bioem.cpp 28.3 KB
Newer Older
Pilar Cossio's avatar
License  
Pilar Cossio committed
1 2 3 4
/* ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
        < BioEM software for Bayesian inference of Electron Microscopy images>
            Copyright (C) 2014 Pilar Cossio, David Rohr and Gerhard Hummer.
            Max Planck Institute of Biophysics, Frankfurt, Germany.
5

Pilar Cossio's avatar
License  
Pilar Cossio committed
6 7 8 9
                See license statement for terms of distribution.

   ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/

10 11 12 13 14 15 16 17
#include <mpi.h>

#define MPI_CHK(expr) \
	if (expr != MPI_SUCCESS) \
	{ \
		fprintf(stderr, "Error in MPI function %s: %d\n", __FILE__, __LINE__); \
	}

18 19 20 21 22 23 24 25 26
#include <fstream>
#include <boost/program_options.hpp>
#include <iostream>
#include <algorithm>
#include <iterator>
#include <stdio.h>
#include <stdlib.h>
#include <string>
#include <cmath>
27

28
#ifdef WITH_OPENMP
29
#include <omp.h>
30
#endif
31 32 33 34 35 36 37 38 39 40

#include <fftw3.h>
#include <math.h>
#include "cmodules/timer.h"

#include "param.h"
#include "bioem.h"
#include "model.h"
#include "map.h"

41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
#ifdef BIOEM_USE_NVTX
#include "nvToolsExt.h"

const uint32_t colors[] = { 0x0000ff00, 0x000000ff, 0x00ffff00, 0x00ff00ff, 0x0000ffff, 0x00ff0000, 0x00ffffff };
const int num_colors = sizeof(colors)/sizeof(colors[0]);

#define cuda_custom_timeslot(name,cid) { \
	int color_id = cid; \
	color_id = color_id%num_colors;\
	nvtxEventAttributes_t eventAttrib = {0}; \
	eventAttrib.version = NVTX_VERSION; \
	eventAttrib.size = NVTX_EVENT_ATTRIB_STRUCT_SIZE; \
	eventAttrib.colorType = NVTX_COLOR_ARGB; \
	eventAttrib.color = colors[color_id]; \
	eventAttrib.messageType = NVTX_MESSAGE_TYPE_ASCII; \
	eventAttrib.message.ascii = name; \
	nvtxRangePushEx(&eventAttrib); \
}
#define cuda_custom_timeslot_end nvtxRangePop();
#else
#define cuda_custom_timeslot(name,cid)
#define cuda_custom_timeslot_end
#endif
64

65 66 67 68 69 70 71 72 73 74 75
#include "bioem_algorithm.h"

using namespace boost;
namespace po = boost::program_options;

using namespace std;

// A helper function of Boost
template<class T>
ostream& operator<<(ostream& os, const vector<T>& v)
{
76 77
	copy(v.begin(), v.end(), ostream_iterator<T>(os, " "));
	return os;
78 79 80 81
}

bioem::bioem()
{
82
	FFTAlgo = getenv("FFTALGO") == NULL ? 1 : atoi(getenv("FFTALGO"));
83
	DebugOutput = getenv("BIOEM_DEBUG_OUTPUT") == NULL ? 2 : atoi(getenv("BIOEM_DEBUG_OUTPUT"));
84 85 86 87 88 89 90 91
}

bioem::~bioem()
{
}

int bioem::configure(int ac, char* av[])
{
David Rohr's avatar
David Rohr committed
92 93 94 95 96
	// **************************************************************************************
	// **** Configuration Routine using boost for extracting parameters, models and maps ****
	// **************************************************************************************
	// ****** And Precalculating necessary grids, map crosscorrelations and kernels  ********
	// *************************************************************************************
97

David Rohr's avatar
David Rohr committed
98 99 100 101 102
	if (mpi_rank == 0)
	{
		// *** Inizialzing default variables ***
		std::string infile, modelfile, mapfile;
		Model.readPDB = false;
103
		param.param_device.writeAngles = false;
David Rohr's avatar
David Rohr committed
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
		param.dumpMap = false;
		param.loadMap = false;
		RefMap.readMRC = false;
		RefMap.readMultMRC = false;

		// *************************************************************************************
		cout << " ++++++++++++ FROM COMMAND LINE +++++++++++\n\n";
		// *************************************************************************************

		// ********************* Command line reading input with BOOST ************************

		try {
			po::options_description desc("Command line inputs");
			desc.add_options()
			("Inputfile", po::value<std::string>(), "(Mandatory) Name of input parameter file")
			("Modelfile", po::value< std::string>() , "(Mandatory) Name of model file")
			("Particlesfile", po::value< std::string>(), "(Mandatory) Name of paricles file")
			("ReadPDB", "(Optional) If reading model file in PDB format")
			("ReadMRC", "(Optional) If reading particle file in MRC format")
			("ReadMultipleMRC", "(Optional) If reading Multiple MRCs")
			("DumpMaps", "(Optional) Dump maps after they were red from maps file")
			("LoadMapDump", "(Optional) Read Maps from dump instead of maps file")
			("help", "(Optional) Produce help message")
			;


