summaryrefslogtreecommitdiffstats
path: root/cuda/2d/fan_bp.cu
blob: 0fbbaa9c5c343e4a66a5a4be79719f5833e395b5 (plain)
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
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
/*
-----------------------------------------------------------------------
Copyright: 2010-2018, iMinds-Vision Lab, University of Antwerp
           2014-2018, CWI, Amsterdam

Contact: astra@astra-toolbox.com
Website: http://www.astra-toolbox.com/

This file is part of the ASTRA Toolbox.


The ASTRA Toolbox 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 ASTRA Toolbox 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 ASTRA Toolbox. If not, see <http://www.gnu.org/licenses/>.

-----------------------------------------------------------------------
*/

#include <cstdio>
#include <cassert>
#include <iostream>

#include "util.h"
#include "arith.h"

#ifdef STANDALONE
#include "testutil.h"
#endif

#define PIXELTRACE


typedef texture<float, 2, cudaReadModeElementType> texture2D;

static texture2D gT_FanProjTexture;


namespace astraCUDA {

const unsigned int g_anglesPerBlock = 16;
const unsigned int g_blockSliceSize = 32;
const unsigned int g_blockSlices = 16;

const unsigned int g_MaxAngles = 2560;

__constant__ float gC_SrcX[g_MaxAngles];
__constant__ float gC_SrcY[g_MaxAngles];
__constant__ float gC_DetSX[g_MaxAngles];
__constant__ float gC_DetSY[g_MaxAngles];
__constant__ float gC_DetUX[g_MaxAngles];
__constant__ float gC_DetUY[g_MaxAngles];


static bool bindProjDataTexture(float* data, unsigned int pitch, unsigned int width, unsigned int height, cudaTextureAddressMode mode = cudaAddressModeBorder)
{
	cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<float>();

	gT_FanProjTexture.addressMode[0] = mode;
	gT_FanProjTexture.addressMode[1] = mode;
	gT_FanProjTexture.filterMode = cudaFilterModeLinear;
	gT_FanProjTexture.normalized = false;

	cudaBindTexture2D(0, gT_FanProjTexture, (const void*)data, channelDesc, width, height, sizeof(float)*pitch);

	// TODO: error value?

	return true;
}

__global__ void devFanBP(float* D_volData, unsigned int volPitch, unsigned int startAngle, const SDimensions dims, float fOutputScale)
{
	const int relX = threadIdx.x;
	const int relY = threadIdx.y;

	int endAngle = startAngle + g_anglesPerBlock;
	if (endAngle > dims.iProjAngles)
		endAngle = dims.iProjAngles;
	const int X = blockIdx.x * g_blockSlices + relX;
	const int Y = blockIdx.y * g_blockSliceSize + relY;

	if (X >= dims.iVolWidth || Y >= dims.iVolHeight)
		return;

	const float fX = ( X - 0.5f*dims.iVolWidth + 0.5f );
	const float fY = - ( Y - 0.5f*dims.iVolHeight + 0.5f );

	float* volData = (float*)D_volData;

	float fVal = 0.0f;
	float fA = startAngle + 0.5f;

	// TODO: Distance correction?

	for (int angle = startAngle; angle < endAngle; ++angle)
	{
		const float fSrcX = gC_SrcX[angle];
		const float fSrcY = gC_SrcY[angle];
		const float fDetSX = gC_DetSX[angle];
		const float fDetSY = gC_DetSY[angle];
		const float fDetUX = gC_DetUX[angle];
		const float fDetUY = gC_DetUY[angle];

		const float fXD = fSrcX - fX;
		const float fYD = fSrcY - fY;

		const float fNum = fDetSY * fXD - fDetSX * fYD + fX*fSrcY - fY*fSrcX;
		const float fDen = fDetUX * fYD - fDetUY * fXD;
		
		const float fT = fNum / fDen;
		fVal += tex2D(gT_FanProjTexture, fT, fA);
		fA += 1.0f;
	}

