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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
|
/*
//
// 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: 2012 $
//
// $LastChangedDate: 2010-07-19 15:14:26 -0700 (Mon, 19 Jul 2010) $
//
// $LastChangedBy: torstenrohlfing $
//
*/
#include "cmtkDeviceHistogram_kernels.h"
#include <cstdio>
#include <cstdlib>
extern __shared__ float shared[];
__global__
void
cmtkDeviceHistogramEntropyKernel( float* result, const float *dataPtr )
{
int tx = threadIdx.x;
// first, load data into shared memory
float *working = &shared[0];
working[tx] = dataPtr[tx];
__syncthreads();
// second, compute sum of all bin values via butterfly
for ( int bit = 1; bit < blockDim.x; bit <<= 1 )
{
const float sum = working[tx] + working[tx^bit];
__syncthreads();
working[tx] = sum;
__syncthreads();
}
// third, normalize
if ( working[tx] )
{
working[tx] = dataPtr[tx] / working[tx];
}
// fourth, do p*log(p)
if ( working[tx] > 0 )
{
working[tx] *= log( working[tx] );
}
else
{
working[tx] = 0;
}
__syncthreads();
// fifth, another butterfly to compute \sum[p*log(p)]
for ( int bit = 1; bit < blockDim.x; bit <<= 1 )
{
const float sum = working[tx] + working[tx^bit];
__syncthreads();
working[tx] = sum;
__syncthreads();
}
result[tx] = -working[tx];
}
void
cmtkDeviceHistogramEntropy( float* result, const float* dataPtr, int numberOfBins )
{
dim3 dimBlock( numberOfBins, 1 );
dim3 dimGrid( 1, 1 );
// start kernel and allocate shared memory for "numberOfBins" floats
cmtkDeviceHistogramEntropyKernel<<<dimGrid,dimBlock,numberOfBins*sizeof(float)>>>( result, dataPtr );
}
__global__
void
cmtkDeviceHistogramConsolidate( float* histPtr, float* localPtr, const int numberOfBins, const int numberOfThreads )
{
int tx = threadIdx.x;
// finally, add all thread working histograms to output histogram
for ( int idx = 1+tx; idx <= numberOfBins; idx += blockDim.x )
{
float sum = 0;
for ( int hx = 0; hx < numberOfThreads; ++hx )
{
sum += localPtr[ idx + hx*(1+numberOfBins) ];
}
histPtr[idx-1] += sum;
}
}
__global__
void
cmtkDeviceHistogramPopulateKernel( float* histPtr, float* localPtr, const float *dataPtr, const float rangeFrom, const float rangeTo, const int numberOfBins, const int numberOfSamples )
{
int tx = threadIdx.x;
int offs = tx + blockDim.x * blockIdx.x;
int skip = blockDim.x * gridDim.x;
// working histogram for this thread
float* working = localPtr + (numberOfBins+1)*offs;
// start by resetting all histogram bins to 0
for ( int i = 0; i < numberOfBins; ++i )
working[i] = 0;
// populate histogram bins
const float binScale = (numberOfBins-1) / (rangeTo - rangeFrom);
for ( int offset = offs; offset < numberOfSamples; offset += skip )
{
int index = 1+truncf( fmaxf( 0, fminf( numberOfBins-1, (dataPtr[offset] - rangeFrom) * binScale ) ) ); // 1+... for consistency with masked computation; bin0 is ignored in final analysis.
++working[ index ];
}
}
__global__
void
cmtkDeviceHistogramPopulateLogKernel( float* histPtr, float* localPtr, const float *dataPtr, const float rangeFrom, const float rangeTo, const int numberOfBins, const int numberOfSamples )
{
int tx = threadIdx.x;
int offs = tx + blockDim.x * blockIdx.x;
int skip = blockDim.x * gridDim.x;
// working histogram for this thread
float* working = localPtr + (numberOfBins+1)*offs;
// start by resetting all histogram bins to 0
for ( int i = 0; i < numberOfBins; ++i )
working[i] = 0;
// populate histogram bins
const float binScale = (numberOfBins-1) / (rangeTo - rangeFrom);
const float logNumBins = log( static_cast<float>( numberOfBins ) );
for ( int offset = offs; offset < numberOfSamples; offset += skip )
{
int index = 1+truncf( (numberOfBins-1) * fmaxf( 0, fminf( 1, log((1 + dataPtr[offset]-rangeFrom)*binScale)/logNumBins ) ) );
++working[ index ];
}
}
void
cmtkDeviceHistogramPopulate( float* histPtr, const float* dataPtr, const float rangeFrom, const float rangeTo, const bool logScale, const int numberOfBins, const int numberOfSamples )
{
dim3 dimBlock( 256, 1 );
dim3 dimGrid( 16, 1 );
const size_t nThreads = 16*256;
const size_t lBytes = sizeof(float) * (numberOfBins+1) * nThreads;
float* localPtr;
if ( cudaMalloc( &localPtr, lBytes ) != cudaSuccess )
{
fprintf( stderr, "ERROR: cudaMalloc() failed with error %s\n",cudaGetErrorString( cudaGetLastError() ) );
exit( 1 );
}
