def create_pdf(self, iterable, tbins, trange, qbins, qrange, nreps=1): """Returns tuple: 1D array of channel hit counts, 3D array of (channel, time, charge) pdfs.""" first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) pdf_config = (tbins, trange, qbins, qrange) if pdf_config != self.pdf_config: self.pdf_config = pdf_config self.gpu_pdf.setup_pdf(self.detector.num_channels(), tbins, trange, qbins, qrange) else: self.gpu_pdf.clear_pdf() if nreps > 1: iterable = itertoolset.repeating_iterator(iterable, nreps) for ev in iterable: gpu_photons = gpu.GPUPhotons(ev.photons_beg) gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) self.gpu_daq.begin_acquire() self.gpu_daq.acquire(gpu_photons, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) gpu_channels = self.gpu_daq.end_acquire() self.gpu_pdf.add_hits_to_pdf(gpu_channels) return self.gpu_pdf.get_pdfs()
def eval_kernel(self, event_channels, kernel_iterable, trange, qrange, nreps=1, ndaq=1, naverage=1, time_only=True): """Returns tuple: 1D array of channel hit counts, 1D array of PDF probability densities.""" self.gpu_pdf_kernel.setup_kernel(event_channels.hit, event_channels.t, event_channels.q) first_element, kernel_iterable = itertoolset.peek(kernel_iterable) if isinstance(first_element, event.Event): kernel_iterable = \ self.photon_generator.generate_events(kernel_iterable) # Evaluate likelihood using this bandwidth for ev in kernel_iterable: gpu_photons = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps) gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) for gpu_photon_slice in gpu_photons.iterate_copies(): for idaq in xrange(ndaq): self.gpu_daq.begin_acquire() self.gpu_daq.acquire(gpu_photon_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) gpu_channels = self.gpu_daq.end_acquire() self.gpu_pdf_kernel.accumulate_kernel(gpu_channels) return self.gpu_pdf_kernel.get_kernel_eval()
def setup_kernel(self, event_channels, bandwidth_iterable, trange, qrange, nreps=1, ndaq=1, time_only=True, scale_factor=1.0): '''Call this before calling eval_pdf_kernel(). Sets up the event information and computes an appropriate kernel bandwidth''' nchannels = len(event_channels.hit) self.gpu_pdf_kernel.setup_moments(nchannels, trange, qrange, time_only=time_only) # Compute bandwidth first_element, bandwidth_iterable = itertoolset.peek(bandwidth_iterable) if isinstance(first_element, event.Event): bandwidth_iterable = \ self.photon_generator.generate_events(bandwidth_iterable) for ev in bandwidth_iterable: gpu_photons = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps) gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) for gpu_photon_slice in gpu_photons.iterate_copies(): for idaq in xrange(ndaq): self.gpu_daq.begin_acquire() self.gpu_daq.acquire(gpu_photon_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) gpu_channels = self.gpu_daq.end_acquire() self.gpu_pdf_kernel.accumulate_moments(gpu_channels) self.gpu_pdf_kernel.compute_bandwidth(event_channels.hit, event_channels.t, event_channels.q, scale_factor=scale_factor)
def simulate(self, iterable, keep_photons_beg=False, keep_photons_end=False, run_daq=True, max_steps=100 ): try: if isinstance(iterable, event.Photons): raise TypeError # Kludge because Photons looks iterable else: first_element, iterable = itertoolset.peek(iterable) except TypeError: first_element, iterable = iterable, [iterable] t_photon_start = time.time() if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) elif isinstance(first_element, GPUPhotons): print "GPU Photons" iterable = (event.Event(photons_beg=x) for x in iterable) # hacky!!! elif isinstance(first_element, event.Vertex): iterable = (event.Event(primary_vertex=vertex, vertices=[vertex]) for vertex in iterable) iterable = self.photon_generator.generate_events(iterable) t_photon_end = time.time() print "Photon Load Time: ",t_photon_end-t_photon_start," sec" for ev in iterable: photons = ev.photons_beg if isinstance(photons,event.Photons): gpu_photons = GPUPhotons(photons,cl_context=self.context) elif isinstance(photons,GPUPhotons): gpu_photons = photons ev.photons_beg = photons.get() gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, max_steps=max_steps, cl_context=self.context) ev.nphotons = len(ev.photons_beg.pos) if not keep_photons_beg: ev.photons_beg = None if keep_photons_end: ev.photons_end = gpu_photons.get() # Skip running DAQ if we don't have one if hasattr(self, 'gpu_daq') and run_daq: t_daq_start = time.time() self.gpu_daq.begin_acquire( cl_context=self.context ) self.gpu_daq.acquire(gpu_photons, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, cl_context=self.context ) gpu_channels = self.gpu_daq.end_acquire( cl_context=self.context ) ev.channels = gpu_channels.get() t_daq_end = time.time() print "DAQ readout time: ",t_daq_end-t_daq_start," sec" yield ev
