sz_float_pwr.c 24.4 KB
Newer Older
T
tickduan 已提交
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
/**
 *  @file sz_float_pwr.c
 *  @author Sheng Di, Dingwen Tao, Xin Liang, Xiangyu Zou, Tao Lu, Wen Xia, Xuan Wang, Weizhe Zhang
 *  @date Aug, 2016
 *  @brief SZ_Init, Compression and Decompression functions
 * This file contains the compression/decompression functions related to point-wise relative errors
 *  (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */


#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <math.h>
#include "sz.h"
#include "CompressElement.h"
#include "DynamicByteArray.h"
#include "DynamicIntArray.h"
#include "TightDataPointStorageF.h"
#include "sz_float.h"
#include "sz_float_pwr.h"
#include "zlib.h"
#include "rw.h"
#include "utility.h"

void compute_segment_precisions_float_1D(float *oriData, size_t dataLength, float* pwrErrBound, unsigned char* pwrErrBoundBytes, double globalPrecision)
{
	size_t i = 0, j = 0, k = 0;
	float realPrecision = oriData[0]!=0?fabs(confparams_cpr->pw_relBoundRatio*oriData[0]):confparams_cpr->pw_relBoundRatio; 
	float approxPrecision;
	unsigned char realPrecBytes[4];
	float curPrecision;
	float curValue;
	float sum = 0;
	for(i=0;i<dataLength;i++)
	{
		curValue = oriData[i];
		if(i%confparams_cpr->segment_size==0&&i>0)
		{
			//get two first bytes of the realPrecision
			if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE)
			{
				realPrecision = sum/confparams_cpr->segment_size;
				sum = 0;			
			}
			realPrecision *= confparams_cpr->pw_relBoundRatio;
			
			if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL)
				realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; 
			else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL)
				realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision;
				
			floatToBytes(realPrecBytes, realPrecision);
			realPrecBytes[2] = realPrecBytes[3] = 0;
			approxPrecision = bytesToFloat(realPrecBytes);
			//put the realPrecision in float* pwrErBound
			pwrErrBound[j++] = approxPrecision;
			//put the two bytes in pwrErrBoundBytes
			pwrErrBoundBytes[k++] = realPrecBytes[0];
			pwrErrBoundBytes[k++] = realPrecBytes[1];
			
			realPrecision = fabs(curValue);
		}
		
		if(curValue!=0)
		{
			curPrecision = fabs(curValue);
			
			switch(confparams_cpr->pwr_type)
			{
			case SZ_PWR_MIN_TYPE: 
				if(realPrecision>curPrecision)
					realPrecision = curPrecision;	
				break;
			case SZ_PWR_AVG_TYPE:
				sum += curPrecision;
				break;
			case SZ_PWR_MAX_TYPE:
				if(realPrecision<curPrecision)
					realPrecision = curPrecision;					
				break;
			}
		}
	}
	if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE)
	{
		int size = dataLength%confparams_cpr->segment_size==0?confparams_cpr->segment_size:dataLength%confparams_cpr->segment_size;
		realPrecision = sum/size;		
	}	
	if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL)
		realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; 
	else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL)
		realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision;	
	floatToBytes(realPrecBytes, realPrecision);
	realPrecBytes[2] = realPrecBytes[3] = 0;
	approxPrecision = bytesToFloat(realPrecBytes);
	//put the realPrecision in float* pwrErBound
	pwrErrBound[j++] = approxPrecision;
	//put the two bytes in pwrErrBoundBytes
	pwrErrBoundBytes[k++] = realPrecBytes[0];
	pwrErrBoundBytes[k++] = realPrecBytes[1];
}