if ( cudaMemset( localPtr, 0, lBytes ) != cudaSuccess )
{
fprintf( stderr, "ERROR: cudaMemset() failed with error %s\n",cudaGetErrorString( cudaGetLastError() ) );
exit( 1 );
}
if ( logScale )
cmtkDeviceHistogramPopulateLogKernel<<<dimGrid,dimBlock>>>( histPtr, localPtr, dataPtr, rangeFrom, rangeTo, numberOfBins, numberOfSamples );
else
cmtkDeviceHistogramPopulateKernel<<<dimGrid,dimBlock>>>( histPtr, localPtr, dataPtr, rangeFrom, rangeTo, numberOfBins, numberOfSamples );
cudaError_t kernelError = cudaGetLastError();
if ( kernelError != cudaSuccess )
{
fprintf( stderr, "ERROR: CUDA kernel failed with error %s\n", cudaGetErrorString( kernelError ) );
exit( 1 );
}
dim3 dimBlock2( 256, 1 );
dim3 dimGrid2( 1, 1 );
cmtkDeviceHistogramConsolidate<<<dimGrid2,dimBlock2>>>( histPtr, localPtr, numberOfBins, nThreads );
kernelError = cudaGetLastError();
if ( kernelError != cudaSuccess )
{
fprintf( stderr, "ERROR: CUDA kernel failed with error %s\n", cudaGetErrorString( kernelError ) );
exit( 1 );
}
cudaFree( localPtr );
}
__global__
void
cmtkDeviceHistogramPopulateWithMaskKernel( float* histPtr, float* localPtr, const float *dataPtr, const int *maskPtr, const float rangeFrom, const float rangeTo, const int numberOfBins, const int numberOfSamples )
{
int tx = threadIdx.x;
int offs = tx + blockDim.x * blockIdx.x;
int skip = blockDim.x * gridDim.x;
// working histogram for this thread in shared memory
float* working = localPtr + (numberOfBins+1)*offs;
// populate histogram bins
const float binScale = (numberOfBins-1) / (rangeTo - rangeFrom);
for ( int offset = offs; offset < numberOfSamples; offset += skip )
{
const float d = (dataPtr[offset] - rangeFrom) * binScale;
const float m = maskPtr[offset];
float binIndex = fmaxf( 0, fminf( numberOfBins-1, d ) );
const int index = truncf( (1+binIndex) * m );
++working[ index ];
}
}
__global__
void
cmtkDeviceHistogramPopulateLogWithMaskKernel( float* histPtr, float* localPtr, const float *dataPtr, const int *maskPtr, const float rangeFrom, const float rangeTo, const int numberOfBins, const int numberOfSamples )
{
int tx = threadIdx.x;
int offs = tx + blockDim.x * blockIdx.x;
int skip = blockDim.x * gridDim.x;
// working histogram for this thread in shared memory
float* working = localPtr + (numberOfBins+1)*offs;
// populate histogram bins
const float binScale = (numberOfBins-1) / (rangeTo - rangeFrom);
const float logNumBins = log( static_cast<float>( numberOfBins ) );
for ( int offset = offs; offset < numberOfSamples; offset += skip )
{
const float d = log((1 + dataPtr[offset]-rangeFrom)*binScale);
const float m = maskPtr[offset];
float binIndex = (numberOfBins-1) * fmaxf( 0, fminf( 1, d / logNumBins ) );
const int index = truncf( (1+binIndex) * m );
++working[ index ];
}
}
void
cmtkDeviceHistogramPopulate( float* histPtr, const float* dataPtr, const int* maskPtr, const float rangeFrom, const float rangeTo, const bool logScale, const int numberOfBins, const int numberOfSamples )
{
dim3 dimBlock( 256, 1 );
dim3 dimGrid( 16, 1 );
const size_t nThreads = 16*256;
const size_t lBytes = sizeof(float) * (numberOfBins+1) * nThreads;
float* localPtr;
if ( cudaMalloc( &localPtr, lBytes ) != cudaSuccess )
{
fprintf( stderr, "ERROR: cudaMalloc() failed with error %s\n",cudaGetErrorString( cudaGetLastError() ) );
exit( 1 );
}
if ( cudaMemset( localPtr, 0, lBytes ) != cudaSuccess )
{
fprintf( stderr, "ERROR: cudaMemset() failed with error %s\n",cudaGetErrorString( cudaGetLastError() ) );
exit( 1 );
}
if ( logScale )
cmtkDeviceHistogramPopulateLogWithMaskKernel<<<dimGrid,dimBlock>>>( histPtr, localPtr, dataPtr, maskPtr, rangeFrom, rangeTo, numberOfBins, numberOfSamples );
else
cmtkDeviceHistogramPopulateWithMaskKernel<<<dimGrid,dimBlock>>>( histPtr, localPtr, dataPtr, maskPtr, rangeFrom, rangeTo, numberOfBins, numberOfSamples );
cudaError_t kernelError = cudaGetLastError();
if ( kernelError != cudaSuccess )
{
fprintf( stderr, "ERROR: CUDA kernel failed with error %s\n", cudaGetErrorString( kernelError ) );
exit( 1 );
}
dim3 dimBlock2( 256, 1 );
dim3 dimGrid2( 1, 1 );
cmtkDeviceHistogramConsolidate<<<dimGrid2,dimBlock2>>>( histPtr, localPtr, numberOfBins, nThreads );
kernelError = cudaGetLastError();
if ( kernelError != cudaSuccess )
{
fprintf( stderr, "ERROR: CUDA kernel failed with error %s\n", cudaGetErrorString( kernelError ) );
exit( 1 );
}
cudaFree( localPtr );
}
|
