sz_float.c 65.7 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
/**
 *  @file sz_float.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
 *  (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */


#include <stddef.h>
#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 "szd_float.h"
#include "szd_float_pwr.h"
#include "zlib.h"
#include "rw.h"
#include "sz_float_ts.h"
#include "utility.h"
#include "CacheTable.h"
#include "MultiLevelCacheTableWideInterval.h"
#include "sz_stats.h"

T
tickduan 已提交
34 35 36
# define MIN(_a, _b) (((_a) < (_b)) ? (_a) : (_b))


T
tickduan 已提交
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609
unsigned char* SZ_skip_compress_float(float* data, size_t dataLength, size_t* outSize)
{
	*outSize = dataLength*sizeof(float);
	unsigned char* out = (unsigned char*)malloc(dataLength*sizeof(float));
	memcpy(out, data, dataLength*sizeof(float));
	return out;
}

void computeReqLength_float(double realPrecision, short radExpo, int* reqLength, float* medianValue)
{
	short reqExpo = getPrecisionReqLength_double(realPrecision);
	*reqLength = 9+radExpo - reqExpo+1; //radExpo-reqExpo == reqMantiLength
	if(*reqLength<9)
		*reqLength = 9;
	if(*reqLength>32)
	{	
		*reqLength = 32;
		*medianValue = 0;
	}			
}

inline short computeReqLength_float_MSST19(double realPrecision)
{
	short reqExpo = getPrecisionReqLength_float(realPrecision);
	return 9-reqExpo;
}

unsigned int optimize_intervals_float_1D(float *oriData, size_t dataLength, double realPrecision)
{	
	size_t i = 0, radiusIndex;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = dataLength/confparams_cpr->sampleDistance;
	for(i=2;i<dataLength;i++)
	{
		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;
}

unsigned int optimize_intervals_float_2D(float *oriData, size_t r1, size_t r2, double realPrecision)
{	
	size_t i,j, index;
	size_t radiusIndex;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = (r1-1)*(r2-1)/confparams_cpr->sampleDistance;

	//float max = oriData[0];
	//float min = oriData[0];

	for(i=1;i<r1;i++)
	{
		for(j=1;j<r2;j++)
		{
			if((i+j)%confparams_cpr->sampleDistance==0)
			{
				index = i*r2+j;
				pred_value = oriData[index-1] + oriData[index-r2] - oriData[index-r2-1];
				pred_err = fabs(pred_value - oriData[index]);
				radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
				if(radiusIndex>=confparams_cpr->maxRangeRadius)
					radiusIndex = confparams_cpr->maxRangeRadius - 1;
				intervals[radiusIndex]++;

			//	if (max < oriData[index]) max = oriData[index];
			//	if (min > oriData[index]) min = oriData[index];
			}			
		}
	}
	//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;

	//	struct timeval costStart, costEnd;
	//	double cost_est = 0;
	//
	//	gettimeofday(&costStart, NULL);
	//
	//	//compute estimate of bit-rate and distortion
	//	double est_br = 0;
	//	double est_psnr = 0;
	//	double c1 = log2(targetCount)+1;
	//	double c2 = -20.0*log10(realPrecision) + 20.0*log10(max-min) + 10.0*log10(3);
	//
	//	for (i = 0; i < powerOf2/2; i++)
	//	{
	//		int count = intervals[i];
	//		if (count != 0)
	//			est_br += count*log2(count);
	//		est_psnr += count;
	//	}
	//
	//	//compute estimate of bit-rate
	//	est_br -= c1*est_psnr;
	//	est_br /= totalSampleSize;
	//	est_br = -est_br;
	//
	//	//compute estimate of psnr
	//	est_psnr /= totalSampleSize;
	//	printf ("sum of P(i) = %lf\n", est_psnr);
	//	est_psnr = -10.0*log10(est_psnr);
	//	est_psnr += c2;
	//
	//	printf ("estimate bitrate = %.2f\n", est_br);
	//	printf ("estimate psnr = %.2f\n",est_psnr);
	//
	//	gettimeofday(&costEnd, NULL);
	//	cost_est = ((costEnd.tv_sec*1000000+costEnd.tv_usec)-(costStart.tv_sec*1000000+costStart.tv_usec))/1000000.0;
	//
	//	printf ("analysis time = %f\n", cost_est);

	free(intervals);
	//printf("confparams_cpr->maxRangeRadius = %d, accIntervals=%d, powerOf2=%d\n", confparams_cpr->maxRangeRadius, accIntervals, powerOf2);
	return powerOf2;
}

unsigned int optimize_intervals_float_3D(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision)
{	
	size_t i,j,k, index;
	size_t radiusIndex;
	size_t r23=r2*r3;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = (r1-1)*(r2-1)*(r3-1)/confparams_cpr->sampleDistance;

	//float max = oriData[0];
	//float min = oriData[0];

	for(i=1;i<r1;i++)
	{
		for(j=1;j<r2;j++)
		{
			for(k=1;k<r3;k++)
			{			
				if((i+j+k)%confparams_cpr->sampleDistance==0)
				{
					index = i*r23+j*r3+k;
					pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r23] 
					- oriData[index-1-r23] - oriData[index-r3-1] - oriData[index-r3-r23] + oriData[index-r3-r23-1];
					pred_err = fabs(pred_value - oriData[index]);
					radiusIndex = (pred_err/realPrecision+1)/2;
					if(radiusIndex>=confparams_cpr->maxRangeRadius)
					{
						radiusIndex = confparams_cpr->maxRangeRadius - 1;
						//printf("radiusIndex=%d\n", radiusIndex);
					}
					intervals[radiusIndex]++;

					//	if (max < oriData[index]) max = oriData[index];
					//	if (min > oriData[index]) min = oriData[index];
				}
			}
		}
	}
	//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;
	
	//	struct timeval costStart, costEnd;
	//	double cost_est = 0;
	//
	//	gettimeofday(&costStart, NULL);
	//
	//	//compute estimate of bit-rate and distortion
	//	double est_br = 0;
	//	double est_psnr = 0;
	//	double c1 = log2(targetCount)+1;
	//	double c2 = -20.0*log10(realPrecision) + 20.0*log10(max-min) + 10.0*log10(3);
	//
	//	for (i = 0; i < powerOf2/2; i++)
	//	{
	//		int count = intervals[i];
	//		if (count != 0)
	//			est_br += count*log2(count);
	//		est_psnr += count;
	//	}
	//
	//	//compute estimate of bit-rate
	//	est_br -= c1*est_psnr;
	//	est_br /= totalSampleSize;
	//	est_br = -est_br;
	//
	//	//compute estimate of psnr
	//	est_psnr /= totalSampleSize;
	//	printf ("sum of P(i) = %lf\n", est_psnr);
	//	est_psnr = -10.0*log10(est_psnr);
	//	est_psnr += c2;
	//
	//	printf ("estimate bitrate = %.2f\n", est_br);
	//	printf ("estimate psnr = %.2f\n",est_psnr);
	//
	//	gettimeofday(&costEnd, NULL);
	//	cost_est = ((costEnd.tv_sec*1000000+costEnd.tv_usec)-(costStart.tv_sec*1000000+costStart.tv_usec))/1000000.0;
	//
	//	printf ("analysis time = %f\n", cost_est);

	free(intervals);
	//printf("targetCount=%d, sum=%d, totalSampleSize=%d, ratio=%f, accIntervals=%d, powerOf2=%d\n", targetCount, sum, totalSampleSize, (double)sum/(double)totalSampleSize, accIntervals, powerOf2);
	return powerOf2;
}


unsigned int optimize_intervals_float_4D(float *oriData, size_t r1, size_t r2, size_t r3, size_t r4, double realPrecision)
{
	size_t i,j,k,l, index;
	size_t radiusIndex;
	size_t r234=r2*r3*r4;
	size_t r34=r3*r4;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = (r1-1)*(r2-1)*(r3-1)*(r4-1)/confparams_cpr->sampleDistance;
	for(i=1;i<r1;i++)
	{
		for(j=1;j<r2;j++)
		{
			for(k=1;k<r3;k++)
			{
				for (l=1;l<r4;l++)
				{
					if((i+j+k+l)%confparams_cpr->sampleDistance==0)
					{
						index = i*r234+j*r34+k*r4+l;
						pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r34]
								- oriData[index-1-r34] - oriData[index-r4-1] - oriData[index-r4-r34] + oriData[index-r4-r34-1];
						pred_err = fabs(pred_value - oriData[index]);
						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);
	return powerOf2;
}