unsigned int optimize_intervals_float_1D_pwr(float *oriData, size_t dataLength, float* pwrErrBound)
{	
	size_t i = 0, j = 0;
	float realPrecision = pwrErrBound[j++];	
	unsigned long radiusIndex;
	float pred_value = 0, pred_err;
	int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int));
	int totalSampleSize = dataLength/confparams_cpr->sampleDistance;
	for(i=2;i<dataLength;i++)
	{
		if(i%confparams_cpr->segment_size==0)
			realPrecision = pwrErrBound[j++];
		if(i%confparams_cpr->sampleDistance==0)
		{
			//pred_value = 2*oriData[i-1] - oriData[i-2];
			pred_value = oriData[i-1];
			pred_err = fabs(pred_value - oriData[i]);
			radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
			if(radiusIndex>=confparams_cpr->maxRangeRadius)
				radiusIndex = confparams_cpr->maxRangeRadius - 1;			
			intervals[radiusIndex]++;
		}
	}
	//compute the appropriate number
	size_t targetCount = totalSampleSize*confparams_cpr->predThreshold;
	size_t sum = 0;
	for(i=0;i<confparams_cpr->maxRangeRadius;i++)
	{
		sum += intervals[i];
		if(sum>targetCount)
			break;
	}
	if(i>=confparams_cpr->maxRangeRadius)
		i = confparams_cpr->maxRangeRadius-1;

	unsigned int accIntervals = 2*(i+1);
	unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
	
	if(powerOf2<32)
		powerOf2 = 32;
	
	free(intervals);
	//printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2);
	return powerOf2;
}

T
tickduan 已提交
153
void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(unsigned char* newByteData, float *oriData, double globalPrecision, 
T
tickduan 已提交
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
size_t dataLength, size_t *outSize, float min, float max)
{
	size_t pwrLength = dataLength%confparams_cpr->segment_size==0?dataLength/confparams_cpr->segment_size:dataLength/confparams_cpr->segment_size+1;
	float* pwrErrBound = (float*)malloc(sizeof(float)*pwrLength);
	size_t pwrErrBoundBytes_size = sizeof(unsigned char)*pwrLength*2;
	unsigned char* pwrErrBoundBytes = (unsigned char*)malloc(pwrErrBoundBytes_size);
	
	compute_segment_precisions_float_1D(oriData, dataLength, pwrErrBound, pwrErrBoundBytes, globalPrecision);
	
	unsigned int quantization_intervals;
	if(exe_params->optQuantMode==1)
	{
		quantization_intervals = optimize_intervals_float_1D_pwr(oriData, dataLength, pwrErrBound);	
		updateQuantizationInfo(quantization_intervals);
	}
	else
		quantization_intervals = exe_params->intvCapacity;
	size_t i = 0, j = 0;
	int reqLength;
	float realPrecision = pwrErrBound[j++];	
	float medianValue = 0;
	float radius = fabs(max)<fabs(min)?fabs(min):fabs(max);
	short radExpo = getExponent_float(radius);
	
	computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue);

	int* type = (int*) malloc(dataLength*sizeof(int));
	//type[dataLength]=0;
		
	float* spaceFillingValue = oriData; //
	
	DynamicByteArray *resiBitLengthArray;
	new_DBA(&resiBitLengthArray, DynArrayInitLen);
	
	DynamicIntArray *exactLeadNumArray;
	new_DIA(&exactLeadNumArray, DynArrayInitLen);
	
	DynamicByteArray *exactMidByteArray;
	new_DBA(&exactMidByteArray, DynArrayInitLen);
	
	DynamicIntArray *resiBitArray;
	new_DIA(&resiBitArray, DynArrayInitLen);
	
	type[0] = 0;
	
	unsigned char preDataBytes[4] = {0};
	intToBytes_bigEndian(preDataBytes, 0);
	
	int reqBytesLength = reqLength/8;
	int resiBitsLength = reqLength%8;
	float last3CmprsData[3] = {0};

	FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
	LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
						
	//add the first data	
	addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength);
	compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
	updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
	memcpy(preDataBytes,vce->curBytes,4);
	addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
	listAdd_float(last3CmprsData, vce->data);
	//printf("%.30G\n",last3CmprsData[0]);	
		