			po::positional_options_description p;
			p.add("Inputfile", -1);
			p.add("Modelfile", -1);
			p.add("Particlesfile", -1);
			p.add("ReadPDB", -1);
			p.add("ReadMRC", -1);
			p.add("ReadMultipleMRC", -1);
			p.add("DumpMaps", -1);
			p.add("LoadMapDump", -1);

			po::variables_map vm;
			po::store(po::command_line_parser(ac, av).
					  options(desc).positional(p).run(), vm);
			po::notify(vm);

			if((ac < 6)) {
				std::cout << desc << std::endl;
				return 1;
			}
			if (vm.count("help")) {
				cout << "Usage: options_description [options]\n";
				cout << desc;
				return 1;
			}

			if (vm.count("Inputfile"))
			{
				cout << "Input file is: ";
				cout << vm["Inputfile"].as< std::string >() << "\n";
				infile = vm["Inputfile"].as< std::string >();
			}
			if (vm.count("Modelfile"))
			{
				cout << "Model file is: "
					 << vm["Modelfile"].as<  std::string  >() << "\n";
				modelfile = vm["Modelfile"].as<  std::string  >();
			}

			if (vm.count("ReadPDB"))
			{
				cout << "Reading model file in PDB format.\n";
				Model.readPDB = true;
			}

			if (vm.count("ReadMRC"))
			{
				cout << "Reading particle file in MRC format.\n";
				RefMap.readMRC=true;
			}

			if (vm.count("ReadMultipleMRC"))
			{
				cout << "Reading Multiple MRCs.\n";
				RefMap.readMultMRC=true;
			}

			if (vm.count("DumpMaps"))
			{
				cout << "Dumping Maps after reading from file.\n";
				param.dumpMap = true;
			}

			if (vm.count("LoadMapDump"))
			{
				cout << "Loading Map dump.\n";
				param.loadMap = true;
			}

			if (vm.count("Particlesfile"))
			{
				cout << "Paricle file is: "
					 << vm["Particlesfile"].as< std::string >() << "\n";
				mapfile = vm["Particlesfile"].as< std::string >();
			}
David Rohr's avatar
David Rohr committed
204
		}
David Rohr's avatar
David Rohr committed
205
		catch(std::exception& e)
206
		{
David Rohr's avatar
David Rohr committed
207 208
			cout << e.what() << "\n";
			return 1;
209
		}
David Rohr's avatar
David Rohr committed
210 211 212 213 214 215
			//check for consitency in multiple MRCs
			if(  RefMap.readMultMRC && not(RefMap.readMRC) ){
			 cout << "For Multiple MRCs command --ReadMRC is necesary too";
			 exit(1);
			}
		// ********************* Reading Parameter Input ***************************
216
		param.readParameters(infile.c_str());
David Rohr's avatar
David Rohr committed
217 218

		// ********************* Reading Model Input ******************************
219
		Model.readModel(modelfile.c_str());
David Rohr's avatar
David Rohr committed
220 221

		// ********************* Reading Particle Maps Input **********************
222
		RefMap.readRefMaps(param, mapfile.c_str());
David Rohr's avatar
David Rohr committed
223
	}
224

David Rohr's avatar
David Rohr committed
225
#ifdef WITH_MPI
226 227 228 229
	MPI_Bcast(&param, sizeof(param), MPI_BYTE, 0, MPI_COMM_WORLD);
	//refCtf, CtfParam, angles automatically filled by precalculare function below
	
	MPI_Bcast(&Model, sizeof(Model), MPI_BYTE, 0, MPI_COMM_WORLD);
David Rohr's avatar
David Rohr committed
230
	if (mpi_rank != 0) Model.points = (bioem_model::bioem_model_point*) mallocchk(sizeof(bioem_model::bioem_model_point) * Model.nPointsModel);
231 232
	MPI_Bcast(Model.points, sizeof(bioem_model::bioem_model_point) * Model.nPointsModel, MPI_BYTE, 0, MPI_COMM_WORLD);
	
David Rohr's avatar
David Rohr committed
233 234 235
	MPI_Bcast(&RefMap, sizeof(RefMap), MPI_BYTE, 0, MPI_COMM_WORLD);
	if (mpi_rank != 0) RefMap.maps = (myfloat_t*) mallocchk(RefMap.refMapSize * sizeof(myfloat_t) * RefMap.ntotRefMap);
	MPI_Bcast(RefMap.maps, RefMap.refMapSize * sizeof(myfloat_t) * RefMap.ntotRefMap, MPI_BYTE, 0, MPI_COMM_WORLD);
David Rohr's avatar
David Rohr committed
236
#endif
237

David Rohr's avatar
David Rohr committed
238
	// ****************** Precalculating Necessary Stuff *********************
239
	param.PrepareFFTs();
240
	precalculate();
David Rohr's avatar
David Rohr committed
241

242 243
	if (getenv("BIOEM_DEBUG_BREAK"))
	{
244 245 246
		const int cut = atoi(getenv("BIOEM_DEBUG_BREAK"));
		if (param.nTotGridAngles > cut) param.nTotGridAngles = cut;
		if (param.nTotCTFs > cut) param.nTotCTFs = cut;
247
	}
David Rohr's avatar
David Rohr committed
248

