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
|
#!/usr/bin/env python
import numpy as np
import pycuda.driver as cuda
from pycuda import gpuarray as ga
import time
from uncertainties import ufloat
import sys
import itertools
from chroma import gpu
from chroma import camera
from chroma import event
from chroma import sample
from chroma import generator
from chroma import tools
from chroma import optics
from chroma.transform import normalize
# Generator processes need to fork BEFORE the GPU context is setup
g4generator = generator.photon.G4ParallelGenerator(4, optics.water_wcsim)
def intersect(gpu_geometry, number=100, nphotons=500000, nthreads_per_block=64,
max_blocks=1024):
"Returns the average number of ray intersections per second."
distances_gpu = ga.empty(nphotons, dtype=np.float32)
module = gpu.get_cu_module('mesh.h', options=('--use_fast_math',))
gpu_funcs = gpu.GPUFuncs(module)
run_times = []
for i in tools.progress(range(number)):
pos = ga.zeros(nphotons, dtype=ga.vec.float3)
dir = ga.to_gpu(gpu.to_float3(sample.uniform_sphere(nphotons)))
t0 = time.time()
gpu_funcs.distance_to_mesh(np.int32(pos.size), pos, dir, gpu_geometry.gpudata, distances_gpu, block=(nthreads_per_block,1,1), grid=(pos.size//nthreads_per_block+1,1))
cuda.Context.get_current().synchronize()
elapsed = time.time() - t0
if i > 0:
# first kernel call incurs some driver overhead
run_times.append(elapsed)
return nphotons/ufloat((np.mean(run_times),np.std(run_times)))
def load_photons(number=100, nphotons=500000):
"""Returns the average number of photons moved to the GPU device memory
per second."""
pos = np.zeros((nphotons,3))
dir = sample.uniform_sphere(nphotons)
pol = normalize(np.cross(sample.uniform_sphere(nphotons), dir))
wavelengths = np.random.uniform(400,800,size=nphotons)
photons = event.Photons(pos, dir, pol, wavelengths)
run_times = []
for i in tools.progress(range(number)):
t0 = time.time()
gpu_photons = gpu.GPUPhotons(photons)
cuda.Context.get_current().synchronize()
elapsed = time.time() - t0
if i > 0:
# first kernel call incurs some driver overhead
run_times.append(elapsed)
return nphotons/ufloat((np.mean(run_times),np.std(run_times)))
def propagate(gpu_geometry, number=10, nphotons=500000, nthreads_per_block=64,
max_blocks=1024):
"Returns the average number of photons propagated on the GPU per second."
rng_states = gpu.get_rng_states(nthreads_per_block*max_blocks)
run_times = []
for i in tools.progress(range(number)):
pos = np.zeros((nphotons,3))
dir = sample.uniform_sphere(nphotons)
pol = normalize(np.cross(sample.uniform_sphere(nphotons), dir))
wavelengths = np.random.uniform(400,800,size=nphotons)
photons = event.Photons(pos, dir, pol, wavelengths)
gpu_photons = gpu.GPUPhotons(photons)
t0 = time.time()
gpu_photons.propagate(gpu_geometry, rng_states, nthreads_per_block,
max_blocks)
cuda.Context.get_current().synchronize()
elapsed = time.time() - t0
if i > 0:
# first kernel call incurs some driver overhead
run_times.append(elapsed)
return nphotons/ufloat((np.mean(run_times),np.std(run_times)))
@tools.profile_if_possible
def pdf(gpu_geometry, max_pmt_id, npdfs=10, nevents=100, nreps=16, ndaq=1,
nthreads_per_block=64, max_blocks=1024):
"""
Returns the average number of 100 MeV events per second that can be
histogrammed per second.
Args:
- gpu_instance, chroma.gpu.GPU
The GPU instance passed to the GPUGeometry constructor.
- max_pmt_id, int
The channel number of the highest PMT
- npdfs, int
The number of pdf generations to average.
- nevents, int
The number of 100 MeV events to generate for each PDF.
- nreps, int
The number of times to propagate each event and add to PDF
- ndaq, int
The number of times to run the DAQ simulation on the propagated
event and add it to the PDF.