	//add the second data
	type[1] = 0;
	addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength);			
	compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
	updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
	memcpy(preDataBytes,vce->curBytes,4);
	addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
	listAdd_float(last3CmprsData, vce->data);
	//printf("%.30G\n",last3CmprsData[0]);	
	
	int state;
	double checkRadius;
	float curData;
	float pred;
	double predAbsErr;
	checkRadius = (exe_params->intvCapacity-1)*realPrecision;
	double interval = 2*realPrecision;
	int updateReqLength = 0; //a marker: 1 means already updated
	
	for(i=2;i<dataLength;i++)
	{
		curData = spaceFillingValue[i];
		if(i%confparams_cpr->segment_size==0)
		{
			realPrecision = pwrErrBound[j++];
			checkRadius = (exe_params->intvCapacity-1)*realPrecision;
			interval = 2*realPrecision;
			updateReqLength = 0;
		}
		//pred = 2*last3CmprsData[0] - last3CmprsData[1];
		pred = last3CmprsData[0];
		predAbsErr = fabs(curData - pred);	
		if(predAbsErr<checkRadius)
		{
			state = (predAbsErr/realPrecision+1)/2;
			if(curData>=pred)
			{
				type[i] = exe_params->intvRadius+state;
				pred = pred + state*interval;
			}
			else //curData<pred
			{
				type[i] = exe_params->intvRadius-state;
				pred = pred - state*interval;
			}
			listAdd_float(last3CmprsData, pred);			
			continue;
		}
		
		//unpredictable data processing		
T
tickduan 已提交
268 269 270 271 272 273
		if(updateReqLength==0)
		{
			computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue);
			reqBytesLength = reqLength/8;
			resiBitsLength = reqLength%8;
			updateReqLength = 1;		
T
tickduan 已提交
274
		}
T
tickduan 已提交
275 276 277 278 279 280 281 282
		
		type[i] = 0;
		addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength);
		
		compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
		updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
		memcpy(preDataBytes,vce->curBytes,4);
		addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
T
tickduan 已提交
283

T
tickduan 已提交
284 285 286
		listAdd_float(last3CmprsData, vce->data);	
	}//end of for
		
T
tickduan 已提交
287
	int exactDataNum = exactLeadNumArray->size;
T
tickduan 已提交
288
	
T
tickduan 已提交
289
	TightDataPointStorageF* tdps;
T
tickduan 已提交
290 291 292 293 294
			
	new_TightDataPointStorageF2(&tdps, dataLength, exactDataNum, 
			type, exactMidByteArray->array, exactMidByteArray->size,  
			exactLeadNumArray->array,  
			resiBitArray->array, resiBitArray->size, 
T
tickduan 已提交
295 296 297
			resiBitLengthArray->array, resiBitLengthArray->size, 
			realPrecision, medianValue, (char)reqLength, quantization_intervals, pwrErrBoundBytes, pwrErrBoundBytes_size, radExpo);

T
tickduan 已提交
298
	
T
tickduan 已提交
299 300 301 302 303
	//free memory
	free_DBA(resiBitLengthArray);
	free_DIA(exactLeadNumArray);
	free_DIA(resiBitArray);
	free(type);
T
tickduan 已提交
304 305 306 307 308 309 310 311 312 313 314 315 316 317
	
	convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
	
	int floatSize=sizeof(float);
	if(*outSize>dataLength*floatSize)
	{
		size_t k = 0, i;
		tdps->isLossless = 1;
		size_t totalByteLength = 3 + exe_params->SZ_SIZE_TYPE + 1 + floatSize*dataLength;
		*newByteData = (unsigned char*)malloc(totalByteLength);
		