249
	pProb.init(RefMap.ntotRefMap, param.nTotGridAngles, *this);
250

251 252
	deviceInit();

253
	return(0);
254 255
}

256 257 258
void bioem::cleanup()
{
	//Deleting allocated pointers
259
	free_device_host(pProb.ptr);
260 261 262
	RefMap.freePointers();
}

263 264
int bioem::precalculate()
{
David Rohr's avatar
David Rohr committed
265
	// **************************************************************************************
David Rohr's avatar
David Rohr committed
266
	// **Precalculating Routine of Orientation grids, Map crosscorrelations and CTF Kernels**
David Rohr's avatar
David Rohr committed
267
	// **************************************************************************************
268

269 270
	// Generating Grids of orientations
	param.CalculateGridsParam();
271

272 273
	// Precalculating CTF Kernels stored in class Param
	param.CalculateRefCTF();
274

275 276
	//Precalculate Maps
	RefMap.precalculate(param, *this);
277

278
	return(0);
279 280 281 282
}

int bioem::run()
{
David Rohr's avatar
David Rohr committed
283 284 285
	// **************************************************************************************
	// **** Main BioEM routine, projects, convolutes and compares with Map using OpenMP ****
	// **************************************************************************************
286 287 288 289 290 291 292
	
	if (param.printModel)
	{
		
		//....
		return(0);
	}
293

David Rohr's avatar
David Rohr committed
294 295
	// **** If we want to control the number of threads -> omp_set_num_threads(XX); ******
	// ****************** Declarying class of Probability Pointer  *************************
296

David Rohr's avatar
David Rohr committed
297
	if (mpi_rank == 0) printf("\tInitializing Probabilities\n");
298 299 300
	// Inizialzing Probabilites to zero and constant to -Infinity
	for (int iRefMap = 0; iRefMap < RefMap.ntotRefMap; iRefMap ++)
	{
301 302 303 304
		bioem_Probability_map& pProbMap = pProb.getProbMap(iRefMap);

		pProbMap.Total = 0.0;
		pProbMap.Constoadd = -9999999;
305
		if (param.param_device.writeAngles)
306
		{
307 308 309
			for (int iOrient = 0; iOrient < param.nTotGridAngles; iOrient ++)
			{
				bioem_Probability_angle& pProbAngle = pProb.getProbAngle(iRefMap, iOrient);
310

311 312 313
				pProbAngle.forAngles = 0.0;
				pProbAngle.ConstAngle = -99999999;
			}
314 315
		}
	}
David Rohr's avatar
David Rohr committed
316
	// **************************************************************************************
317
	deviceStartRun();
318 319 320 321 322 323 324 325 326 327
	{
		const int count = omp_get_max_threads();
		localCCT = new mycomplex_t*[count];
		lCC = new myfloat_t*[count];
		for (int i = 0;i < count;i++)
		{
			localCCT[i] = (mycomplex_t *) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
			lCC[i] = (myfloat_t *) myfftw_malloc(sizeof(myfloat_t) * param.param_device.NumberPixels * param.param_device.NumberPixels);
		}
	}
328

David Rohr's avatar
David Rohr committed
329
	// ******************************** MAIN CYCLE ******************************************
David Rohr's avatar
David Rohr committed
330

David Rohr's avatar
David Rohr committed
331
	// *** Declaring Private variables for each thread *****
332
	mycomplex_t* proj_mapFFT;
333
	myfloat_t* conv_map = new myfloat_t[param.param_device.NumberPixels * param.param_device.NumberPixels];
334
	mycomplex_t* conv_mapFFT;
335
	myfloat_t sumCONV, sumsquareCONV;
336 337

	//allocating fftw_complex vector
338 339
	proj_mapFFT = (mycomplex_t *) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
	conv_mapFFT = (mycomplex_t *) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
340 341 342

	HighResTimer timer;

David Rohr's avatar
David Rohr committed
343 344 345 346 347 348 349
	if (DebugOutput >= 1 && mpi_rank == 0) printf("\tMain Loop (GridAngles %d, CTFs %d, RefMaps %d, Shifts (%d/%d)²), Pixels %d²\n", param.nTotGridAngles, param.nTotCTFs, RefMap.ntotRefMap, 2 * param.param_device.maxDisplaceCenter + param.param_device.GridSpaceCenter, param.param_device.GridSpaceCenter, param.param_device.NumberPixels);

	const int iOrientStart = (int) ((long long int) mpi_rank * param.nTotGridAngles / mpi_size);
	int iOrientEnd = (int) ((long long int) (mpi_rank + 1) * param.nTotGridAngles / mpi_size);
	if (iOrientEnd > param.nTotGridAngles) iOrientEnd = param.nTotGridAngles;
	
	for (int iOrient = iOrientStart; iOrient < iOrientEnd; iOrient++)
350
	{
David Rohr's avatar
David Rohr committed
351 352
		// ***************************************************************************************
		// ***** Creating Projection for given orientation and transforming to Fourier space *****
353
		if (DebugOutput >= 1) timer.ResetStart();
354
		createProjection(iOrient, proj_mapFFT);
355
		if (DebugOutput >= 1) printf("Time Projection %d: %f\n", iOrient, timer.GetCurrentElapsedTime());
356