	volData[Y*volPitch+X] += fVal * fOutputScale;
}

// supersampling version
__global__ void devFanBP_SS(float* D_volData, unsigned int volPitch, unsigned int startAngle, const SDimensions dims, float fOutputScale)
{
	const int relX = threadIdx.x;
	const int relY = threadIdx.y;

	int endAngle = startAngle + g_anglesPerBlock;
	if (endAngle > dims.iProjAngles)
		endAngle = dims.iProjAngles;
	const int X = blockIdx.x * g_blockSlices + relX;
	const int Y = blockIdx.y * g_blockSliceSize + relY;

	if (X >= dims.iVolWidth || Y >= dims.iVolHeight)
		return;

	const float fXb = ( X - 0.5f*dims.iVolWidth + 0.5f - 0.5f + 0.5f/dims.iRaysPerPixelDim);
	const float fYb = - ( Y - 0.5f*dims.iVolHeight + 0.5f - 0.5f + 0.5f/dims.iRaysPerPixelDim);

	const float fSubStep = 1.0f/dims.iRaysPerPixelDim;

	float* volData = (float*)D_volData;

	fOutputScale /= (dims.iRaysPerPixelDim * dims.iRaysPerPixelDim);

	float fVal = 0.0f;
	float fA = startAngle + 0.5f;

	// TODO: Distance correction?

	for (int angle = startAngle; angle < endAngle; ++angle)
	{
		const float fSrcX = gC_SrcX[angle];
		const float fSrcY = gC_SrcY[angle];
		const float fDetSX = gC_DetSX[angle];
		const float fDetSY = gC_DetSY[angle];
		const float fDetUX = gC_DetUX[angle];
		const float fDetUY = gC_DetUY[angle];

		// TODO: Optimize these loops...
		float fX = fXb;
		for (int iSubX = 0; iSubX < dims.iRaysPerPixelDim; ++iSubX) {
			float fY = fYb;
			for (int iSubY = 0; iSubY < dims.iRaysPerPixelDim; ++iSubY) {
				const float fXD = fSrcX - fX;
				const float fYD = fSrcY - fY;

				const float fNum = fDetSY * fXD - fDetSX * fYD + fX*fSrcY - fY*fSrcX;
				const float fDen = fDetUX * fYD - fDetUY * fXD;
		
				const float fT = fNum / fDen;
				fVal += tex2D(gT_FanProjTexture, fT, fA);
				fY -= fSubStep;
			}
			fX += fSubStep;
		}
		fA += 1.0f;
	}

	volData[Y*volPitch+X] += fVal * fOutputScale;
}


// BP specifically for SART.
// It includes (free) weighting with voxel weight.
// It assumes the proj texture is set up _without_ padding, unlike regular BP.
__global__ void devFanBP_SART(float* D_volData, unsigned int volPitch, const SDimensions dims, float fOutputScale)
{
	const int relX = threadIdx.x;
	const int relY = threadIdx.y;

	const int X = blockIdx.x * g_blockSlices + relX;
	const int Y = blockIdx.y * g_blockSliceSize + relY;

	if (X >= dims.iVolWidth || Y >= dims.iVolHeight)
		return;

	const float fX = ( X - 0.5f*dims.iVolWidth + 0.5f );
	const float fY = - ( Y - 0.5f*dims.iVolHeight + 0.5f );

	float* volData = (float*)D_volData;

	// TODO: Distance correction?

	// TODO: Constant memory vs parameters.
	const float fSrcX = gC_SrcX[0];
	const float fSrcY = gC_SrcY[0];
	const float fDetSX = gC_DetSX[0];
	const float fDetSY = gC_DetSY[0];
	const float fDetUX = gC_DetUX[0];
	const float fDetUY = gC_DetUY[0];

	const float fXD = fSrcX - fX;
	const float fYD = fSrcY - fY;

	const float fNum = fDetSY * fXD - fDetSX * fYD + fX*fSrcY - fY*fSrcX;
	const float fDen = fDetUX * fYD - fDetUY * fXD;
		
	const float fT = fNum / fDen;
	const float fVal = tex2D(gT_FanProjTexture, fT, 0.5f);