		unsigned char dsLengthBytes[exe_params->SZ_SIZE_TYPE];
		intToBytes_bigEndian(dsLengthBytes, dataLength);//4
		for (i = 0; i < 3; i++)//3
T
tickduan 已提交
318
			newByteData[k++] = versionNumber[i];
T
tickduan 已提交
319 320 321
		
		if(exe_params->SZ_SIZE_TYPE==4)
		{
T
tickduan 已提交
322
			newByteData[k++] = 16;	//=00010000	
T
tickduan 已提交
323 324 325
		}
		else 
		{
T
tickduan 已提交
326
			newByteData[k++] = 80;
T
tickduan 已提交
327 328
		}
		for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++)//4 or 8
T
tickduan 已提交
329
			newByteData[k++] = dsLengthBytes[i];
T
tickduan 已提交
330

T
tickduan 已提交
331 332
		
		if(sysEndianType==BIG_ENDIAN_SYSTEM)
T
tickduan 已提交
333
			memcpy(newByteData+4+exe_params->SZ_SIZE_TYPE, oriData, dataLength*floatSize);
T
tickduan 已提交
334 335
		else
		{
T
tickduan 已提交
336
			unsigned char* p = newByteData+4+exe_params->SZ_SIZE_TYPE;
T
tickduan 已提交
337 338 339 340 341
			for(i=0;i<dataLength;i++,p+=floatSize)
				floatToBytes(p, oriData[i]);
		}
		*outSize = totalByteLength;
	}
T
tickduan 已提交
342 343

	free(pwrErrBound);
T
tickduan 已提交
344
	
T
tickduan 已提交
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
	free(vce);
	free(lce);
	free_TightDataPointStorageF(tdps);
	free(exactMidByteArray);
}

void createRangeGroups_float(float** posGroups, float** negGroups, int** posFlags, int** negFlags)
{
	size_t size = GROUP_COUNT*sizeof(float);
	size_t size2 = GROUP_COUNT*sizeof(int);
	*posGroups = (float*)malloc(size);
	*negGroups = (float*)malloc(size);
	*posFlags = (int*)malloc(size2);
	*negFlags = (int*)malloc(size2);
	memset(*posGroups, 0, size);
	memset(*negGroups, 0, size);
	memset(*posFlags, 0, size2);
	memset(*negFlags, 0, size2);
}

void compressGroupIDArray_float(char* groupID, TightDataPointStorageF* tdps)
{
	size_t dataLength = tdps->dataSeriesLength;
	int* standGroupID = (int*)malloc(dataLength*sizeof(int));

	size_t i;
	standGroupID[0] = groupID[0]+GROUP_COUNT; //plus an offset such that it would not be a negative number.
	char lastGroupIDValue = groupID[0], curGroupIDValue;
	int offset = 2*(GROUP_COUNT + 2);
	for(i=1; i<dataLength;i++)
	{
		curGroupIDValue = groupID[i];
		standGroupID[i] = (curGroupIDValue - lastGroupIDValue) + offset; 
		lastGroupIDValue = curGroupIDValue;
	}
	
	unsigned char* out = NULL;
	size_t outSize;
	
	HuffmanTree* huffmanTree = SZ_Reset();
	encode_withTree(huffmanTree, standGroupID, dataLength, &out, &outSize);
	SZ_ReleaseHuffman(huffmanTree);
	
	tdps->pwrErrBoundBytes = out; //groupIDArray
	tdps->pwrErrBoundBytes_size = outSize;
	
	free(standGroupID);
}

TightDataPointStorageF* SZ_compress_float_1D_MDQ_pwrGroup(float* oriData, size_t dataLength, int errBoundMode, 
double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSize, float medianValue_f)
{
	size_t i;
	float *posGroups, *negGroups, *groups;
	float pos_01_group = 0, neg_01_group = 0; //[0,1] and [-1,0]
	int *posFlags, *negFlags, *flags;
	int pos_01_flag = 0, neg_01_flag = 0;
	createRangeGroups_float(&posGroups, &negGroups, &posFlags, &negFlags);
	size_t nbBins = (size_t)(1/pwrErrRatio);
	if(nbBins%2==1)
		nbBins++;
	exe_params->intvRadius = nbBins;