David Rohr's avatar
David Rohr committed
357 358
		// ***************************************************************************************
		// ***** **** Internal Loop over convolutions **** *****
359 360
		for (int iConv = 0; iConv < param.nTotCTFs; iConv++)
		{
David Rohr's avatar
David Rohr committed
361
			// *** Calculating convolutions of projection map and crosscorrelations ***
362

363
			if (DebugOutput >= 2) timer.ResetStart();
364
			createConvolutedProjectionMap(iOrient, iConv, proj_mapFFT, conv_map, conv_mapFFT, sumCONV, sumsquareCONV);
365
			if (DebugOutput >= 2) printf("Time Convolution %d %d: %f\n", iOrient, iConv, timer.GetCurrentElapsedTime());
366

David Rohr's avatar
David Rohr committed
367 368
			// ***************************************************************************************
			// *** Comparing each calculated convoluted map with all experimental maps ***
369
			if (DebugOutput >= 2) timer.ResetStart();
370
			compareRefMaps(iOrient, iConv, conv_map, conv_mapFFT, sumCONV, sumsquareCONV);
371

372 373 374 375 376 377 378 379 380 381 382 383
			if (DebugOutput >= 2)
			{
				const double compTime = timer.GetCurrentElapsedTime();
				const int nShifts = 2 * param.param_device.maxDisplaceCenter / param.param_device.GridSpaceCenter + 1;
				const double nFlops = (double) RefMap.ntotRefMap * (double) nShifts * (double) nShifts *
									  (((double) param.param_device.NumberPixels - (double) param.param_device.maxDisplaceCenter / 2.) * ((double) param.param_device.NumberPixels - (double) param.param_device.maxDisplaceCenter / 2.) * 5. + 25.) / compTime;
				const double nGBs = (double) RefMap.ntotRefMap * (double) nShifts * (double) nShifts *
									(((double) param.param_device.NumberPixels - (double) param.param_device.maxDisplaceCenter / 2.) * ((double) param.param_device.NumberPixels - (double) param.param_device.maxDisplaceCenter / 2.) * 2. + 8.) * (double) sizeof(myfloat_t) / compTime;
				const double nGBs2 = (double) RefMap.ntotRefMap * ((double) param.param_device.NumberPixels * (double) param.param_device.NumberPixels + 8.) * (double) sizeof(myfloat_t) / compTime;

				printf("Time Comparison %d %d: %f sec (%f GFlops, %f GB/s (cached), %f GB/s)\n", iOrient, iConv, compTime, nFlops / 1000000000., nGBs / 1000000000., nGBs2 / 1000000000.);
			}
384 385 386 387 388
		}
	}
	//deallocating fftw_complex vector
	myfftw_free(proj_mapFFT);
	myfftw_free(conv_mapFFT);
389
	delete[] conv_map;
David Rohr's avatar
David Rohr committed
390

391
	deviceFinishRun();
392 393 394 395 396 397 398 399 400 401
	{
		const int count = omp_get_max_threads();
		for (int i = 0;i < count;i++)
		{
			myfftw_free(localCCT[i]);
			myfftw_free(lCC[i]);
		}
		delete[] localCCT;
		delete[] lCC;
	}
402

David Rohr's avatar
David Rohr committed
403
	// ************* Writing Out Probabilities ***************
404

David Rohr's avatar
David Rohr committed
405
	// *** Angular Probability ***
David Rohr's avatar
David Rohr committed
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424
	
#ifdef WITH_MPI
	if (DebugOutput >= 2 && mpi_rank == 0) timer.ResetStart();
	//Reduce Constant and summarize probabilities
	{
		myfloat_t* tmp1 = new myfloat_t[RefMap.ntotRefMap];
		myfloat_t* tmp2 = new myfloat_t[RefMap.ntotRefMap];
		myfloat_t* tmp3 = new myfloat_t[RefMap.ntotRefMap];
		for (int i = 0;i < RefMap.ntotRefMap;i++)
		{
				tmp1[i] = pProb.getProbMap(i).Constoadd;
		}
		MPI_Allreduce(tmp1, tmp2, RefMap.ntotRefMap, MY_MPI_FLOAT, MPI_MAX, MPI_COMM_WORLD);
		for (int i = 0;i < RefMap.ntotRefMap;i++)
		{
			bioem_Probability_map& pProbMap = pProb.getProbMap(i);
			tmp1[i] = pProbMap.Total * exp(pProbMap.Constoadd - tmp2[i]);
		}
		MPI_Reduce(tmp1, tmp3, RefMap.ntotRefMap, MY_MPI_FLOAT, MPI_SUM, 0, MPI_COMM_WORLD);
425