	volData[Y*volPitch+X] += fVal * fOutputScale;
}

// Weighted BP for use in fan beam FBP
// Each pixel/ray is weighted by 1/L^2 where L is the distance to the source.
__global__ void devFanBP_FBPWeighted(float* D_volData, unsigned int volPitch, unsigned int startAngle, const SDimensions dims, float fOutputScale)
{
	const int relX = threadIdx.x;
	const int relY = threadIdx.y;

	int endAngle = startAngle + g_anglesPerBlock;
	if (endAngle > dims.iProjAngles)
		endAngle = dims.iProjAngles;
	const int X = blockIdx.x * g_blockSlices + relX;
	const int Y = blockIdx.y * g_blockSliceSize + relY;

	if (X >= dims.iVolWidth || Y >= dims.iVolHeight)
		return;

	const float fX = ( X - 0.5f*dims.iVolWidth + 0.5f );
	const float fY = - ( Y - 0.5f*dims.iVolHeight + 0.5f );

	float* volData = (float*)D_volData;

	float fVal = 0.0f;
	float fA = startAngle + 0.5f;

	// TODO: Distance correction?

	for (int angle = startAngle; angle < endAngle; ++angle)
	{
		const float fSrcX = gC_SrcX[angle];
		const float fSrcY = gC_SrcY[angle];
		const float fDetSX = gC_DetSX[angle];
		const float fDetSY = gC_DetSY[angle];
		const float fDetUX = gC_DetUX[angle];
		const float fDetUY = gC_DetUY[angle];

		const float fXD = fSrcX - fX;
		const float fYD = fSrcY - fY;

		const float fNum = fDetSY * fXD - fDetSX * fYD + fX*fSrcY - fY*fSrcX;
		const float fDen = fDetUX * fYD - fDetUY * fXD;

		const float fWeight = fXD*fXD + fYD*fYD;
		
		const float fT = fNum / fDen;
		fVal += tex2D(gT_FanProjTexture, fT, fA) / fWeight;
		fA += 1.0f;
	}

	volData[Y*volPitch+X] += fVal * fOutputScale;
}


bool FanBP_internal(float* D_volumeData, unsigned int volumePitch,
           float* D_projData, unsigned int projPitch,
           const SDimensions& dims, const SFanProjection* angles,
           float fOutputScale)
{
	assert(dims.iProjAngles <= g_MaxAngles);

	bindProjDataTexture(D_projData, projPitch, dims.iProjDets, dims.iProjAngles);

	// transfer angles to constant memory
	float* tmp = new float[dims.iProjAngles];

#define TRANSFER_TO_CONSTANT(name) do { for (unsigned int i = 0; i < dims.iProjAngles; ++i) tmp[i] = angles[i].f##name ; cudaMemcpyToSymbol(gC_##name, tmp, dims.iProjAngles*sizeof(float), 0, cudaMemcpyHostToDevice); } while (0)

	TRANSFER_TO_CONSTANT(SrcX);
	TRANSFER_TO_CONSTANT(SrcY);
	TRANSFER_TO_CONSTANT(DetSX);
	TRANSFER_TO_CONSTANT(DetSY);
	TRANSFER_TO_CONSTANT(DetUX);
	TRANSFER_TO_CONSTANT(DetUY);

#undef TRANSFER_TO_CONSTANT

	delete[] tmp;

	dim3 dimBlock(g_blockSlices, g_blockSliceSize);
	dim3 dimGrid((dims.iVolWidth+g_blockSlices-1)/g_blockSlices,
	             (dims.iVolHeight+g_blockSliceSize-1)/g_blockSliceSize);

	cudaStream_t stream;
	cudaStreamCreate(&stream);

	for (unsigned int i = 0; i < dims.iProjAngles; i += g_anglesPerBlock) {
		if (dims.iRaysPerPixelDim > 1)
			devFanBP_SS<<<dimGrid, dimBlock, 0, stream>>>(D_volumeData, volumePitch, i, dims, fOutputScale);
		else
			devFanBP<<<dimGrid, dimBlock, 0, stream>>>(D_volumeData, volumePitch, i, dims, fOutputScale);
	}
	cudaThreadSynchronize();