	int reqLength, status;
	float medianValue = medianValue_f;
	float realPrecision = (float)getRealPrecision_float(valueRangeSize, errBoundMode, absErrBound, relBoundRatio, &status);
	if(realPrecision<0)
		realPrecision = pwrErrRatio;
	float realGroupPrecision; //precision (error) based on group ID
	getPrecisionReqLength_float(realPrecision);
	short radExpo = getExponent_float(valueRangeSize/2);
	short lastGroupNum = 0, groupNum, grpNum = 0;
	
	double* groupErrorBounds = generateGroupErrBounds(errBoundMode, realPrecision, pwrErrRatio);
	exe_params->intvRadius = generateGroupMaxIntervalCount(groupErrorBounds);
	
	computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue);

	int* type = (int*) malloc(dataLength*sizeof(int));
	char *groupID = (char*) malloc(dataLength*sizeof(char));
	char *gp = groupID;
		
	float* spaceFillingValue = oriData; 
	
	DynamicIntArray *exactLeadNumArray;
	new_DIA(&exactLeadNumArray, DynArrayInitLen);
	
	DynamicByteArray *exactMidByteArray;
	new_DBA(&exactMidByteArray, DynArrayInitLen);
	
	DynamicIntArray *resiBitArray;
	new_DIA(&resiBitArray, DynArrayInitLen);
	
	unsigned char preDataBytes[4];
	intToBytes_bigEndian(preDataBytes, 0);
	
	int reqBytesLength = reqLength/8;
	int resiBitsLength = reqLength%8;

	FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
	LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
			
	int state;
	float curData, decValue;
	float pred;
	float predAbsErr;
	double interval = 0;
	
	//add the first data	
	type[0] = 0;
	compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
	updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
	memcpy(preDataBytes,vce->curBytes,4);
	addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
	
	curData = spaceFillingValue[0];
	groupNum = computeGroupNum_float(vce->data);

	if(curData > 0 && groupNum >= 0)
	{
		groups = posGroups;
		flags = posFlags;
		grpNum = groupNum;
	}
	else if(curData < 0 && groupNum >= 0)
	{
		groups = negGroups;
		flags = negFlags;
		grpNum = groupNum;
	}
	else if(curData >= 0 && groupNum == -1)
	{
		groups = &pos_01_group;
		flags = &pos_01_flag;
		grpNum = 0;
	}
	else //curData < 0 && groupNum == -1
	{
		groups = &neg_01_group;
		flags = &neg_01_flag;
		grpNum = 0;
	}

	listAdd_float_group(groups, flags, groupNum, spaceFillingValue[0], vce->data, gp);
	gp++;
	
	for(i=1;i<dataLength;i++)
	{
		curData = oriData[i];
		//printf("i=%d, posGroups[3]=%f, negGroups[3]=%f\n", i, posGroups[3], negGroups[3]);
		
		groupNum = computeGroupNum_float(curData);
		
		if(curData > 0 && groupNum >= 0)
		{
			groups = posGroups;
			flags = posFlags;
			grpNum = groupNum;
		}
		else if(curData < 0 && groupNum >= 0)
		{
			groups = negGroups;
			flags = negFlags;
			grpNum = groupNum;
		}
		else if(curData >= 0 && groupNum == -1)
		{
			groups = &pos_01_group;
			flags = &pos_01_flag;
			grpNum = 0;
		}
		else //curData < 0 && groupNum == -1
		{
			groups = &neg_01_group;
			flags = &neg_01_flag;
			grpNum = 0;
		}

		if(groupNum>=GROUP_COUNT)
		{
			type[i] = 0;
			compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
			updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
			memcpy(preDataBytes,vce->curBytes,4);
			addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
			listAdd_float_group(groups, flags, lastGroupNum, curData, vce->data, gp);	//set the group number to be last one in order to get the groupID array as smooth as possible.		
		}
		else if(flags[grpNum]==0) //the dec value may not be in the same group
		{	
			type[i] = 0;
			compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
			updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
			memcpy(preDataBytes,vce->curBytes,4);
			addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
			//decGroupNum = computeGroupNum_float(vce->data);
			