def simulate(self, iterable, keep_photons_beg=False, keep_photons_end=False, keep_hits=True, keep_flat_hits=True, run_daq=False, max_steps=1000, photons_per_batch=1000000): if isinstance(iterable, event.Photons): first_element, iterable = iterable, [iterable] else: first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) elif isinstance(first_element, event.Vertex): iterable = (event.Event(vertices=[vertex]) for vertex in iterable) iterable = self.photon_generator.generate_events(iterable) nphotons = 0 batch_events = [] for ev in iterable: ev.nphotons = len(ev.photons_beg) ev.photons_beg.evidx[:] = len(batch_events) nphotons += ev.nphotons batch_events.append(ev) #FIXME need an alternate implementation to split an event that is too large if nphotons >= photons_per_batch: yield from self._simulate_batch( batch_events, keep_photons_beg=keep_photons_beg, keep_photons_end=keep_photons_end, keep_hits=keep_hits, keep_flat_hits=keep_flat_hits, run_daq=run_daq, max_steps=max_steps) nphotons = 0 batch_events = [] if len(batch_events) != 0: yield from self._simulate_batch(batch_events, keep_photons_beg=keep_photons_beg, keep_photons_end=keep_photons_end, keep_hits=keep_hits, keep_flat_hits=keep_flat_hits, run_daq=run_daq, max_steps=max_steps)
def simulate(self, iterable, keep_photons_beg=False, keep_photons_end=False, keep_hits=True, run_daq=False, max_steps=100): if isinstance(iterable, event.Photons): first_element, iterable = iterable, [iterable] else: first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) elif isinstance(first_element, event.Vertex): iterable = (event.Event(vertices=[vertex]) for vertex in iterable) iterable = self.photon_generator.generate_events(iterable) for ev in iterable: gpu_photons = gpu.GPUPhotons(ev.photons_beg) gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, max_steps=max_steps) ev.nphotons = len(ev.photons_beg.pos) if not keep_photons_beg: ev.photons_beg = None if keep_photons_end: ev.photons_end = gpu_photons.get() if hasattr(self.detector, 'num_channels') and keep_hits: ev.hits = gpu_photons.get_hits(self.gpu_geometry) # Skip running DAQ if we don't have one # Disabled by default because incredibly special-case if hasattr(self, 'gpu_daq') and run_daq: self.gpu_daq.begin_acquire() self.gpu_daq.acquire( gpu_photons, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) gpu_channels = self.gpu_daq.end_acquire() ev.channels = gpu_channels.get() yield ev
def simulate(self, iterable, keep_photons_beg=False, keep_photons_end=False, keep_hits=True, run_daq=False, max_steps=100): if isinstance(iterable, event.Photons): first_element, iterable = iterable, [iterable] else: first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) elif isinstance(first_element, event.Vertex): iterable = (event.Event(vertices=[vertex]) for vertex in iterable) iterable = self.photon_generator.generate_events(iterable) for ev in iterable: gpu_photons = gpu.GPUPhotons(ev.photons_beg) gpu_photons.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, max_steps=max_steps) ev.nphotons = len(ev.photons_beg.pos) if not keep_photons_beg: ev.photons_beg = None if keep_photons_end: ev.photons_end = gpu_photons.get() if hasattr(self.detector, 'num_channels') and keep_hits: ev.hits = gpu_photons.get_hits(self.gpu_geometry) # Skip running DAQ if we don't have one # Disabled by default because incredibly special-case if hasattr(self, 'gpu_daq') and run_daq: self.gpu_daq.begin_acquire() self.gpu_daq.acquire(gpu_photons, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) gpu_channels = self.gpu_daq.end_acquire() ev.channels = gpu_channels.get() yield ev