David Rohr's avatar
David Rohr committed
426
		//Find MaxProb
427
		MPI_Status mpistatus;
David Rohr's avatar
David Rohr committed
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446
		{	
			int* tmpi1 = new int[RefMap.ntotRefMap];
			int* tmpi2 = new int[RefMap.ntotRefMap];
			for (int i = 0;i < RefMap.ntotRefMap;i++)
			{
				bioem_Probability_map& pProbMap = pProb.getProbMap(i);
				tmpi1[i] = tmp2[i] <= pProbMap.Constoadd ? mpi_rank : -1;
			}
			MPI_Allreduce(tmpi1, tmpi2, RefMap.ntotRefMap, MPI_INT, MPI_MAX, MPI_COMM_WORLD);
			for (int i = 0;i < RefMap.ntotRefMap;i++)
			{
				if (tmpi2[i] == -1)
				{
					if (mpi_rank == 0) printf("Error: Could not find highest probability\n");
				}
				else if (tmpi2[i] != 0) //Skip if rank 0 already has highest probability
				{
					if (mpi_rank == 0)
					{
447
						MPI_Recv(&pProb.getProbMap(i).max, sizeof(pProb.getProbMap(i).max), MPI_BYTE, tmpi2[i], i, MPI_COMM_WORLD, &mpistatus);
David Rohr's avatar
David Rohr committed
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
					}
					else if (mpi_rank == tmpi2[i])
					{
						MPI_Send(&pProb.getProbMap(i).max, sizeof(pProb.getProbMap(i).max), MPI_BYTE, 0, i, MPI_COMM_WORLD);
					}
				}
			}
			delete[] tmpi1;
			delete[] tmpi2;
		}
		
		if (mpi_rank == 0)
		{
			for (int i = 0;i < RefMap.ntotRefMap;i++)
			{
					bioem_Probability_map& pProbMap = pProb.getProbMap(i);
					pProbMap.Total = tmp3[i];
					pProbMap.Constoadd = tmp2[i];
			}
		}
		
		delete[] tmp1;
		delete[] tmp2;
		delete[] tmp3;
		if (DebugOutput >= 2 && mpi_rank == 0) printf("Time MPI Reduction: %f\n", timer.GetCurrentElapsedTime());
	}
	
	//Angle Reduction and Probability summation for individual angles
	if (param.param_device.writeAngles)
477
	{
David Rohr's avatar
David Rohr committed
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
		const int count = RefMap.ntotRefMap * param.nTotGridAngles;
		myfloat_t* tmp1 = new myfloat_t[count];
		myfloat_t* tmp2 = new myfloat_t[count];
		myfloat_t* tmp3 = new myfloat_t[count];
		for (int i = 0;i < RefMap.ntotRefMap;i++)
		{
				tmp1[i] = pProb.getProbMap(i).Constoadd;
		}
		MPI_Allreduce(tmp1, tmp2, count, MY_MPI_FLOAT, MPI_MAX, MPI_COMM_WORLD);
		for (int i = 0;i < RefMap.ntotRefMap;i++)
		{
			for (int j = 0;j < param.nTotGridAngles;j++)
			{
				bioem_Probability_angle& pProbAngle = pProb.getProbAngle(i, j);
				tmp1[i * param.nTotGridAngles + j] = pProbAngle.forAngles * exp(pProbAngle.ConstAngle - tmp2[i * param.nTotGridAngles + j]);
			}
		}
		MPI_Reduce(tmp1, tmp3, count, MY_MPI_FLOAT, MPI_SUM, 0, MPI_COMM_WORLD);
		if (mpi_rank == 0)
		{
			for (int i = 0;i < RefMap.ntotRefMap;i++)
			{
				for (int j = 0;j < param.nTotGridAngles;j++)
				{
					bioem_Probability_angle& pProbAngle = pProb.getProbAngle(i, j);
					pProbAngle.forAngles = tmp3[i * param.nTotGridAngles + j];
					pProbAngle.ConstAngle = tmp2[i * param.nTotGridAngles + j];
				}
			}
		}
		delete[] tmp1;
		delete[] tmp2;
		delete[] tmp3;	
511
	}
David Rohr's avatar
David Rohr committed
512
#endif
513

David Rohr's avatar
David Rohr committed
514
	if (mpi_rank == 0)
515
	{
David Rohr's avatar
David Rohr committed
516 517 518 519 520 521 522 523 524 525 526 527
		ofstream angProbfile;
		if(param.param_device.writeAngles)
		{
			angProbfile.open ("ANG_PROB");
		}

		ofstream outputProbFile;
		outputProbFile.open ("Output_Probabilities");
		for (int iRefMap = 0; iRefMap < RefMap.ntotRefMap; iRefMap ++)
		{
			// **** Total Probability ***
			bioem_Probability_map& pProbMap = pProb.getProbMap(iRefMap);
528

David Rohr's avatar
David Rohr committed
529
			outputProbFile << "RefMap " << iRefMap << " Probability  "  << log(pProbMap.Total) + pProbMap.Constoadd + 0.5 * log(M_PI) + (1 - param.param_device.Ntotpi * 0.5)*(log(2 * M_PI) + 1) + log(param.param_device.volu) << " Constant " << pProbMap.Constoadd  << "\n";
530

David Rohr's avatar
David Rohr committed
531
			outputProbFile << "RefMap " << iRefMap << " Maximizing Param: ";
532