TightDataPointStorageF* SZ_compress_float_1D_MDQ(float *oriData, 
size_t dataLength, float realPrecision, float valueRangeSize, float medianValue_f)
{
#ifdef HAVE_TIMECMPR	
	float* decData = NULL;
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
		decData = (float*)(multisteps->hist_data);
#endif	
	
	unsigned int quantization_intervals;
	if(exe_params->optQuantMode==1)
		quantization_intervals = optimize_intervals_float_1D_opt(oriData, dataLength, realPrecision);
	else
		quantization_intervals = exe_params->intvCapacity;
	//updateQuantizationInfo(quantization_intervals);	
	int intvRadius = quantization_intervals/2;

	size_t i;
	int reqLength;
	float medianValue = medianValue_f;
	short radExpo = getExponent_float(valueRangeSize/2);
	
	computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue);	

	int* type = (int*) malloc(dataLength*sizeof(int));
		
	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;
	float last3CmprsData[3] = {0};

	FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
	LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
				
	//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);
	listAdd_float(last3CmprsData, vce->data);
#ifdef HAVE_TIMECMPR	
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
		decData[0] = vce->data;
#endif		
		
	//add the second data
	type[1] = 0;
	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);
#ifdef HAVE_TIMECMPR	
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
		decData[1] = vce->data;
#endif
	int state;
	float checkRadius;
	float curData;
	float pred = last3CmprsData[0];
	float predAbsErr;
	checkRadius = (quantization_intervals-1)*realPrecision;
	float interval = 2*realPrecision;
	
	float recip_precision = 1/realPrecision;
	
	for(i=2;i<dataLength;i++)
	{	
		curData = spaceFillingValue[i];
		//pred = 2*last3CmprsData[0] - last3CmprsData[1];
		//pred = last3CmprsData[0];
		predAbsErr = fabsf(curData - pred);	
		if(predAbsErr<checkRadius)
		{
			state = ((int)(predAbsErr*recip_precision+1))>>1;
			if(curData>=pred)
			{
				type[i] = intvRadius+state;
				pred = pred + state*interval;
			}
			else //curData<pred
			{
				type[i] = intvRadius-state;
				pred = pred - state*interval;
			}
				
			//double-check the prediction error in case of machine-epsilon impact	
			if(fabs(curData-pred)>realPrecision)
			{	
				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(last3CmprsData, vce->data);	
				pred = vce->data;
#ifdef HAVE_TIMECMPR					
				if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
					decData[i] = vce->data;
#endif					
			}
			else
			{
				//listAdd_float(last3CmprsData, pred);
#ifdef HAVE_TIMECMPR					
				if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
					decData[i] = pred;			
#endif	
			}	
			continue;
		}
		
		//unpredictable data processing		
		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(last3CmprsData, vce->data);
		pred = vce->data;
#ifdef HAVE_TIMECMPR
		if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
			decData[i] = vce->data;
#endif	
		
	}//end of for
		
//	char* expSegmentsInBytes;
//	int expSegmentsInBytes_size = convertESCToBytes(esc, &expSegmentsInBytes);
	size_t exactDataNum = exactLeadNumArray->size;
	
	TightDataPointStorageF* tdps;
			
	new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, 
			type, exactMidByteArray->array, exactMidByteArray->size,  
			exactLeadNumArray->array,  
			resiBitArray->array, resiBitArray->size, 
			resiBitsLength,
			realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0);

//sdi:Debug
/*	int sum =0;
	for(i=0;i<dataLength;i++)
		if(type[i]==0) sum++;
	printf("opt_quantizations=%d, exactDataNum=%zu, sum=%d\n",quantization_intervals, exactDataNum, sum);
*/	
	//free memory
	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;
}

void SZ_compress_args_float_StoreOriData(float* oriData, size_t dataLength, unsigned char** newByteData, size_t *outSize)
{	
	int floatSize=sizeof(float);	
	size_t k = 0, i;
	size_t totalByteLength = 3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1 + floatSize*dataLength;
	/*No need to malloc because newByteData should always already be allocated with no less totalByteLength.*/
	//*newByteData = (unsigned char*)malloc(totalByteLength);
	
	unsigned char dsLengthBytes[8];
	for (i = 0; i < 3; i++)//3
		(*newByteData)[k++] = versionNumber[i];

	if(exe_params->SZ_SIZE_TYPE==4)//1
		(*newByteData)[k++] = 16; //00010000
	else
		(*newByteData)[k++] = 80;	//01010000: 01000000 indicates the SZ_SIZE_TYPE=8
	
	convertSZParamsToBytes(confparams_cpr, &((*newByteData)[k]));
	k = k + MetaDataByteLength;	
	
	sizeToBytes(dsLengthBytes,dataLength); //SZ_SIZE_TYPE: 4 or 8	
	for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++)
		(*newByteData)[k++] = dsLengthBytes[i];
		
	if(sysEndianType==BIG_ENDIAN_SYSTEM)
		memcpy((*newByteData)+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE, oriData, dataLength*floatSize);
	else
	{
		unsigned char* p = (*newByteData)+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE;
		for(i=0;i<dataLength;i++,p+=floatSize)
			floatToBytes(p, oriData[i]);
	}	
	*outSize = totalByteLength;
}

char SZ_compress_args_float_NoCkRngeNoGzip_1D(int cmprType, unsigned char** newByteData, float *oriData, 
size_t dataLength, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f)
{		
	char compressionType = 0;	
	TightDataPointStorageF* tdps = NULL;	

#ifdef HAVE_TIMECMPR
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
	{
		int timestep = sz_tsc->currentStep;
		if(cmprType == SZ_PERIO_TEMPORAL_COMPRESSION)
		{
			if(timestep % confparams_cpr->snapshotCmprStep != 0)
			{
				tdps = SZ_compress_float_1D_MDQ_ts(oriData, dataLength, multisteps, realPrecision, valueRangeSize, medianValue_f);
				compressionType = 1; //time-series based compression 
			}
			else
			{	
				tdps = SZ_compress_float_1D_MDQ(oriData, dataLength, realPrecision, valueRangeSize, medianValue_f);
				compressionType = 0; //snapshot-based compression
				multisteps->lastSnapshotStep = timestep;
			}
		}
		else if(cmprType == SZ_FORCE_SNAPSHOT_COMPRESSION)
		{
			tdps = SZ_compress_float_1D_MDQ(oriData, dataLength, realPrecision, valueRangeSize, medianValue_f);
			compressionType = 0; //snapshot-based compression
			multisteps->lastSnapshotStep = timestep;			
		}
		else if(cmprType == SZ_FORCE_TEMPORAL_COMPRESSION)
		{
			tdps = SZ_compress_float_1D_MDQ_ts(oriData, dataLength, multisteps, realPrecision, valueRangeSize, medianValue_f);
			compressionType = 1; //time-series based compression 			
		}		
	}
	else
#endif
		tdps = SZ_compress_float_1D_MDQ(oriData, dataLength, realPrecision, 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);
	return compressionType;
}

T
tickduan 已提交
610 611 612
/*MSST19*/
TightDataPointStorageF* SZ_compress_float_1D_MDQ_MSST19(float *oriData, 
size_t dataLength, double realPrecision, float valueRangeSize, float medianValue_f)
T
tickduan 已提交
613
{
T
tickduan 已提交
614 615
#ifdef HAVE_TIMECMPR	
	float* decData = NULL;
T
tickduan 已提交
616 617 618
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
		decData = (float*)(multisteps->hist_data);
#endif	
T
tickduan 已提交
619 620 621

	//struct ClockPoint clockPointBuild;
	//TimeDurationStart("build", &clockPointBuild);
T
tickduan 已提交
622 623
	unsigned int quantization_intervals;
	if(exe_params->optQuantMode==1)
T
tickduan 已提交
624
		quantization_intervals = optimize_intervals_float_1D_opt_MSST19(oriData, dataLength, realPrecision);
T
tickduan 已提交
625 626
	else
		quantization_intervals = exe_params->intvCapacity;
T
tickduan 已提交
627
	//updateQuantizationInfo(quantization_intervals);
T
tickduan 已提交
628 629
	int intvRadius = quantization_intervals/2;
	