"""
rng_states = gpu.get_rng_states(nthreads_per_block*max_blocks)
gpu_daq = gpu.GPUDaq(gpu_geometry, max_pmt_id)
gpu_pdf = gpu.GPUPDF()
gpu_pdf.setup_pdf(max_pmt_id, 100, (-0.5, 999.5), 10, (-0.5, 9.5))
run_times = []
for i in tools.progress(range(npdfs)):
t0 = time.time()
gpu_pdf.clear_pdf()
vertex_gen = generator.vertex.constant_particle_gun('e-', (0,0,0),
(1,0,0), 100)
vertex_iter = itertools.islice(vertex_gen, nevents)
for ev in g4generator.generate_events(vertex_iter):
gpu_photons = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps)
gpu_photons.propagate(gpu_geometry, rng_states,
nthreads_per_block, max_blocks)
for gpu_photon_slice in gpu_photons.iterate_copies():
for idaq in xrange(ndaq):
gpu_channels = gpu_daq.acquire(gpu_photon_slice, rng_states,
nthreads_per_block, max_blocks)
gpu_pdf.add_hits_to_pdf(gpu_channels, nthreads_per_block)
hitcount, pdf = gpu_pdf.get_pdfs()
elapsed = time.time() - t0
if i > 0:
# first kernel call incurs some driver overhead
run_times.append(elapsed)
return nevents*nreps*ndaq/ufloat((np.mean(run_times),np.std(run_times)))
def pdf_eval(gpu_geometry, max_pmt_id, npdfs=10, nevents=25, nreps=16, ndaq=128,
nthreads_per_block=64, max_blocks=1024):
"""
Returns the average number of 100 MeV events that can be
histogrammed per second.
Args:
- gpu_instance, chroma.gpu.GPU
The GPU instance passed to the GPUGeometry constructor.
- max_pmt_id, int
The channel number of the highest PMT
- npdfs, int
The number of pdf generations to average.
- nevents, int
The number of 100 MeV events to generate for each PDF.
- nreps, int
The number of times to propagate each event and add to PDF
- ndaq, int
The number of times to run the DAQ simulation on the propagated
event and add it to the PDF.
"""
rng_states = gpu.get_rng_states(nthreads_per_block*max_blocks)
# Make data event
data_ev = g4generator.generate_events(itertools.islice(generator.vertex.constant_particle_gun('e-', (0,0,0),
(1,0,0), 100),
1)).next()
gpu_photons = gpu.GPUPhotons(data_ev.photons_beg)
gpu_photons.propagate(gpu_geometry, rng_states,
nthreads_per_block, max_blocks)
gpu_daq = gpu.GPUDaq(gpu_geometry, max_pmt_id)
data_ev_channels = gpu_daq.acquire(gpu_photons, rng_states, nthreads_per_block, max_blocks).get()
# Setup PDF evaluation
gpu_pdf = gpu.GPUPDF()
gpu_pdf.setup_pdf_eval(data_ev_channels.hit,
data_ev_channels.t,
data_ev_channels.q,
0.05,
(-0.5, 999.5),
1.0,
(-0.5, 20),
min_bin_content=20,
time_only=True)
run_times = []
for i in tools.progress(range(npdfs)):
t0 = time.time()
gpu_pdf.clear_pdf_eval()
vertex_gen = generator.vertex.constant_particle_gun('e-', (0,0,0),
(1,0,0), 100)
vertex_iter = itertools.islice(vertex_gen, nevents)
for ev in g4generator.generate_events(vertex_iter):
gpu_photons = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps)
gpu_photons.propagate(gpu_geometry, rng_states,
nthreads_per_block, max_blocks)
for gpu_photon_slice in gpu_photons.iterate_copies():
for idaq in xrange(ndaq):
gpu_channels = gpu_daq.acquire(gpu_photon_slice, rng_states,
nthreads_per_block, max_blocks)
gpu_pdf.accumulate_pdf_eval(gpu_channels, nthreads_per_block)
cuda.Context.get_current().synchronize()
elapsed = time.time() - t0
if i > 0:
# first kernel call incurs some driver overhead
run_times.append(elapsed)
return nevents*nreps*ndaq/ufloat((np.mean(run_times),np.std(run_times)))
if __name__ == '__main__':
from chroma import detectors
import gc
lbne = detectors.lbne()
lbne.build(bits=11)
context = gpu.create_cuda_context()
gpu_geometry = gpu.GPUGeometry(lbne)
print '%s ray intersections/sec.' % \
tools.ufloat_to_str(intersect(gpu_geometry))
# run garbage collection since there is a reference loop
# in the GPUArray class.
gc.collect()
print '%s photons loaded/sec.' % tools.ufloat_to_str(load_photons())
gc.collect()
print '%s photons propagated/sec.' % \
tools.ufloat_to_str(propagate(gpu_geometry))
gc.collect()
print '%s 100 MeV events histogrammed/s' % \
tools.ufloat_to_str(pdf(gpu_geometry, max(lbne.pmtids)))
print '%s 100 MeV events/s accumulated in PDF evaluation data structure (100 GEANT4 x 16 Chroma x 128 DAQ)' % \
tools.ufloat_to_str(pdf_eval(gpu_geometry, max(lbne.pmtids)))
context.pop()
|
