-
Notifications
You must be signed in to change notification settings - Fork 92
/
ForwardProjectorByBinParallelproj.cxx
212 lines (176 loc) · 7.21 KB
/
ForwardProjectorByBinParallelproj.cxx
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
//
//
/*!
\file
\ingroup Parallelproj
\brief non-inline implementations for stir::ForwardProjectorByBinParallelproj
\author Richard Brown
\author Kris Thielemans
*/
/*
Copyright (C) 2019, 2021 University College London
This file is part of STIR.
SPDX-License-Identifier: Apache-2.0
See STIR/LICENSE.txt for details
*/
#include "stir/recon_buildblock/Parallelproj_projector/ForwardProjectorByBinParallelproj.h"
#include "stir/recon_buildblock/Parallelproj_projector/ParallelprojHelper.h"
#include "stir/ProjDataInMemory.h"
#include "stir/RelatedViewgrams.h"
#include "stir/ProjDataInfoCylindricalNoArcCorr.h"
#include "stir/recon_buildblock/TrivialDataSymmetriesForBins.h"
#include "stir/info.h"
#include "stir/error.h"
#include "stir/recon_array_functions.h"
#ifdef parallelproj_built_with_CUDA
#include "parallelproj_cuda.h"
#else
#include "parallelproj_c.h"
#endif
START_NAMESPACE_STIR
//////////////////////////////////////////////////////////
const char * const
ForwardProjectorByBinParallelproj::registered_name =
"Parallelproj";
ForwardProjectorByBinParallelproj::ForwardProjectorByBinParallelproj() :
_cuda_verbosity(true), _use_truncation(true), _num_gpu_chunks(1)
{
this->_already_set_up = false;
this->_do_not_setup_helper = false;
}
ForwardProjectorByBinParallelproj::~ForwardProjectorByBinParallelproj()
{
}
void
ForwardProjectorByBinParallelproj::
initialise_keymap()
{
parser.add_start_key("Forward Projector Using Parallelproj Parameters");
parser.add_stop_key("End Forward Projector Using Parallelproj Parameters");
parser.add_key("verbosity", &_cuda_verbosity);
parser.add_key("num_gpu_chunks", &_num_gpu_chunks);
}
void
ForwardProjectorByBinParallelproj::
set_defaults()
{
_cuda_verbosity = true;
_use_truncation = true;
_num_gpu_chunks = 1;
}
void
ForwardProjectorByBinParallelproj::set_helper(shared_ptr<detail::ParallelprojHelper> helper)
{
this->_helper = helper;
this->_do_not_setup_helper = true;
}
void
ForwardProjectorByBinParallelproj::
set_up(const shared_ptr<const ProjDataInfo>& proj_data_info_sptr,
const shared_ptr<const DiscretisedDensity<3,float> >& density_info_sptr)
{
ForwardProjectorByBin::set_up(proj_data_info_sptr,density_info_sptr);
check(*proj_data_info_sptr, *_density_sptr);
_symmetries_sptr.reset(new TrivialDataSymmetriesForBins(proj_data_info_sptr));
#if 0
shared_ptr<const ProjDataInfoCylindricalNoArcCorr>
proj_data_info_cy_no_ar_cor_sptr(
dynamic_pointer_cast<const ProjDataInfoCylindricalNoArcCorr>(
proj_data_info_sptr));
if (is_null_ptr(proj_data_info_cy_no_ar_cor_sptr))
error("ForwardProjectorByBinParallelproj: Failed casting to ProjDataInfoCylindricalNoArcCorr");
#endif
// Initialise projected_data_sptr from this->_proj_data_info_sptr
_projected_data_sptr.reset(
new ProjDataInMemory(this->_density_sptr->get_exam_info_sptr(), proj_data_info_sptr));
if (!this->_do_not_setup_helper)
_helper = std::make_shared<detail::ParallelprojHelper>(*proj_data_info_sptr, *density_info_sptr);
}
const DataSymmetriesForViewSegmentNumbers *
ForwardProjectorByBinParallelproj::
get_symmetries_used() const
{
if (!this->_already_set_up)
error("ForwardProjectorByBin method called without calling set_up first.");