T
tickduan 已提交
630 631 632 633 634 635 636 637 638 639 640 641 642
	double* precisionTable = (double*)malloc(sizeof(double) * quantization_intervals);
	double inv = 2.0-pow(2, -(confparams_cpr->plus_bits));
    for(int i=0; i<quantization_intervals; i++){
        double test = pow((1+realPrecision), inv*(i - intvRadius));
        precisionTable[i] = test;
//        if(i>30000 && i<40000)
//			printf("%d %.30G\n", i, test);
    }
    //float smallest_precision = precisionTable[0], largest_precision = precisionTable[quantization_intervals-1];
	struct TopLevelTableWideInterval levelTable;
    MultiLevelCacheTableWideIntervalBuild(&levelTable, precisionTable, quantization_intervals, realPrecision, confparams_cpr->plus_bits);

	size_t i;
T
tickduan 已提交
643 644
	int reqLength;
	float medianValue = medianValue_f;
T
tickduan 已提交
645 646 647 648
	//float medianInverse = 1 / medianValue_f;
	//short radExpo = getExponent_float(valueRangeSize/2);
	
	reqLength = computeReqLength_float_MSST19(realPrecision);	
T
tickduan 已提交
649 650 651 652

	int* type = (int*) malloc(dataLength*sizeof(int));
		
	float* spaceFillingValue = oriData; //
T
tickduan 已提交
653
	
T
tickduan 已提交
654
	DynamicIntArray *exactLeadNumArray;
T
tickduan 已提交
655
	new_DIA(&exactLeadNumArray, dataLength/2/8);
T
tickduan 已提交
656 657
	
	DynamicByteArray *exactMidByteArray;
T
tickduan 已提交
658
	new_DBA(&exactMidByteArray, dataLength/2);
T
tickduan 已提交
659 660 661 662 663 664 665 666 667
	
	DynamicIntArray *resiBitArray;
	new_DIA(&resiBitArray, DynArrayInitLen);
	
	unsigned char preDataBytes[4];
	intToBytes_bigEndian(preDataBytes, 0);
	
	int reqBytesLength = reqLength/8;
	int resiBitsLength = reqLength%8;
T
tickduan 已提交
668 669 670
	float last3CmprsData[3] = {0};

	//size_t miss=0, hit=0;
T
tickduan 已提交
671 672 673

	FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
	LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
T
tickduan 已提交
674 675
				
	//add the first data	
T
tickduan 已提交
676
	type[0] = 0;
T
tickduan 已提交
677
	compressSingleFloatValue_MSST19(vce, spaceFillingValue[0], realPrecision, reqLength, reqBytesLength, resiBitsLength);
T
tickduan 已提交
678 679 680
	updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
	memcpy(preDataBytes,vce->curBytes,4);
	addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
T
tickduan 已提交
681 682
	listAdd_float(last3CmprsData, vce->data);
	//miss++;
T
tickduan 已提交
683 684 685
#ifdef HAVE_TIMECMPR	
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
		decData[0] = vce->data;
T
tickduan 已提交
686 687 688 689 690 691 692 693 694 695
#endif		
		
	//add the second data
	type[1] = 0;
	compressSingleFloatValue_MSST19(vce, spaceFillingValue[1], realPrecision, 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);
	//miss++;
T
tickduan 已提交
696 697
#ifdef HAVE_TIMECMPR	
	if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
T
tickduan 已提交
698
		decData[1] = vce->data;
T
tickduan 已提交
699
#endif
T
tickduan 已提交
700 701 702 703
	int state;
	//double checkRadius;
	float curData;
	float pred = vce->data;
T
tickduan 已提交
704

T
tickduan 已提交
705
    double predRelErrRatio;
T
tickduan 已提交
706

T
tickduan 已提交
707 708 709 710 711 712 713 714 715
	const uint64_t top = levelTable.topIndex, base = levelTable.baseIndex;
	const uint64_t range = top - base;
	const int bits = levelTable.bits;
	uint64_t* const buffer = (uint64_t*)&predRelErrRatio;
	const int shift = 52-bits;
	uint64_t expoIndex, mantiIndex;
	uint16_t* tables[range+1];
	for(int i=0; i<=range; i++){
		tables[i] = levelTable.subTables[i].table;
T
tickduan 已提交
716 717
	}

T
tickduan 已提交
718 719 720 721
	for(i=2;i<dataLength;i++)
	{
		curData = spaceFillingValue[i];
		predRelErrRatio = curData / pred;
T
tickduan 已提交
722

T
tickduan 已提交
723 724 725 726 727 728
		expoIndex = ((*buffer & 0x7fffffffffffffff) >> 52) - base;
		if(expoIndex <= range){
			mantiIndex = (*buffer & 0x000fffffffffffff) >> shift;
			state = tables[expoIndex][mantiIndex];
		}else{
			state = 0;
T
tickduan 已提交
729
		}
T
tickduan 已提交
730 731

		if(state)
T
tickduan 已提交
732
		{
T
tickduan 已提交
733 734 735 736
			type[i] = state;
			pred *= precisionTable[state];
			//hit++;
			continue;
T
tickduan 已提交
737 738
		}

T
tickduan 已提交
739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
		//unpredictable data processing
		type[i] = 0;
		compressSingleFloatValue_MSST19(vce, curData, realPrecision, reqLength, reqBytesLength, resiBitsLength);
		updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
		memcpy(preDataBytes,vce->curBytes,4);
		addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
		pred =  vce->data;
		//miss++;
#ifdef HAVE_TIMECMPR
		if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
			decData[i] = vce->data;
#endif	
		
	}//end of for
		
//	printf("miss:%d, hit:%d\n", miss, hit);
T
tickduan 已提交
755 756 757 758 759 760 761 762 763

	size_t exactDataNum = exactLeadNumArray->size;
	
	TightDataPointStorageF* tdps;
			
	new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, 
			type, exactMidByteArray->array, exactMidByteArray->size,  
			exactLeadNumArray->array,  
			resiBitArray->array, resiBitArray->size, 
T
tickduan 已提交
764
			resiBitsLength,
T
tickduan 已提交
765
			realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0);
T
tickduan 已提交
766
    tdps->plus_bits = confparams_cpr->plus_bits;
T
tickduan 已提交
767 768 769 770
	
	//free memory
	free_DIA(exactLeadNumArray);
	free_DIA(resiBitArray);
T
tickduan 已提交
771
	free(type);	
T
tickduan 已提交
772
	free(vce);
T
tickduan 已提交
773
	free(lce);	
T
tickduan 已提交
774
	free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps);
T
tickduan 已提交
775 776 777
	free(precisionTable);
	freeTopLevelTableWideInterval(&levelTable);
	return tdps;
T
tickduan 已提交
778 779 780
}


T
tickduan 已提交
781 782 783 784 785 786 787
void SZ_compress_args_float_withinRange(unsigned char** newByteData, float *oriData, size_t dataLength, size_t *outSize)
{
	TightDataPointStorageF* tdps = (TightDataPointStorageF*) malloc(sizeof(TightDataPointStorageF));
	tdps->rtypeArray = NULL;
	tdps->typeArray = NULL;	
	tdps->leadNumArray = NULL;
	tdps->residualMidBits = NULL;
T
tickduan 已提交
788
	
T
tickduan 已提交
789 790 791 792 793 794 795 796
	tdps->allSameData = 1;
	tdps->dataSeriesLength = dataLength;
	tdps->exactMidBytes = (unsigned char*)malloc(sizeof(unsigned char)*4);
	tdps->pwrErrBoundBytes = NULL;
	tdps->isLossless = 0;
	float value = oriData[0];
	floatToBytes(tdps->exactMidBytes, value);
	tdps->exactMidBytes_size = 4;
T
tickduan 已提交
797
	
T
tickduan 已提交
798 799 800 801 802 803 804 805
	size_t tmpOutSize;
	//unsigned char *tmpByteData;
	convertTDPStoFlatBytes_float(tdps, newByteData, &tmpOutSize);

	//*newByteData = (unsigned char*)malloc(sizeof(unsigned char)*12); //for floating-point data (1+3+4+4)
	//memcpy(*newByteData, tmpByteData, 12);
	*outSize = tmpOutSize; //8+SZ_SIZE_TYPE; //8==3+1+4(float_size)
	free_TightDataPointStorageF(tdps);	
T
tickduan 已提交
806 807
}