			//if(decGroupNum < groupNum)
			//	decValue = curData>0?pow(2, groupNum):-pow(2, groupNum);
			//else if(decGroupNum > groupNum)
			//	decValue = curData>0?pow(2, groupNum+1):-pow(2, groupNum+1);
			//else
			//	decValue = vce->data;
			
			decValue = vce->data;	
			listAdd_float_group(groups, flags, groupNum, curData, decValue, gp);
			lastGroupNum = curData>0?groupNum + 2: -(groupNum+2);
		}
		else //if flags[groupNum]==1, the dec value must be in the same group
		{
			pred = groups[grpNum];
			predAbsErr = fabs(curData - pred);
			realGroupPrecision = groupErrorBounds[grpNum]; //compute real error bound
			interval = realGroupPrecision*2;
			state = (predAbsErr/realGroupPrecision+1)/2;
			if(curData>=pred)
			{
				type[i] = exe_params->intvRadius+state;
				decValue = pred + state*interval;
			}
			else //curData<pred
			{
				type[i] = exe_params->intvRadius-state;
				decValue = pred - state*interval;
			}
			//decGroupNum = computeGroupNum_float(pred);
			
			if((decValue>0&&curData<0)||(decValue<0&&curData>=0))
				decValue = 0;
			//else
			//{
			//	if(decGroupNum < groupNum)
			//		decValue = curData>0?pow(2, groupNum):-pow(2, groupNum);
			//	else if(decGroupNum > groupNum)
			//		decValue = curData>0?pow(2, groupNum+1):-pow(2, groupNum+1);
			//	else
			//		decValue = pred;				
			//}
			
			if(fabs(curData-decValue)>realGroupPrecision)
			{	
				type[i] = 0;
				compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength);
				updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
				memcpy(preDataBytes,vce->curBytes,4);
				addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);

				decValue = vce->data;	
			}
			
			listAdd_float_group(groups, flags, groupNum, curData, decValue, gp);			
			lastGroupNum = curData>=0?groupNum + 2: -(groupNum+2);			
		}
		gp++;	

	}
	
	int exactDataNum = exactLeadNumArray->size;
	
	TightDataPointStorageF* tdps;
			
	//combineTypeAndGroupIDArray(nbBins, dataLength, &type, groupID);

	new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, 
			type, exactMidByteArray->array, exactMidByteArray->size,  
			exactLeadNumArray->array,  
			resiBitArray->array, resiBitArray->size, 
			resiBitsLength, 
			realPrecision, medianValue, (char)reqLength, nbBins, NULL, 0, radExpo);	
	
	compressGroupIDArray_float(groupID, tdps);
	
	free(posGroups);
	free(negGroups);
	free(posFlags);
	free(negFlags);
	free(groupID);
	free(groupErrorBounds);
	
	free_DIA(exactLeadNumArray);
	free_DIA(resiBitArray);
	free(type);	
	free(vce);
	free(lce);	
	free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps);	
	
	return tdps;
}

T
tickduan 已提交
633
void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(unsigned char* newByteData, float *oriData,
T
tickduan 已提交
634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649
size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSize, float medianValue_f, size_t *outSize)
{
        TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ_pwrGroup(oriData, dataLength, confparams_cpr->errorBoundMode, 
        absErrBound, relBoundRatio, pwrErrRatio, 
        valueRangeSize, medianValue_f);

        convertTDPStoFlatBytes_float(tdps, newByteData, outSize);