David Rohr's avatar
David Rohr committed
533 534 535 536 537 538 539 540 541 542 543
			// *** Param that maximize probability****
			outputProbFile << (pProbMap.Constoadd + 0.5 * log(M_PI) + (1 - param.param_device.Ntotpi * 0.5) * (log(2 * M_PI) + 1) + log(param.param_device.volu)) << " ";
			outputProbFile << param.angles[pProbMap.max.max_prob_orient].pos[0] << " ";
			outputProbFile << param.angles[pProbMap.max.max_prob_orient].pos[1] << " ";
			outputProbFile << param.angles[pProbMap.max.max_prob_orient].pos[2] << " ";
			outputProbFile << param.CtfParam[pProbMap.max.max_prob_conv].pos[0] << " ";
			outputProbFile << param.CtfParam[pProbMap.max.max_prob_conv].pos[1] << " ";
			outputProbFile << param.CtfParam[pProbMap.max.max_prob_conv].pos[2] << " ";
			outputProbFile << pProbMap.max.max_prob_cent_x << " ";
			outputProbFile << pProbMap.max.max_prob_cent_y;
			outputProbFile << "\n";
544

David Rohr's avatar
David Rohr committed
545
			// *** For individual files*** //angProbfile.open ("ANG_PROB_"iRefMap);
546

David Rohr's avatar
David Rohr committed
547
			if(param.param_device.writeAngles)
548
			{
David Rohr's avatar
David Rohr committed
549 550 551
				for (int iOrient = 0; iOrient < param.nTotGridAngles; iOrient++)
				{
					bioem_Probability_angle& pProbAngle = pProb.getProbAngle(iRefMap, iOrient);
552

David Rohr's avatar
David Rohr committed
553 554
					angProbfile << " " << iRefMap << " " << param.angles[iOrient].pos[0] << " " << param.angles[iOrient].pos[1] << " " << param.angles[iOrient].pos[2] << " " << log(pProbAngle.forAngles) + pProbAngle.ConstAngle + 0.5 * log(M_PI) + (1 - param.param_device.Ntotpi * 0.5)*(log(2 * M_PI) + 1) + log(param.param_device.volu) << " " << log(param.param_device.volu) << "\n";
				}
555 556
			}
		}
557

David Rohr's avatar
David Rohr committed
558 559 560 561 562
		if(param.param_device.writeAngles)
		{
			angProbfile.close();
		}
		outputProbFile.close();
563
	}
564

565
	return(0);
566 567
}

568
int bioem::compareRefMaps(int iOrient, int iConv, const myfloat_t* conv_map, mycomplex_t* localmultFFT, myfloat_t sumC, myfloat_t sumsquareC, const int startMap)
569
{
David Rohr's avatar
David Rohr committed
570 571
	//***************************************************************************************
	//***** BioEM routine for comparing reference maps to convoluted maps *****
572
	if (FFTAlgo)
573
	{
David Rohr's avatar
David Rohr committed
574
		//With FFT Algorithm
575 576
		#pragma omp parallel for
		for (int iRefMap = startMap; iRefMap < RefMap.ntotRefMap; iRefMap ++)
577
		{
578
			const int num = omp_get_thread_num();
579
			calculateCCFFT(iRefMap, iOrient, iConv, sumC, sumsquareC, localmultFFT, localCCT[num], lCC[num]);
580 581 582
		}
	}
	else
583
	{
David Rohr's avatar
David Rohr committed
584
		//Without FFT Algorithm
585
		#pragma omp parallel for
586
		for (int iRefMap = startMap; iRefMap < RefMap.ntotRefMap; iRefMap ++)
587
		{
588
			compareRefMapShifted < -1 > (iRefMap, iOrient, iConv, conv_map, pProb, param.param_device, RefMap);
589 590 591 592 593
		}
	}
	return(0);
}

594
inline void bioem::calculateCCFFT(int iRefMap, int iOrient, int iConv, myfloat_t sumC, myfloat_t sumsquareC, mycomplex_t* localConvFFT, mycomplex_t* localCCT, myfloat_t* lCC)
595
{
David Rohr's avatar
David Rohr committed
596 597
	//***************************************************************************************
	//***** Calculating cross correlation in FFTALGOrithm *****
Pilar Cossio's avatar
Pilar Cossio committed
598

599
	const mycomplex_t* RefMapFFT = &RefMap.RefMapsFFT[iRefMap * param.FFTMapSize];
600
	for(int i = 0; i < param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D; i++)
601
	{
602 603
		localCCT[i][0] = localConvFFT[i][0] * RefMapFFT[i][0] + localConvFFT[i][1] * RefMapFFT[i][1];
		localCCT[i][1] = localConvFFT[i][1] * RefMapFFT[i][0] - localConvFFT[i][0] * RefMapFFT[i][1];
604 605
	}

606
	myfftw_execute_dft_c2r(param.fft_plan_c2r_backward, localCCT, lCC);
607

608
	doRefMapFFT(iRefMap, iOrient, iConv, lCC, sumC, sumsquareC, pProb, param.param_device, RefMap);
609
}
610