T
tickduan 已提交
808 809 810
int SZ_compress_args_float(int cmprType, int withRegression, unsigned char** newByteData, float *oriData, 
size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, size_t *outSize, 
int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRatio)
T
tickduan 已提交
811
{
T
tickduan 已提交
812 813
	confparams_cpr->errorBoundMode = errBoundMode; //this is used to print the metadata if needed...
	if(errBoundMode==PW_REL)
T
tickduan 已提交
814
	{
T
tickduan 已提交
815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839
		confparams_cpr->pw_relBoundRatio = pwRelBoundRatio;	
	}
	int status = SZ_SCES;
	size_t dataLength = computeDataLength(r5,r4,r3,r2,r1);
	
	if(dataLength <= MIN_NUM_OF_ELEMENTS)
	{
		*newByteData = SZ_skip_compress_float(oriData, dataLength, outSize);
		return status;
	}
	
	float valueRangeSize = 0, medianValue = 0;
	
	unsigned char * signs = NULL;
	bool positive = true;
	float nearZero = 0.0;
	float min = 0;
	if(pwRelBoundRatio < 0.000009999)
		confparams_cpr->accelerate_pw_rel_compression = 0;
	if(confparams_cpr->errorBoundMode == PW_REL && confparams_cpr->accelerate_pw_rel_compression)
	{
		signs = (unsigned char *) malloc(dataLength);
		memset(signs, 0, dataLength);
		min = computeRangeSize_float_MSST19(oriData, dataLength, &valueRangeSize, &medianValue, signs, &positive, &nearZero);
	}
T
tickduan 已提交
840
	else
T
tickduan 已提交
841 842 843 844 845 846 847 848
		min = computeRangeSize_float(oriData, dataLength, &valueRangeSize, &medianValue);	
	float max = min+valueRangeSize;
	confparams_cpr->fmin = min;
	confparams_cpr->fmax = max;
	
	double realPrecision = 0; 
	
	if(confparams_cpr->errorBoundMode==PSNR)
T
tickduan 已提交
849
	{
T
tickduan 已提交
850 851 852 853 854 855 856 857 858
		confparams_cpr->errorBoundMode = ABS;
		realPrecision = confparams_cpr->absErrBound = computeABSErrBoundFromPSNR(confparams_cpr->psnr, (double)confparams_cpr->predThreshold, (double)valueRangeSize);
		//printf("realPrecision=%lf\n", realPrecision);
	}
	else if(confparams_cpr->errorBoundMode==NORM) //norm error = sqrt(sum((xi-xi_)^2))
	{
		confparams_cpr->errorBoundMode = ABS;
		realPrecision = confparams_cpr->absErrBound = computeABSErrBoundFromNORM_ERR(confparams_cpr->normErr, dataLength);
		//printf("realPrecision=%lf\n", realPrecision);				
T
tickduan 已提交
859 860 861
	}
	else
	{
T
tickduan 已提交
862 863 864 865 866 867 868 869
		realPrecision = getRealPrecision_float(valueRangeSize, errBoundMode, absErr_Bound, relBoundRatio, &status);
		confparams_cpr->absErrBound = realPrecision;
	}	
	if(valueRangeSize <= realPrecision)
	{
		if(confparams_cpr->errorBoundMode>=PW_REL && confparams_cpr->accelerate_pw_rel_compression == 1)
			free(signs);		
		SZ_compress_args_float_withinRange(newByteData, oriData, dataLength, outSize);
T
tickduan 已提交
870
	}
T
tickduan 已提交
871
	else
T
tickduan 已提交
872
	{
T
tickduan 已提交
873 874 875 876
		size_t tmpOutSize = 0;
		unsigned char* tmpByteData;
		
		if (r2==0)
T
tickduan 已提交
877
		{
T
tickduan 已提交
878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905
			if(confparams_cpr->errorBoundMode>=PW_REL)
			{
				if(confparams_cpr->accelerate_pw_rel_compression && confparams_cpr->maxRangeRadius <= 32768)
					SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log_MSST19(&tmpByteData, oriData, pwRelBoundRatio, r1, &tmpOutSize, valueRangeSize, medianValue, signs, &positive, min, max, nearZero);
				else
					SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r1, &tmpOutSize, min, max);
					//SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio, valueRangeSize, medianValue, &tmpOutSize);
			}
			else
#ifdef HAVE_TIMECMPR
				if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
					multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_1D(cmprType, &tmpByteData, oriData, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
				else
#endif				
					{
#ifdef HAVE_RANDOMACCESS						
						if(confparams_cpr->randomAccess == 0)
						{
#endif							
							SZ_compress_args_float_NoCkRngeNoGzip_1D(cmprType, &tmpByteData, oriData, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
							if(tmpOutSize>=dataLength*sizeof(float) + 3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1)
								SZ_compress_args_float_StoreOriData(oriData, dataLength, &tmpByteData, &tmpOutSize);
#ifdef HAVE_RANDOMACCESS
						}
						else
							tmpByteData = SZ_compress_float_1D_MDQ_decompression_random_access_with_blocked_regression(oriData, r1, realPrecision, &tmpOutSize);			
#endif							
					}
T
tickduan 已提交
906 907 908
		}
		else
		{
T
tickduan 已提交
909 910
			printf("Error: doesn't support 5 dimensions for now.\n");
			status = SZ_DERR; //dimension error
T
tickduan 已提交
911 912
		}

T
tickduan 已提交
913 914 915 916
		//
		//Call Zstd or Gzip to do the further compression.
		//
		if(confparams_cpr->szMode==SZ_BEST_SPEED)
T
tickduan 已提交
917
		{
T
tickduan 已提交
918 919
			*outSize = tmpOutSize;
			*newByteData = tmpByteData;
T
tickduan 已提交
920
		}
T
tickduan 已提交
921
		else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION || confparams_cpr->szMode==SZ_TEMPORAL_COMPRESSION)
T
tickduan 已提交
922
		{
T
tickduan 已提交
923 924
			*outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
			free(tmpByteData);
T
tickduan 已提交
925
		}
T
tickduan 已提交
926
		else
T
tickduan 已提交
927
		{
T
tickduan 已提交
928 929
			printf("Error: Wrong setting of confparams_cpr->szMode in the float compression.\n");
			status = SZ_MERR; //mode error			
T
tickduan 已提交
930 931
		}
	}
T
tickduan 已提交
932 933 934
	
	return status;
}
T
tickduan 已提交
935

T
tickduan 已提交
936 937 938 939 940 941 942 943 944 945
//TODO
int SZ_compress_args_float_subblock(unsigned char* compressedBytes, float *oriData,
size_t r5, size_t r4, size_t r3, size_t r2, size_t r1,
size_t s5, size_t s4, size_t s3, size_t s2, size_t s1,
size_t e5, size_t e4, size_t e3, size_t e2, size_t e1,
size_t *outSize, int errBoundMode, double absErr_Bound, double relBoundRatio)
{
	int status = SZ_SCES;
	float valueRangeSize = 0, medianValue = 0;
	computeRangeSize_float_subblock(oriData, &valueRangeSize, &medianValue, r5, r4, r3, r2, r1, s5, s4, s3, s2, s1, e5, e4, e3, e2, e1);
T
tickduan 已提交
946

T
tickduan 已提交
947
	double realPrecision = getRealPrecision_float(valueRangeSize, errBoundMode, absErr_Bound, relBoundRatio, &status);
T
tickduan 已提交
948

T
tickduan 已提交
949
	if(valueRangeSize <= realPrecision)
T
tickduan 已提交
950
	{
T
tickduan 已提交
951 952 953 954 955 956
		//TODO
		//SZ_compress_args_float_withinRange_subblock();
	}
	else
	{
		if (r2==0)
T
tickduan 已提交
957
		{
T
tickduan 已提交
958 959 960 961 962 963 964 965
			if(errBoundMode>=PW_REL)
			{
				//TODO
				//SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_subblock();
				printf ("Current subblock version does not support point-wise relative error bound.\n");
			}
			else
				SZ_compress_args_float_NoCkRnge_1D_subblock(compressedBytes, oriData, realPrecision, outSize, valueRangeSize, medianValue, r1, s1, e1);
T
tickduan 已提交
966 967 968
		}
		else
		{
T
tickduan 已提交
969 970
			printf("Error: doesn't support 5 dimensions for now.\n");
			status = SZ_DERR; //dimension error
T
tickduan 已提交
971 972
		}
	}
T
tickduan 已提交
973 974
	return status;
}
T
tickduan 已提交
975