        if(*outSize>3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1 + sizeof(float)*dataLength)
                SZ_compress_args_float_StoreOriData(oriData, dataLength, newByteData, outSize);

        free_TightDataPointStorageF(tdps);
}

#include <stdbool.h>

T
tickduan 已提交
650
void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log(unsigned char* newByteData, float *oriData, double pwrErrRatio, size_t dataLength, size_t *outSize, float min, float max){
T
tickduan 已提交
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711

	float * log_data = (float *) malloc(dataLength * sizeof(float));

	unsigned char * signs = (unsigned char *) malloc(dataLength);
	memset(signs, 0, dataLength);
	// preprocess
	float max_abs_log_data;
    if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
    else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
    else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
    float min_log_data = max_abs_log_data;
	bool positive = true;
	for(size_t i=0; i<dataLength; i++){
		if(oriData[i] < 0){
			signs[i] = 1;
			log_data[i] = -oriData[i];
			positive = false;
		}
		else
			log_data[i] = oriData[i];
		if(log_data[i] > 0){
			log_data[i] = log2(log_data[i]);
			if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
			if(log_data[i] < min_log_data) min_log_data = log_data[i];
		}
	}

	float valueRangeSize, medianValue_f;
	computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f);	
	if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
	double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7;
	for(size_t i=0; i<dataLength; i++){
		if(oriData[i] == 0){
			log_data[i] = min_log_data - 2.0001*realPrecision;
		}
	}

    TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ(log_data, dataLength, realPrecision, valueRangeSize, medianValue_f);
    tdps->minLogValue = min_log_data - 1.0001*realPrecision;
    free(log_data);
    if(!positive){
	    unsigned char * comp_signs;
		// compress signs
		unsigned long signSize = sz_lossless_compress(ZSTD_COMPRESSOR, 3, signs, dataLength, &comp_signs);
		tdps->pwrErrBoundBytes = comp_signs;
		tdps->pwrErrBoundBytes_size = signSize;
	}
	else{
		tdps->pwrErrBoundBytes = NULL;
		tdps->pwrErrBoundBytes_size = 0;
	}
	free(signs);

    convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
    if(*outSize>3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1 + sizeof(float)*dataLength)
            SZ_compress_args_float_StoreOriData(oriData, dataLength, newByteData, outSize);

    free_TightDataPointStorageF(tdps);
}


T
tickduan 已提交
712
void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log_MSST19(unsigned char* newByteData, float *oriData, double pwrErrRatio, size_t dataLength, size_t *outSize, float valueRangeSize, float medianValue_f,
T
tickduan 已提交
713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
																unsigned char* signs, bool* positive, float min, float max, float nearZero){
	float multiplier = pow((1+pwrErrRatio), -3.0001);
	for(int i=0; i<dataLength; i++){
		if(oriData[i] == 0){
			oriData[i] = nearZero * multiplier;
		}
	}

	float median_log = sqrt(fabs(nearZero * max));

	TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ_MSST19(oriData, dataLength, pwrErrRatio, valueRangeSize, median_log);

	tdps->minLogValue = nearZero / ((1+pwrErrRatio)*(1+pwrErrRatio));
	if(!(*positive)){
		unsigned char * comp_signs;
		// compress signs
		unsigned long signSize = sz_lossless_compress(ZSTD_COMPRESSOR, 3, signs, dataLength, &comp_signs);
		tdps->pwrErrBoundBytes = comp_signs;
		tdps->pwrErrBoundBytes_size = signSize;
	}
	else{
		tdps->pwrErrBoundBytes = NULL;
		tdps->pwrErrBoundBytes_size = 0;
	}
	free(signs);

	convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
	if(*outSize>3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1 + sizeof(float)*dataLength)
		SZ_compress_args_float_StoreOriData(oriData, dataLength, newByteData, outSize);

	free_TightDataPointStorageF(tdps);
}