	cudaTextForceKernelsCompletion();

	cudaStreamDestroy(stream);

	return true;
}

bool FanBP_FBPWeighted_internal(float* D_volumeData, unsigned int volumePitch,
           float* D_projData, unsigned int projPitch,
           const SDimensions& dims, const SFanProjection* angles,
           float fOutputScale)
{
	assert(dims.iProjAngles <= g_MaxAngles);

	bindProjDataTexture(D_projData, projPitch, dims.iProjDets, dims.iProjAngles);

	// transfer angles to constant memory
	float* tmp = new float[dims.iProjAngles];

#define TRANSFER_TO_CONSTANT(name) do { for (unsigned int i = 0; i < dims.iProjAngles; ++i) tmp[i] = angles[i].f##name ; cudaMemcpyToSymbol(gC_##name, tmp, dims.iProjAngles*sizeof(float), 0, cudaMemcpyHostToDevice); } while (0)

	TRANSFER_TO_CONSTANT(SrcX);
	TRANSFER_TO_CONSTANT(SrcY);
	TRANSFER_TO_CONSTANT(DetSX);
	TRANSFER_TO_CONSTANT(DetSY);
	TRANSFER_TO_CONSTANT(DetUX);
	TRANSFER_TO_CONSTANT(DetUY);

#undef TRANSFER_TO_CONSTANT

	delete[] tmp;

	dim3 dimBlock(g_blockSlices, g_blockSliceSize);
	dim3 dimGrid((dims.iVolWidth+g_blockSlices-1)/g_blockSlices,
	             (dims.iVolHeight+g_blockSliceSize-1)/g_blockSliceSize);

	cudaStream_t stream;
	cudaStreamCreate(&stream);

	for (unsigned int i = 0; i < dims.iProjAngles; i += g_anglesPerBlock) {
		devFanBP_FBPWeighted<<<dimGrid, dimBlock, 0, stream>>>(D_volumeData, volumePitch, i, dims, fOutputScale);
	}
	cudaThreadSynchronize();

	cudaTextForceKernelsCompletion();

	cudaStreamDestroy(stream);

	return true;
}

// D_projData is a pointer to one padded sinogram line
bool FanBP_SART(float* D_volumeData, unsigned int volumePitch,
                float* D_projData, unsigned int projPitch,
                unsigned int angle,
                const SDimensions& dims, const SFanProjection* angles,
                float fOutputScale)
{
	// only one angle
	bindProjDataTexture(D_projData, projPitch, dims.iProjDets, 1, cudaAddressModeClamp);

	// transfer angle to constant memory
#define TRANSFER_TO_CONSTANT(name) do { cudaMemcpyToSymbol(gC_##name, &(angles[angle].f##name), sizeof(float), 0, cudaMemcpyHostToDevice); } while (0)

	TRANSFER_TO_CONSTANT(SrcX);
	TRANSFER_TO_CONSTANT(SrcY);
	TRANSFER_TO_CONSTANT(DetSX);
	TRANSFER_TO_CONSTANT(DetSY);
	TRANSFER_TO_CONSTANT(DetUX);
	TRANSFER_TO_CONSTANT(DetUY);

#undef TRANSFER_TO_CONSTANT

	dim3 dimBlock(g_blockSlices, g_blockSliceSize);
	dim3 dimGrid((dims.iVolWidth+g_blockSlices-1)/g_blockSlices,
	             (dims.iVolHeight+g_blockSliceSize-1)/g_blockSliceSize);

	devFanBP_SART<<<dimGrid, dimBlock>>>(D_volumeData, volumePitch, dims, fOutputScale);
	cudaThreadSynchronize();

	cudaTextForceKernelsCompletion();