611
int bioem::createProjection(int iMap, mycomplex_t* mapFFT)
612
{
David Rohr's avatar
David Rohr committed
613
	// **************************************************************************************
David Rohr's avatar
David Rohr committed
614 615
	// ****  BioEM Create Projection routine in Euler angle predefined grid******************
	// ********************* and turns projection into Fourier space ************************
David Rohr's avatar
David Rohr committed
616
	// **************************************************************************************
617

618 619
	cuda_custom_timeslot("Projection", 0);

620 621
	myfloat3_t RotatedPointsModel[Model.nPointsModel];
	myfloat_t rotmat[3][3];
622
	myfloat_t alpha, gam, beta;
623
	myfloat_t* localproj;
624

625
	localproj = lCC[omp_get_thread_num()];
626
	memset(localproj, 0, param.param_device.NumberPixels * param.param_device.NumberPixels * sizeof(*localproj));
627

628 629 630
	alpha = param.angles[iMap].pos[0];
	beta = param.angles[iMap].pos[1];
	gam = param.angles[iMap].pos[2];
631

David Rohr's avatar
David Rohr committed
632
	// **** To see how things are going: cout << "Id " << omp_get_thread_num() <<  " Angs: " << alpha << " " << beta << " " << gam << "\n"; ***
633

David Rohr's avatar
David Rohr committed
634
	// ********** Creat Rotation with pre-defiend grid of orientations**********
635 636 637 638 639 640 641 642 643 644 645
	rotmat[0][0] = cos(gam) * cos(alpha) - cos(beta) * sin(alpha) * sin(gam);
	rotmat[0][1] = cos(gam) * sin(alpha) + cos(beta) * cos(alpha) * sin(gam);
	rotmat[0][2] = sin(gam) * sin(beta);
	rotmat[1][0] = -sin(gam) * cos(alpha) - cos(beta) * sin(alpha) * cos(gam);
	rotmat[1][1] = -sin(gam) * sin(alpha) + cos(beta) * cos(alpha) * cos(gam);
	rotmat[1][2] = cos(gam) * sin(beta);
	rotmat[2][0] = sin(beta) * sin(alpha);
	rotmat[2][1] = -sin(beta) * cos(alpha);
	rotmat[2][2] = cos(beta);

	for(int n = 0; n < Model.nPointsModel; n++)
646
	{
647 648 649
		RotatedPointsModel[n].pos[0] = 0.0;
		RotatedPointsModel[n].pos[1] = 0.0;
		RotatedPointsModel[n].pos[2] = 0.0;
650
	}
651
	for(int n = 0; n < Model.nPointsModel; n++)
652
	{
653
		for(int k = 0; k < 3; k++)
654
		{
655
			for(int j = 0; j < 3; j++)
656
			{
657
				RotatedPointsModel[n].pos[k] += rotmat[k][j] * Model.points[n].point.pos[j];
658 659 660 661 662 663
			}
		}
	}

	int i, j;

David Rohr's avatar
David Rohr committed
664
	// ************ Projection over the Z axis********************
665
	for(int n = 0; n < Model.nPointsModel; n++)
666 667
	{
		//Getting pixel that represents coordinates & shifting the start at to Numpix/2,Numpix/2 )
668 669
		i = floor(RotatedPointsModel[n].pos[0] / param.pixelSize + (myfloat_t) param.param_device.NumberPixels / 2.0f + 0.5f);
		j = floor(RotatedPointsModel[n].pos[1] / param.pixelSize + (myfloat_t) param.param_device.NumberPixels / 2.0f + 0.5f);
670

671 672 673 674 675 676
		if (i < 0 || j < 0 || i >= param.param_device.NumberPixels || j >= param.param_device.NumberPixels)
		{
			if (DebugOutput >= 3) cout << "Model Point out of map: " << i << ", " << j << "\n";
			continue;
		}

677
		localproj[i * param.param_device.NumberPixels + j] += Model.points[n].density / Model.NormDen;
678 679
	}

David Rohr's avatar
David Rohr committed
680
	// **** Output Just to check****
681
#ifdef PILAR_DEBUG
682
	if(iMap == 10)
683 684 685 686 687 688
	{
		ofstream myexamplemap;
		ofstream myexampleRot;
		myexamplemap.open ("MAP_i10");
		myexampleRot.open ("Rot_i10");
		myexamplemap << "ANGLES " << alpha << " " << beta << " " << gam << "\n";
689
		for(int k = 0; k < param.param_device.NumberPixels; k++)
690
		{
691
			for(int j = 0; j < param.param_device.NumberPixels; j++) myexamplemap << "\nMAP " << k << " " << j << " " << localproj[k * param.param_device.NumberPixels + j];
692 693
		}
		myexamplemap << " \n";
694
		for(int n = 0; n < Model.nPointsModel; n++)myexampleRot << "\nCOOR " << RotatedPointsModel[n].pos[0] << " " << RotatedPointsModel[n].pos[1] << " " << RotatedPointsModel[n].pos[2];
695 696 697
		myexamplemap.close();
		myexampleRot.close();
	}
698
#endif
699