T
tickduan 已提交
976 977 978 979
void SZ_compress_args_float_NoCkRnge_1D_subblock(unsigned char* compressedBytes, float *oriData, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f,
size_t r1, size_t s1, size_t e1)
{
	TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ_subblock(oriData, realPrecision, valueRangeSize, medianValue_f, r1, s1, e1);
T
tickduan 已提交
980

T
tickduan 已提交
981 982 983 984 985 986 987 988 989 990 991 992 993 994
	if (confparams_cpr->szMode==SZ_BEST_SPEED)
		convertTDPStoFlatBytes_float_args(tdps, compressedBytes, outSize);
	else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
	{
		unsigned char *tmpCompBytes;
		size_t tmpOutSize;
		convertTDPStoFlatBytes_float(tdps, &tmpCompBytes, &tmpOutSize);
		*outSize = zlib_compress3(tmpCompBytes, tmpOutSize, compressedBytes, confparams_cpr->gzipMode);
		free(tmpCompBytes);
	}
	else
	{
		printf ("Error: Wrong setting of confparams_cpr->szMode in the double compression.\n");
	}
T
tickduan 已提交
995

T
tickduan 已提交
996 997 998
	//TODO
//	if(*outSize>dataLength*sizeof(float))
//		SZ_compress_args_float_StoreOriData(oriData, dataLength, newByteData, outSize);
T
tickduan 已提交
999

T
tickduan 已提交
1000
	free_TightDataPointStorageF(tdps);
T
tickduan 已提交
1001 1002
}

T
tickduan 已提交
1003 1004

unsigned int optimize_intervals_float_1D_subblock(float *oriData, double realPrecision, size_t r1, size_t s1, size_t e1)
T
tickduan 已提交
1005
{
T
tickduan 已提交
1006 1007
	size_t dataLength = e1 - s1 + 1;
	oriData = oriData + s1;
T
tickduan 已提交
1008

T
tickduan 已提交
1009 1010 1011 1012 1013 1014 1015
	size_t i = 0;
	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));
	size_t totalSampleSize = dataLength/confparams_cpr->sampleDistance;
	for(i=2;i<dataLength;i++)
T
tickduan 已提交
1016
	{
T
tickduan 已提交
1017
		if(i%confparams_cpr->sampleDistance==0)
T
tickduan 已提交
1018
		{
T
tickduan 已提交
1019 1020 1021 1022 1023 1024 1025
			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]++;
T
tickduan 已提交
1026 1027
		}
	}
T
tickduan 已提交
1028 1029 1030 1031
	//compute the appropriate number
	size_t targetCount = totalSampleSize*confparams_cpr->predThreshold;
	size_t sum = 0;
	for(i=0;i<confparams_cpr->maxRangeRadius;i++)
T
tickduan 已提交
1032
	{
T
tickduan 已提交
1033 1034 1035
		sum += intervals[i];
		if(sum>targetCount)
			break;
T
tickduan 已提交
1036
	}
T
tickduan 已提交
1037 1038
	if(i>=confparams_cpr->maxRangeRadius)
		i = confparams_cpr->maxRangeRadius-1;
T
tickduan 已提交
1039

T
tickduan 已提交
1040 1041 1042 1043 1044 1045 1046 1047 1048
	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
1049 1050
}

T
tickduan 已提交
1051 1052 1053

TightDataPointStorageF* SZ_compress_float_1D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f,
size_t r1, size_t s1, size_t e1)
T
tickduan 已提交
1054
{
T
tickduan 已提交
1055
	size_t dataLength = e1 - s1 + 1;
T
tickduan 已提交
1056 1057
	unsigned int quantization_intervals;
	if(exe_params->optQuantMode==1)
T
tickduan 已提交
1058
		quantization_intervals = optimize_intervals_float_1D_subblock(oriData, realPrecision, r1, s1, e1);
T
tickduan 已提交
1059 1060
	else
		quantization_intervals = exe_params->intvCapacity;
T
tickduan 已提交
1061
	//updateQuantizationInfo(quantization_intervals);
T
tickduan 已提交
1062 1063
	int intvRadius = quantization_intervals/2;

T
tickduan 已提交
1064
	size_t i; 
T
tickduan 已提交
1065 1066 1067
	int reqLength;
	float medianValue = medianValue_f;
	short radExpo = getExponent_float(valueRangeSize/2);
T
tickduan 已提交
1068

T
tickduan 已提交
1069 1070 1071 1072
	computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue);

	int* type = (int*) malloc(dataLength*sizeof(int));

T
tickduan 已提交
1073
	float* spaceFillingValue = oriData + s1;
T
tickduan 已提交
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083

	DynamicIntArray *exactLeadNumArray;
	new_DIA(&exactLeadNumArray, DynArrayInitLen);

	DynamicByteArray *exactMidByteArray;
	new_DBA(&exactMidByteArray, DynArrayInitLen);

	DynamicIntArray *resiBitArray;
	new_DIA(&resiBitArray, DynArrayInitLen);

T
tickduan 已提交
1084 1085
	type[0] = 0;

T
tickduan 已提交
1086 1087 1088 1089 1090
	unsigned char preDataBytes[4];
	intToBytes_bigEndian(preDataBytes, 0);

	int reqBytesLength = reqLength/8;
	int resiBitsLength = reqLength%8;
T
tickduan 已提交
1091
	float last3CmprsData[3] = {0};
T
tickduan 已提交
1092 1093 1094 1095

	FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
	LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));

T
tickduan 已提交
1096 1097 1098 1099 1100 1101
	//add the first data
	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);
T
tickduan 已提交
1102

T
tickduan 已提交
1103 1104 1105 1106 1107 1108 1109
	//add the second data
	type[1] = 0;
	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);
T
tickduan 已提交
1110

T
tickduan 已提交
1111 1112 1113 1114 1115 1116 1117
	int state;
	double checkRadius;
	float curData;
	float pred;
	float predAbsErr;
	checkRadius = (quantization_intervals-1)*realPrecision;
	double interval = 2*realPrecision;
T
tickduan 已提交
1118

T
tickduan 已提交
1119 1120 1121 1122 1123 1124
	for(i=2;i<dataLength;i++)
	{
		curData = spaceFillingValue[i];
		pred = 2*last3CmprsData[0] - last3CmprsData[1];
		predAbsErr = fabs(curData - pred);
		if(predAbsErr<=checkRadius)
T
tickduan 已提交
1125
		{
T
tickduan 已提交
1126 1127
			state = (predAbsErr/realPrecision+1)/2;
			if(curData>=pred)
T
tickduan 已提交
1128
			{
T
tickduan 已提交
1129 1130
				type[i] = intvRadius+state;
				pred = pred + state*interval;
T
tickduan 已提交
1131 1132 1133
			}
			else
			{
T
tickduan 已提交
1134 1135
				type[i] = intvRadius-state;
				pred = pred - state*interval;
T
tickduan 已提交
1136 1137
			}

T
tickduan 已提交
1138 1139 1140
			listAdd_float(last3CmprsData, pred);
			continue;
		}
T
tickduan 已提交
1141

T
tickduan 已提交
1142 1143 1144 1145 1146 1147
		//unpredictable data processing
		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);
T
tickduan 已提交
1148

T
tickduan 已提交
1149 1150
		listAdd_float(last3CmprsData, vce->data);
	}
T
tickduan 已提交
1151

T
tickduan 已提交
1152
	size_t exactDataNum = exactLeadNumArray->size;
T
tickduan 已提交
1153

T
tickduan 已提交
1154
	TightDataPointStorageF* tdps;
T
tickduan 已提交
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173

	new_TightDataPointStorageF(&tdps, dataLength, exactDataNum,
			type, exactMidByteArray->array, exactMidByteArray->size,
			exactLeadNumArray->array,
			resiBitArray->array, resiBitArray->size,
			resiBitsLength,
			realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0);

	//free memory
	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
1174 1175 1176 1177 1178 1179 1180 1181
unsigned int optimize_intervals_float_1D_opt_MSST19(float *oriData, size_t dataLength, double realPrecision)
{	
	size_t i = 0, radiusIndex;
	float pred_value = 0;
	double pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = 0;//dataLength/confparams_cpr->sampleDistance;
T
tickduan 已提交
1182