	return true;
}

bool FanBP(float* D_volumeData, unsigned int volumePitch,
           float* D_projData, unsigned int projPitch,
           const SDimensions& dims, const SFanProjection* angles,
           float fOutputScale)
{
	for (unsigned int iAngle = 0; iAngle < dims.iProjAngles; iAngle += g_MaxAngles) {
		SDimensions subdims = dims;
		unsigned int iEndAngle = iAngle + g_MaxAngles;
		if (iEndAngle >= dims.iProjAngles)
			iEndAngle = dims.iProjAngles;
		subdims.iProjAngles = iEndAngle - iAngle;

		bool ret;
		ret = FanBP_internal(D_volumeData, volumePitch,
		                  D_projData + iAngle * projPitch, projPitch,
		                  subdims, angles + iAngle, fOutputScale);
		if (!ret)
			return false;
	}
	return true;
}

bool FanBP_FBPWeighted(float* D_volumeData, unsigned int volumePitch,
           float* D_projData, unsigned int projPitch,
           const SDimensions& dims, const SFanProjection* angles,
           float fOutputScale)
{
	for (unsigned int iAngle = 0; iAngle < dims.iProjAngles; iAngle += g_MaxAngles) {
		SDimensions subdims = dims;
		unsigned int iEndAngle = iAngle + g_MaxAngles;
		if (iEndAngle >= dims.iProjAngles)
			iEndAngle = dims.iProjAngles;
		subdims.iProjAngles = iEndAngle - iAngle;

		bool ret;
		ret = FanBP_FBPWeighted_internal(D_volumeData, volumePitch,
		                  D_projData + iAngle * projPitch, projPitch,
		                  subdims, angles + iAngle, fOutputScale);

		if (!ret)
			return false;
	}
	return true;
}


}

#ifdef STANDALONE

using namespace astraCUDA;

int main()
{
	float* D_volumeData;
	float* D_projData;

	SDimensions dims;
	dims.iVolWidth = 128;
	dims.iVolHeight = 128;
	dims.iProjAngles = 180;
	dims.iProjDets = 256;
	dims.fDetScale = 1.0f;
	dims.iRaysPerDet = 1;
	unsigned int volumePitch, projPitch;

	SFanProjection projs[180];

	projs[0].fSrcX = 0.0f;
	projs[0].fSrcY = 1536.0f;
	projs[0].fDetSX = 128.0f;
	projs[0].fDetSY = -512.0f;
	projs[0].fDetUX = -1.0f;
	projs[0].fDetUY = 0.0f;

#define ROTATE0(name,i,alpha) do { projs[i].f##name##X = projs[0].f##name##X * cos(alpha) - projs[0].f##name##Y * sin(alpha); projs[i].f##name##Y = projs[0].f##name##X * sin(alpha) + projs[0].f##name##Y * cos(alpha); } while(0)

	for (int i = 1; i < 180; ++i) {
		ROTATE0(Src, i, i*2*M_PI/180);
		ROTATE0(DetS, i, i*2*M_PI/180);
		ROTATE0(DetU, i, i*2*M_PI/180);
	}

#undef ROTATE0

	allocateVolume(D_volumeData, dims.iVolWidth, dims.iVolHeight, volumePitch);
	printf("pitch: %u\n", volumePitch);

	allocateVolume(D_projData, dims.iProjDets, dims.iProjAngles, projPitch);
	printf("pitch: %u\n", projPitch);

	unsigned int y, x;
	float* sino = loadImage("sino.png", y, x);

	float* img = new float[dims.iVolWidth*dims.iVolHeight];

	memset(img, 0, dims.iVolWidth*dims.iVolHeight*sizeof(float));

	copyVolumeToDevice(img, dims.iVolWidth, dims.iVolWidth, dims.iVolHeight, D_volumeData, volumePitch);
	copySinogramToDevice(sino, dims.iProjDets, dims.iProjDets, dims.iProjAngles, D_projData, projPitch);

	FanBP(D_volumeData, volumePitch, D_projData, projPitch, dims, projs, 1.0f);

	copyVolumeFromDevice(img, dims.iVolWidth, dims.iVolWidth, dims.iVolHeight, D_volumeData, volumePitch);

	saveImage("vol.png",dims.iVolHeight,dims.iVolWidth,img);

	return 0;
}
#endif