def eval_pdf(self, event_channels, iterable, min_twidth, trange, min_qwidth, qrange, min_bin_content=100, nreps=1, ndaq=1, nscatter=1, time_only=True): """Returns tuple: 1D array of channel hit counts, 1D array of PDF probability densities.""" ndaq_per_rep = 64 ndaq_reps = ndaq // ndaq_per_rep gpu_daq = gpu.GPUDaq(self.gpu_geometry, ndaq=ndaq_per_rep) self.gpu_pdf.setup_pdf_eval(event_channels.hit, event_channels.t, event_channels.q, min_twidth, trange, min_qwidth, qrange, min_bin_content=min_bin_content, time_only=True) first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) for ev in iterable: gpu_photons_no_scatter = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps) gpu_photons_scatter = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps * nscatter) gpu_photons_no_scatter.propagate( self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, use_weights=True, scatter_first=-1, max_steps=10) gpu_photons_scatter.propagate( self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, use_weights=True, scatter_first=1, max_steps=5) nphotons = gpu_photons_no_scatter.true_nphotons # same for scatter for i in range(gpu_photons_no_scatter.ncopies): start_photon = i * nphotons gpu_photon_no_scatter_slice = gpu_photons_no_scatter.select( event.SURFACE_DETECT, start_photon=start_photon, nphotons=nphotons) gpu_photon_scatter_slices = [ gpu_photons_scatter.select( event.SURFACE_DETECT, start_photon=(nscatter * i + j) * nphotons, nphotons=nphotons) for j in range(nscatter) ] if len(gpu_photon_no_scatter_slice) == 0: continue #weights = gpu_photon_slice.weights.get() #print 'weights', weights.min(), weights.max() for j in range(ndaq_reps): gpu_daq.begin_acquire() gpu_daq.acquire(gpu_photon_no_scatter_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) for scatter_slice in gpu_photon_scatter_slices: gpu_daq.acquire( scatter_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, weight=1.0 / nscatter) gpu_channels = gpu_daq.end_acquire() self.gpu_pdf.accumulate_pdf_eval( gpu_channels, nthreads_per_block=ndaq_per_rep) return self.gpu_pdf.get_pdf_eval()
def eval_pdf(self, event_channels, iterable, min_twidth, trange, min_qwidth, qrange, min_bin_content=100, nreps=1, ndaq=1, nscatter=1, time_only=True): """Returns tuple: 1D array of channel hit counts, 1D array of PDF probability densities.""" ndaq_per_rep = 64 ndaq_reps = ndaq // ndaq_per_rep gpu_daq = gpu.GPUDaq(self.gpu_geometry, ndaq=ndaq_per_rep) self.gpu_pdf.setup_pdf_eval(event_channels.hit, event_channels.t, event_channels.q, min_twidth, trange, min_qwidth, qrange, min_bin_content=min_bin_content, time_only=True) first_element, iterable = itertoolset.peek(iterable) if isinstance(first_element, event.Event): iterable = self.photon_generator.generate_events(iterable) elif isinstance(first_element, event.Photons): iterable = (event.Event(photons_beg=x) for x in iterable) for ev in iterable: gpu_photons_no_scatter = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps) gpu_photons_scatter = gpu.GPUPhotons(ev.photons_beg, ncopies=nreps*nscatter) gpu_photons_no_scatter.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, use_weights=True, scatter_first=-1, max_steps=10) gpu_photons_scatter.propagate(self.gpu_geometry, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, use_weights=True, scatter_first=1, max_steps=5) nphotons = gpu_photons_no_scatter.true_nphotons # same for scatter for i in xrange(gpu_photons_no_scatter.ncopies): start_photon = i * nphotons gpu_photon_no_scatter_slice = gpu_photons_no_scatter.select(event.SURFACE_DETECT, start_photon=start_photon, nphotons=nphotons) gpu_photon_scatter_slices = [gpu_photons_scatter.select(event.SURFACE_DETECT, start_photon=(nscatter*i+j)*nphotons, nphotons=nphotons) for j in xrange(nscatter)] if len(gpu_photon_no_scatter_slice) == 0: continue #weights = gpu_photon_slice.weights.get() #print 'weights', weights.min(), weights.max() for j in xrange(ndaq_reps): gpu_daq.begin_acquire() gpu_daq.acquire(gpu_photon_no_scatter_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks) for scatter_slice in gpu_photon_scatter_slices: gpu_daq.acquire(scatter_slice, self.rng_states, nthreads_per_block=self.nthreads_per_block, max_blocks=self.max_blocks, weight=1.0/nscatter) gpu_channels = gpu_daq.end_acquire() self.gpu_pdf.accumulate_pdf_eval(gpu_channels, nthreads_per_block=ndaq_per_rep) return self.gpu_pdf.get_pdf_eval()