T
tickduan 已提交
1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198
	float * data_pos = oriData + 2;
	float divider = log2(1+realPrecision)*2;
	int tempIndex = 0;
	while(data_pos - oriData < dataLength){
		if(*data_pos == 0){
    		data_pos += confparams_cpr->sampleDistance;
            continue;
		}	
	    tempIndex++;
		totalSampleSize++;
		pred_value = data_pos[-1];
		pred_err = fabs((double)*data_pos / pred_value);
		radiusIndex = (unsigned long)fabs(log2(pred_err)/divider+0.5);
		if(radiusIndex>=confparams_cpr->maxRangeRadius)
			radiusIndex = confparams_cpr->maxRangeRadius - 1;			
		intervals[radiusIndex]++;
T
tickduan 已提交
1199

T
tickduan 已提交
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218
		data_pos += confparams_cpr->sampleDistance;
	}
	//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;
T
tickduan 已提交
1219
	
T
tickduan 已提交
1220 1221
	free(intervals);
	return powerOf2;
T
tickduan 已提交
1222 1223 1224
}


T
tickduan 已提交
1225 1226 1227 1228 1229 1230 1231
unsigned int optimize_intervals_float_1D_opt(float *oriData, size_t dataLength, double realPrecision)
{	
	size_t i = 0, radiusIndex;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
	size_t totalSampleSize = 0;//dataLength/confparams_cpr->sampleDistance;
T
tickduan 已提交
1232

T
tickduan 已提交
1233 1234 1235 1236 1237 1238 1239 1240 1241
	float * data_pos = oriData + 2;
	while(data_pos - oriData < dataLength){
		totalSampleSize++;
		pred_value = data_pos[-1];
		pred_err = fabs(pred_value - *data_pos);
		radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
		if(radiusIndex>=confparams_cpr->maxRangeRadius)
			radiusIndex = confparams_cpr->maxRangeRadius - 1;			
		intervals[radiusIndex]++;
T
tickduan 已提交
1242

T
tickduan 已提交
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255
		data_pos += confparams_cpr->sampleDistance;
	}
	//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;
T
tickduan 已提交
1256
		
T
tickduan 已提交
1257 1258
	unsigned int accIntervals = 2*(i+1);
	unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
T
tickduan 已提交
1259
	
T
tickduan 已提交
1260 1261
	if(powerOf2<32)
		powerOf2 = 32;
T
tickduan 已提交
1262
	
T
tickduan 已提交
1263 1264 1265
	free(intervals);
	return powerOf2;
}
T
tickduan 已提交
1266 1267 1268



T
tickduan 已提交
1269
size_t SZ_compress_float_1D_MDQ_RA_block(float * block_ori_data, float * mean, size_t dim_0, size_t block_dim_0, double realPrecision, int * type, float * unpredictable_data){
T
tickduan 已提交
1270

T
tickduan 已提交
1271 1272
	mean[0] = block_ori_data[0];
	unsigned short unpredictable_count = 0;
T
tickduan 已提交
1273 1274

	float curData;
T
tickduan 已提交
1275 1276 1277 1278 1279 1280 1281 1282
	double itvNum;
	double diff;
	float last_over_thres = mean[0];
	float pred1D;
	size_t type_index = 0;
	float * data_pos = block_ori_data;
	for(size_t i=0; i<block_dim_0; i++){
		curData = *data_pos;
T
tickduan 已提交
1283

T
tickduan 已提交
1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294
		pred1D = last_over_thres;
		diff = curData - pred1D;
		itvNum = fabs(diff)/realPrecision + 1;
		if (itvNum < exe_params->intvCapacity){
			if (diff < 0) itvNum = -itvNum;
			type[type_index] = (int) (itvNum/2) + exe_params->intvRadius;	
			last_over_thres = pred1D + 2 * (type[type_index] - exe_params->intvRadius) * realPrecision;
			if(fabs(curData-last_over_thres)>realPrecision){
				type[type_index] = 0;
				last_over_thres = curData;
				unpredictable_data[unpredictable_count ++] = curData;
T
tickduan 已提交
1295 1296 1297
			}

		}
T
tickduan 已提交
1298 1299 1300 1301 1302 1303 1304
		else{
			type[type_index] = 0;
			unpredictable_data[unpredictable_count ++] = curData;
			last_over_thres = curData;
		}
		type_index ++;
		data_pos ++;
T
tickduan 已提交
1305
	}
T
tickduan 已提交
1306
	return unpredictable_count;
T
tickduan 已提交
1307 1308 1309

}

T
tickduan 已提交
1310 1311 1312 1313 1314 1315
/*The above code is for sz 1.4.13; the following code is for sz 2.0*/
static unsigned int optimize_intervals_float_1D_with_freq_and_dense_pos(float *oriData, size_t r1, double realPrecision, float * dense_pos, float * max_freq, float * mean_freq)
{	
	float mean = 0.0;
	size_t len = r1;
	size_t mean_distance = (int) (sqrt(len));
T
tickduan 已提交
1316 1317

	float * data_pos = oriData;
T
tickduan 已提交
1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328
	size_t mean_count = 0;
	while(data_pos - oriData < len){
		mean += *data_pos;
		mean_count ++;
		data_pos += mean_distance;
	}
	if(mean_count > 0) mean /= mean_count;
	size_t range = 8192;
	size_t radius = 4096;
	size_t * freq_intervals = (size_t *) malloc(range*sizeof(size_t));
	memset(freq_intervals, 0, range*sizeof(size_t));
T
tickduan 已提交
1329

T
tickduan 已提交
1330 1331 1332
	unsigned int maxRangeRadius = confparams_cpr->maxRangeRadius;
	int sampleDistance = confparams_cpr->sampleDistance;
	float predThreshold = confparams_cpr->predThreshold;
T
tickduan 已提交
1333

T
tickduan 已提交
1334 1335 1336 1337 1338
	size_t i;
	size_t radiusIndex;
	float pred_value = 0, pred_err;
	size_t *intervals = (size_t*)malloc(maxRangeRadius*sizeof(size_t));
	memset(intervals, 0, maxRangeRadius*sizeof(size_t));
T
tickduan 已提交
1339

T
tickduan 已提交
1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352
	float mean_diff;
	ptrdiff_t freq_index;
	size_t freq_count = 0;
	size_t sample_count = 0;
	data_pos = oriData + 1;
	while(data_pos - oriData < len){
		pred_value = data_pos[-1];
		pred_err = fabs(pred_value - *data_pos);
		if(pred_err < realPrecision) freq_count ++;
		radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
		if(radiusIndex>=maxRangeRadius)
			radiusIndex = maxRangeRadius - 1;
		intervals[radiusIndex]++;
T
tickduan 已提交
1353

T
tickduan 已提交
1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364
		mean_diff = *data_pos - mean;
		if(mean_diff > 0) freq_index = (ptrdiff_t)(mean_diff/realPrecision) + radius;
		else freq_index = (ptrdiff_t)(mean_diff/realPrecision) - 1 + radius;
		if(freq_index <= 0){
			freq_intervals[0] ++;
		}
		else if(freq_index >= range){
			freq_intervals[range - 1] ++;
		}
		else{
			freq_intervals[freq_index] ++;
T
tickduan 已提交
1365
		}
T
tickduan 已提交
1366 1367
		data_pos += sampleDistance;
		sample_count ++;
T
tickduan 已提交
1368
	}
T
tickduan 已提交
1369
	*max_freq = freq_count * 1.0/ sample_count;
T
tickduan 已提交
1370

T
tickduan 已提交
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
	//compute the appropriate number
	size_t targetCount = sample_count*predThreshold;
	size_t sum = 0;
	for(i=0;i<maxRangeRadius;i++)
	{
		sum += intervals[i];
		if(sum>targetCount)
			break;
	}
	if(i>=maxRangeRadius)
		i = maxRangeRadius-1;
	unsigned int accIntervals = 2*(i+1);
	unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
T
tickduan 已提交
1384

T
tickduan 已提交
1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397
	if(powerOf2<32)
		powerOf2 = 32;

	// collect frequency
	size_t max_sum = 0;
	size_t max_index = 0;
	size_t tmp_sum;
	size_t * freq_pos = freq_intervals + 1;
	for(size_t i=1; i<range-2; i++){
		tmp_sum = freq_pos[0] + freq_pos[1];
		if(tmp_sum > max_sum){
			max_sum = tmp_sum;
			max_index = i;
T
tickduan 已提交
1398
		}
T
tickduan 已提交
1399
		freq_pos ++;
T
tickduan 已提交
1400
	}
T
tickduan 已提交
1401 1402
	*dense_pos = mean + realPrecision * (ptrdiff_t)(max_index + 1 - radius);
	*mean_freq = max_sum * 1.0 / sample_count;
T
tickduan 已提交
1403