return _symmetries_sptr.get();
}
void
ForwardProjectorByBinParallelproj::
actual_forward_project(RelatedViewgrams<float>& viewgrams,
const int min_axial_pos_num, const int max_axial_pos_num,
const int min_tangential_pos_num, const int max_tangential_pos_num)
{
if ((min_axial_pos_num != this->_proj_data_info_sptr->get_min_axial_pos_num(viewgrams.get_basic_segment_num())) ||
(max_axial_pos_num != this->_proj_data_info_sptr->get_max_axial_pos_num(viewgrams.get_basic_segment_num())) ||
(min_tangential_pos_num != this->_proj_data_info_sptr->get_min_tangential_pos_num()) ||
(max_tangential_pos_num != this->_proj_data_info_sptr->get_max_tangential_pos_num()))
error("STIR wrapping of Parallelproj projectors current only handles projecting all data");
viewgrams = _projected_data_sptr->get_related_viewgrams(
viewgrams.get_basic_view_segment_num(), _symmetries_sptr);
}
void
ForwardProjectorByBinParallelproj::
set_input(const DiscretisedDensity<3,float> & density)
{
ForwardProjectorByBin::set_input(density);
// Before forward projection, we enforce a truncation outside of the FOV.
// This is because the parallelproj projector seems to have some trouble at the edges and this
// could cause some voxel values to spiral out of control.
//if (_use_truncation)
{
const float radius = this->_proj_data_info_sptr->get_scanner_sptr()->get_inner_ring_radius();
const float image_radius = _helper->voxsize[2]*_helper->imgdim[2]/2;
truncate_rim(*_density_sptr, static_cast<int>(std::max((image_radius-radius) / _helper->voxsize[2],0.F)));
}
std::vector<float> image_vec(density.size_all());
std::copy(_density_sptr->begin_all(), _density_sptr->end_all(), image_vec.begin());
#if 0
// needed to set output to zero as parallelproj accumulates but is no longer the case
_projected_data_sptr->fill(0.F);
#endif
info("Calling parallelproj forward",2);
#ifdef parallelproj_built_with_CUDA
long long num_image_voxel = static_cast<long long>(image_vec.size());
long long num_lors = static_cast<long long>(_projected_data_sptr->get_proj_data_info_sptr()->size_all());
long long num_lors_per_chunk_floor = num_lors / _num_gpu_chunks;
long long remainder = num_lors % _num_gpu_chunks;
long long num_lors_per_chunk;
long long offset = 0;
// send image to all visible CUDA devices
float** image_on_cuda_devices;
image_on_cuda_devices = copy_float_array_to_all_devices(image_vec.data(), num_image_voxel);
// do (chuck-wise) projection on the CUDA devices
for(int chunk_num = 0; chunk_num < _num_gpu_chunks; chunk_num++){
if(chunk_num < remainder){
num_lors_per_chunk = num_lors_per_chunk_floor + 1;
}
else{
num_lors_per_chunk = num_lors_per_chunk_floor;
}
joseph3d_fwd_cuda(_helper->xstart.data() + 3*offset,
_helper->xend.data() + 3*offset,
image_on_cuda_devices,
_helper->origin.data(),
_helper->voxsize.data(),
_projected_data_sptr->get_data_ptr() + offset,
num_lors_per_chunk,
_helper->imgdim.data(),
/*threadsperblock*/ 64);
offset += num_lors_per_chunk;
}
// free image array from CUDA devices
free_float_array_on_all_devices(image_on_cuda_devices);
#else
joseph3d_fwd(_helper->xstart.data(),
_helper->xend.data(),
image_vec.data(),
_helper->origin.data(),
_helper->voxsize.data(),
_projected_data_sptr->get_data_ptr(),
static_cast<long long>(_projected_data_sptr->get_proj_data_info_sptr()->size_all()),
_helper->imgdim.data());
#endif
info("done", 2);
_projected_data_sptr->release_data_ptr();
}
END_NAMESPACE_STIR