David Rohr's avatar
David Rohr committed
700 701
	// ***** Converting projection to Fourier Space for Convolution later with kernel****
	// ********** Omp Critical is necessary with FFTW*******
702
	myfftw_execute_dft_r2c(param.fft_plan_r2c_forward, localproj, mapFFT);
703

704 705
	cuda_custom_timeslot_end;

706 707 708
	return(0);
}

709
int bioem::createConvolutedProjectionMap(int iMap, int iConv, mycomplex_t* lproj, myfloat_t* Mapconv, mycomplex_t* localmultFFT, myfloat_t& sumC, myfloat_t& sumsquareC)
710
{
David Rohr's avatar
David Rohr committed
711 712
	// **************************************************************************************
	// ****  BioEM Create Convoluted Projection Map routine, multiplies in Fourier **********
David Rohr's avatar
David Rohr committed
713 714
	// **************** calculated Projection with convoluted precalculated Kernel***********
	// *************** and Backtransforming it to real Space ********************************
David Rohr's avatar
David Rohr committed
715
	// **************************************************************************************
716

717 718
	cuda_custom_timeslot("Convolution", 1);

719
	mycomplex_t* tmp = localCCT[omp_get_thread_num()];
720

David Rohr's avatar
David Rohr committed
721
	// **** Multiplying FFTmap with corresponding kernel ****
722
	const mycomplex_t* refCTF = &param.refCTF[iConv * param.FFTMapSize];
723
	for(int i = 0; i < param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D; i++)
724
	{
725 726 727
		localmultFFT[i][0] = lproj[i][0] * refCTF[i][0] + lproj[i][1] * refCTF[i][1];
		localmultFFT[i][1] = lproj[i][1] * refCTF[i][0] - lproj[i][0] * refCTF[i][1];
		// cout << "GG " << i << " " << j << " " << refCTF[i][0] << " " << refCTF[i][1] <<" " <<lproj[i][0] <<" " <<lproj[i][1] << "\n";
728 729
	}

David Rohr's avatar
David Rohr committed
730
	// *** Calculating Cross-correlations of cal-convoluted map with its self *****
731 732
	sumC = localmultFFT[0][0];
	
733
	sumsquareC = 0;
734
	if (FFTAlgo)
735
	{
736 737 738 739 740 741 742 743 744 745 746 747 748
		for(int i = 0; i < param.param_device.NumberPixels; i++)
		{
			for (int j = 1;j < param.param_device.NumberFFTPixels1D;j++)
			{
				int k = i * param.param_device.NumberFFTPixels1D + j;
				sumsquareC += (localmultFFT[k][0] * localmultFFT[k][0] + localmultFFT[k][1] * localmultFFT[k][1]) * 2;
			}
			int k = i * param.param_device.NumberFFTPixels1D;
			sumsquareC += localmultFFT[k][0] * localmultFFT[k][0] + localmultFFT[k][1] * localmultFFT[k][1];
		}
		
		myfloat_t norm2 = (myfloat_t) (param.param_device.NumberPixels * param.param_device.NumberPixels);
		sumsquareC = sumsquareC / norm2;
749
	}
750 751 752 753 754 755 756
	else
	{
		//FFTW_C2R will destroy the input array, so we have to work on a copy here
		memcpy(tmp, localmultFFT, sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);

		// **** Bringing convoluted Map to real Space ****
		myfftw_execute_dft_c2r(param.fft_plan_c2r_backward, tmp, Mapconv);
757

758 759 760 761 762 763 764 765 766 767 768
		for(int i = 0; i < param.param_device.NumberPixels * param.param_device.NumberPixels; i++)
		{
			sumsquareC += Mapconv[i] * Mapconv[i];
			sumC += Mapconv[i];
		}

		myfloat_t norm2 = (myfloat_t) (param.param_device.NumberPixels * param.param_device.NumberPixels);
		myfloat_t norm4 = norm2 * norm2;
		sumsquareC = sumsquareC / norm4;
	}
	
769 770
	cuda_custom_timeslot_end;

771
	return(0);
772 773
}

774
int bioem::calcross_cor(myfloat_t* localmap, myfloat_t& sum, myfloat_t& sumsquare)
775
{
David Rohr's avatar
David Rohr committed
776
	// *********************** Routine to calculate Cross correlations***********************
777

778 779
	sum = 0.0;
	sumsquare = 0.0;
780 781 782 783 784
	for (int i = 0; i < param.param_device.NumberPixels; i++)
	{
		for (int j = 0; j < param.param_device.NumberPixels; j++)
		{
			// Calculate Sum of pixels
785
			sum += localmap[i * param.param_device.NumberPixels + j];
786
			// Calculate Sum of pixels squared
787
			sumsquare += localmap[i * param.param_device.NumberPixels + j] * localmap[i * param.param_device.NumberPixels + j];
788 789 790
		}
	}
	return(0);
791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806
}

int bioem::deviceInit()
{
	return(0);
}

int bioem::deviceStartRun()
{
	return(0);
}

int bioem::deviceFinishRun()
{
	return(0);
}
807 808 809 810 811 812 813 814 815 816

void* bioem::malloc_device_host(size_t size)
{
	return(mallocchk(size));
}

void bioem::free_device_host(void* ptr)
{
	free(ptr);
}