T
tickduan 已提交
1404 1405 1406
	free(freq_intervals);
	free(intervals);
	return powerOf2;
T
tickduan 已提交
1407 1408
}

T
tickduan 已提交
1409 1410 1411

// random access
unsigned char * SZ_compress_float_1D_MDQ_decompression_random_access_with_blocked_regression(float *oriData, size_t r1, double realPrecision, size_t * comp_size){
T
tickduan 已提交
1412 1413 1414 1415 1416 1417 1418 1419

	unsigned int quantization_intervals;
	float sz_sample_correct_freq = -1;//0.5; //-1
	float dense_pos;
	float mean_flush_freq;
	unsigned char use_mean = 0;

	// calculate block dims
T
tickduan 已提交
1420 1421
	size_t num_x;
	size_t block_size = 256;
T
tickduan 已提交
1422 1423
	num_x = (r1 - 1) / block_size + 1;

T
tickduan 已提交
1424 1425 1426
	size_t max_num_block_elements = block_size;
	size_t num_blocks = num_x;
	size_t num_elements = r1;
T
tickduan 已提交
1427 1428 1429 1430 1431 1432 1433 1434

	int * result_type = (int *) malloc(num_blocks*max_num_block_elements * sizeof(int));
	size_t unpred_data_max_size = max_num_block_elements;
	float * result_unpredictable_data = (float *) malloc(unpred_data_max_size * sizeof(float) * num_blocks);
	size_t total_unpred = 0;
	size_t unpredictable_count;
	float * data_pos = oriData;
	int * type = result_type;
T
tickduan 已提交
1435
	float * reg_params = (float *) malloc(num_blocks * 2 * sizeof(float));
T
tickduan 已提交
1436 1437 1438
	float * reg_params_pos = reg_params;
	// move regression part out
	size_t params_offset_b = num_blocks;
T
tickduan 已提交
1439
	float * pred_buffer = (float *) malloc((block_size+1)*sizeof(float));
T
tickduan 已提交
1440 1441 1442
	float * pred_buffer_pos = NULL;
	float * block_data_pos_x = NULL;
	for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462
		data_pos = oriData + i*block_size;
		pred_buffer_pos = pred_buffer;
		block_data_pos_x = data_pos;
		// use the buffer as block_size
		for(int ii=0; ii<block_size; ii++){
			*pred_buffer_pos = *block_data_pos_x;
			pred_buffer_pos ++;
			if(i*block_size + ii + 1 < r1) block_data_pos_x ++;
		}
		/*Calculate regression coefficients*/
		{
			float * cur_data_pos = pred_buffer;
			float fx = 0.0;
			float f = 0;
			float curData;
			for(size_t i=0; i<block_size; i++){
				curData = *cur_data_pos;
				fx += curData * i;
				f += curData;
				cur_data_pos ++;
T
tickduan 已提交
1463
			}
T
tickduan 已提交
1464 1465 1466
			float coeff = 1.0 / block_size;
			reg_params_pos[0] = (2 * fx / (block_size - 1) - f) * 6 * coeff / (block_size + 1);
			reg_params_pos[params_offset_b] = f * coeff - (block_size - 1) * reg_params_pos[0] / 2;
T
tickduan 已提交
1467
		}
T
tickduan 已提交
1468
		reg_params_pos ++;
T
tickduan 已提交
1469 1470 1471
	}
	if(exe_params->optQuantMode==1)
	{
T
tickduan 已提交
1472
		quantization_intervals = optimize_intervals_float_1D_with_freq_and_dense_pos(oriData, r1, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
T
tickduan 已提交
1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503
		if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
		updateQuantizationInfo(quantization_intervals);
	}	
	else{
		quantization_intervals = exe_params->intvCapacity;
	}

	float mean = 0;
	if(use_mean){
		// compute mean
		double sum = 0.0;
		size_t mean_count = 0;
		for(size_t i=0; i<num_elements; i++){
			if(fabs(oriData[i] - dense_pos) < realPrecision){
				sum += oriData[i];
				mean_count ++;
			}
		}
		if(mean_count > 0) mean = sum / mean_count;
	}

	double tmp_realPrecision = realPrecision;

	// use two prediction buffers for higher performance
	float * unpredictable_data = result_unpredictable_data;
	unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
	memset(indicator, 0, num_blocks * sizeof(unsigned char));
	unsigned char * indicator_pos = indicator;

	int intvCapacity = quantization_intervals; //exe_params->intvCapacity;
	int intvRadius = intvCapacity/2; //exe_params->intvRadius;	
T
tickduan 已提交
1504
	float noise = realPrecision * 0.5;
T
tickduan 已提交
1505 1506
	reg_params_pos = reg_params;

T
tickduan 已提交
1507
	memset(pred_buffer, 0, (block_size+1)*sizeof(float));
T
tickduan 已提交
1508
	// select
T
tickduan 已提交
1509
	int sample_distance = sqrt(block_size) + 1;
T
tickduan 已提交
1510 1511
	if(use_mean){
		for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538
			data_pos = oriData + i*block_size;
			// add 1 in x, y offset
			pred_buffer_pos = pred_buffer + 1;
			block_data_pos_x = data_pos;
			for(int ii=0; ii<block_size; ii++){
				*pred_buffer_pos = *block_data_pos_x;
				pred_buffer_pos ++;
				if(i*block_size + ii + 1< r1) block_data_pos_x ++;
			}
			/*sampling and decide which predictor*/
			{
				float * cur_data_pos;
				float curData;
				float pred_reg, pred_sz;
				float err_sz = 0.0, err_reg = 0.0;
				for(int i=2; i<=block_size; i+=sample_distance){
					cur_data_pos = pred_buffer + i;
					curData = *cur_data_pos;
					pred_sz = cur_data_pos[-1];
					pred_reg = reg_params_pos[0] * (i-1) + reg_params_pos[params_offset_b];							
					err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
					err_reg += fabs(pred_reg - curData);								
				}
				*indicator_pos = !(err_reg < err_sz);
			}
			reg_params_pos ++;
			indicator_pos ++;
T
tickduan 已提交
1539 1540 1541 1542
		}// end i
	}
	else{
		for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564
			data_pos = oriData + i*block_size;
			// add 1 in x, y offset
			pred_buffer_pos = pred_buffer + 1;
			block_data_pos_x = data_pos;
			for(int ii=0; ii<block_size; ii++){
				*pred_buffer_pos = *block_data_pos_x;
				pred_buffer_pos ++;
				if(i*block_size + ii + 1< r1) block_data_pos_x ++;
			}
			/*sampling and decide which predictor*/
			{
				float * cur_data_pos;
				float curData;
				float pred_reg, pred_sz;
				float err_sz = 0.0, err_reg = 0.0;
				for(int i=2; i<=block_size; i+=sample_distance){
					cur_data_pos = pred_buffer + i;
					curData = *cur_data_pos;
					pred_sz = cur_data_pos[-1];
					pred_reg = reg_params_pos[0] * (i-1) + reg_params_pos[params_offset_b];							
					err_sz += fabs(pred_sz - curData) + noise;
					err_reg += fabs(pred_reg - curData);								
T
tickduan 已提交
1565
				}
T
tickduan 已提交
1566
				*indicator_pos = !(err_reg < err_sz);
T
tickduan 已提交
1567
			}
T
tickduan 已提交
1568 1569 1570
			reg_params_pos ++;
			indicator_pos ++;
		}// end i
T
tickduan 已提交
1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581
	}

	size_t reg_count = 0;
	for(int i=0; i<num_blocks; i++){
		if(!(indicator[i])){
			reg_params[reg_count] = reg_params[i];
			reg_params[reg_count + params_offset_b] = reg_params[i + params_offset_b];
			reg_count ++;
		}
	}
	//Compress coefficient arrays
T
tickduan 已提交
1582 1583
	double precision_a, precision_b;
	float rel_param_err = 0.1/2;
T
tickduan 已提交
1584
	precision_a = rel_param_err * realPrecision / block_size;
T
tickduan 已提交
1585 1586
	precision_b = rel_param_err * realPrecision;
	float last_coeffcients[2] = {0.0};
T
tickduan 已提交
1587 1588
	int coeff_intvCapacity_sz = 65536;
	int coeff_intvRadius = coeff_intvCapacity_sz / 2;
T
tickduan 已提交
1589 1590 1591 1592 1593 1594 1595
	int * coeff_type[2];
	int * coeff_result_type = (int *) malloc(reg_count*2*sizeof(int));
	float * coeff_unpred_data[2];
	float * coeff_unpredictable_data = (float *) malloc(reg_count*2*sizeof(float));
	double precision[2];
	precision[0] = precision_a, precision[1] = precision_b;
	for(int i=0; i<2; i++){
T
tickduan 已提交
1596 1597 1598 1599
		coeff_type[i] = coeff_result_type + i * reg_count;
		coeff_unpred_data[i] = coeff_unpredictable_data + i * reg_count;
	}
	int coeff_index = 0;
T
tickduan 已提交
1600
	unsigned int coeff_unpredictable_count[2] = {0};
T
tickduan 已提交
1601

