{
const int zz = z + i - kernelCenter;
const float w = ( (zz>=0) && (zz<dims2) ) ? deviceKernel[i] : 0;
sum += w * tex3D( texRef, x, y, zz );
total += w;
}
dest[offs] = sum / total;
}
void
cmtk::DeviceImageConvolution( float* dest, const int* dims3, void* array, const int kernelLengthX, const float* kernelX, const int kernelLengthY, const float* kernelY, const int kernelLengthZ, const float* kernelZ )
{
// Set texture parameters for fixed image indexed access
texRef.addressMode[0] = cudaAddressModeClamp;
texRef.addressMode[1] = cudaAddressModeClamp;
texRef.addressMode[2] = cudaAddressModeClamp;
texRef.filterMode = cudaFilterModePoint;
texRef.normalized = false;
cmtkCheckCallCUDA( cudaBindTextureToArray( texRef, (struct cudaArray*) array, cudaCreateChannelDesc<float>() ) );
const int nPixels = dims3[0] * dims3[1] * dims3[2];
cmtkCheckCallCUDA( cudaMemcpyToSymbol( deviceKernel, kernelX, kernelLengthX * sizeof( float ), 0, cudaMemcpyHostToDevice ) );
dim3 threads( 512 );
dim3 blocks( nPixels/512+1 );
cmtkDeviceImageConvolutionKernelX<<<blocks,threads>>>( dest, dims3[0], dims3[1], dims3[2], kernelLengthX, (kernelLengthX-1)>>1 );
cmtkCheckLastErrorCUDA;
cudaMemcpy3DParms copyParams = {0};
copyParams.srcPtr = make_cudaPitchedPtr( (void*)dest, dims3[0]*sizeof(float), dims3[0], dims3[1] );
copyParams.dstArray = (struct cudaArray*) array;
copyParams.extent = make_cudaExtent( dims3[0], dims3[1], dims3[2] );
copyParams.kind = cudaMemcpyDeviceToDevice;
cmtkCheckCallCUDA( cudaMemcpy3D( ©Params ) );
cmtkCheckCallCUDA( cudaMemcpyToSymbol( deviceKernel, kernelY, kernelLengthY * sizeof( float ), 0, cudaMemcpyHostToDevice ) );
cmtkDeviceImageConvolutionKernelY<<<blocks,threads>>>( dest, dims3[0], dims3[1], dims3[2], kernelLengthY, (kernelLengthY-1)>>1 );
cmtkCheckLastErrorCUDA;
cmtkCheckCallCUDA( cudaMemcpy3D( ©Params ) );
cmtkCheckCallCUDA( cudaMemcpyToSymbol( deviceKernel, kernelZ, kernelLengthZ * sizeof( float ), 0, cudaMemcpyHostToDevice ) );
cmtkDeviceImageConvolutionKernelZ<<<blocks,threads>>>( dest, dims3[0], dims3[1], dims3[2], kernelLengthZ, (kernelLengthZ-1)>>1 );
cmtkCheckLastErrorCUDA;
cmtkCheckCallCUDA( cudaUnbindTexture( texRef ) );
}
void
cmtk::DeviceImageConvolutionInPlace( const int* dims3, void* array, const int kernelLengthX, const float* kernelX, const int kernelLengthY, const float* kernelY, const int kernelLengthZ, const float* kernelZ )
{
const int nPixels = dims3[0] * dims3[1] * dims3[2];
cmtk::DeviceMemory<float> temporary( nPixels );
// call out-of-place-place convolution
DeviceImageConvolution( temporary.Ptr(), dims3, array, kernelLengthX, kernelX, kernelLengthY, kernelY, kernelLengthZ, kernelZ );
// copy back into original array
cudaMemcpy3DParms copyParams = {0};
copyParams.srcPtr = make_cudaPitchedPtr( (void*)temporary.Ptr(), dims3[0]*sizeof(float), dims3[0], dims3[1] );
copyParams.dstArray = (struct cudaArray*) array;
copyParams.extent = make_cudaExtent( dims3[0], dims3[1], dims3[2] );
copyParams.kind = cudaMemcpyDeviceToDevice;
cmtkCheckCallCUDA( cudaMemcpy3D( ©Params ) );
cmtkCheckCallCUDA( cudaUnbindTexture( texRef ) );
}
|