T
tickduan 已提交
1602 1603
	float * reg_params_separte[2];
	for(int i=0; i<2; i++){
T
tickduan 已提交
1604 1605 1606 1607 1608 1609
		reg_params_separte[i] = reg_params + i * num_blocks;
	}
	for(size_t i=0; i<reg_count; i++){
		// for each coeff
		float cur_coeff;
		double diff, itvNum;
T
tickduan 已提交
1610
		for(int e=0; e<2; e++){
T
tickduan 已提交
1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642
			cur_coeff = reg_params_separte[e][i];
			diff = cur_coeff - last_coeffcients[e];
			itvNum = fabs(diff)/precision[e] + 1;
			if (itvNum < coeff_intvCapacity_sz){
				if (diff < 0) itvNum = -itvNum;
				coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
				last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
				//ganrantee compression error against the case of machine-epsilon
				if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){	
					coeff_type[e][coeff_index] = 0;
					last_coeffcients[e] = cur_coeff;	
					coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
				}					
			}
			else{
				coeff_type[e][coeff_index] = 0;
				last_coeffcients[e] = cur_coeff;
				coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
			}
			reg_params_separte[e][i] = last_coeffcients[e];
		}
		coeff_index ++;
	}
	// pred & quantization
	int * blockwise_unpred_count = (int *) malloc(num_blocks * sizeof(int));
	int * blockwise_unpred_count_pos = blockwise_unpred_count;
	reg_params_pos = reg_params;
	indicator_pos = indicator;
	if(use_mean){
		int intvCapacity_sz = intvCapacity - 2;
		type = result_type;
		for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674
			data_pos = oriData + i*block_size;
			// add 1 in x, y offset
			pred_buffer_pos = pred_buffer + 1;
			block_data_pos_x = data_pos;
			for(int ii=0; ii<block_size; ii++){
				*pred_buffer_pos = *block_data_pos_x;
				pred_buffer_pos ++;
				if(i*block_size + ii + 1< r1) block_data_pos_x ++;
			}
			if(!(*indicator_pos)){
				float curData;
				float pred;
				double itvNum;
				double diff;
				size_t index = 0;
				size_t block_unpredictable_count = 0;
				float * cur_data_pos = pred_buffer + 1;
				for(size_t ii=0; ii<block_size; ii++){
					curData = *cur_data_pos;
					pred = reg_params_pos[0] * ii + reg_params_pos[params_offset_b];									
					diff = curData - pred;
					itvNum = fabs(diff)/tmp_realPrecision + 1;
					if (itvNum < intvCapacity){
						if (diff < 0) itvNum = -itvNum;
						type[index] = (int) (itvNum/2) + intvRadius;
						pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
						//ganrantee comporession error against the case of machine-epsilon
						if(fabs(curData - pred)>tmp_realPrecision){	
							type[index] = 0;
							pred = curData;
							unpredictable_data[block_unpredictable_count ++] = curData;
						}		
T
tickduan 已提交
1675 1676
					}
					else{
T
tickduan 已提交
1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718
						type[index] = 0;
						pred = curData;
						unpredictable_data[block_unpredictable_count ++] = curData;
					}
					index ++;
					cur_data_pos ++;
				}
				reg_params_pos ++;
				total_unpred += block_unpredictable_count;
				unpredictable_data += block_unpredictable_count;
				*blockwise_unpred_count_pos = block_unpredictable_count;
			}
			else{
				// use SZ
				// SZ predication
				unpredictable_count = 0;
				float * cur_data_pos = pred_buffer + 1;
				float curData;
				float pred3D;
				double itvNum, diff;
				size_t index = 0;
				for(size_t ii=0; ii<block_size; ii++){
					curData = *cur_data_pos;
					if(fabs(curData - mean) <= realPrecision){
						type[index] = 1;
						*cur_data_pos = mean;
					}
					else
					{
						pred3D = cur_data_pos[-1];
						diff = curData - pred3D;
						itvNum = fabs(diff)/realPrecision + 1;
						if (itvNum < intvCapacity_sz){
							if (diff < 0) itvNum = -itvNum;
							type[index] = (int) (itvNum/2) + intvRadius;
							*cur_data_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
							//ganrantee comporession error against the case of machine-epsilon
							if(fabs(curData - *cur_data_pos)>tmp_realPrecision){	
								type[index] = 0;
								*cur_data_pos = curData;	
								unpredictable_data[unpredictable_count ++] = curData;
							}					
T
tickduan 已提交
1719
						}
T
tickduan 已提交
1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
						else{
							type[index] = 0;
							*cur_data_pos = curData;
							unpredictable_data[unpredictable_count ++] = curData;
						}
					}
					index ++;
					cur_data_pos ++;
				}
				total_unpred += unpredictable_count;
				unpredictable_data += unpredictable_count;
				*blockwise_unpred_count_pos = unpredictable_count;
			}// end SZ
			blockwise_unpred_count_pos ++;
			type += block_size;
			indicator_pos ++;
T
tickduan 已提交
1736 1737 1738
		}// end i
	}
	else{
T
tickduan 已提交
1739
		int intvCapacity_sz = intvCapacity;
T
tickduan 已提交
1740 1741
		type = result_type;
		for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773
			data_pos = oriData + i*block_size;
			// add 1 in x, y offset
			pred_buffer_pos = pred_buffer + 1;
			block_data_pos_x = data_pos;
			for(int ii=0; ii<block_size; ii++){
				*pred_buffer_pos = *block_data_pos_x;
				pred_buffer_pos ++;
				if(i*block_size + ii + 1< r1) block_data_pos_x ++;
			}
			if(!(*indicator_pos)){
				float curData;
				float pred;
				double itvNum;
				double diff;
				size_t index = 0;
				size_t block_unpredictable_count = 0;
				float * cur_data_pos = pred_buffer + 1;
				for(size_t ii=0; ii<block_size; ii++){
					curData = *cur_data_pos;
					pred = reg_params_pos[0] * ii + reg_params_pos[params_offset_b];									
					diff = curData - pred;
					itvNum = fabs(diff)/tmp_realPrecision + 1;
					if (itvNum < intvCapacity){
						if (diff < 0) itvNum = -itvNum;
						type[index] = (int) (itvNum/2) + intvRadius;
						pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
						//ganrantee comporession error against the case of machine-epsilon
						if(fabs(curData - pred)>tmp_realPrecision){	
							type[index] = 0;
							pred = curData;
							unpredictable_data[block_unpredictable_count ++] = curData;
						}		
T
tickduan 已提交
1774 1775
					}
					else{
T
tickduan 已提交
1776 1777 1778 1779 1780 1781
						type[index] = 0;
						pred = curData;
						unpredictable_data[block_unpredictable_count ++] = curData;
					}
					index ++;
					cur_data_pos ++;
T
tickduan 已提交
1782
				}
T
tickduan 已提交
1783 1784 1785 1786
				reg_params_pos ++;
				total_unpred += block_unpredictable_count;
				unpredictable_data += block_unpredictable_count;
				*blockwise_unpred_count_pos = block_unpredictable_count;
T
tickduan 已提交
1787
			}
T
tickduan 已提交
1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828
			else{
				// use SZ
				// SZ predication
				unpredictable_count = 0;
				float * cur_data_pos = pred_buffer + 1;
				float curData;
				float pred3D;
				double itvNum, diff;
				size_t index = 0;
				for(size_t ii=0; ii<block_size; ii++){
					curData = *cur_data_pos;					
					pred3D = cur_data_pos[-1];
					diff = curData - pred3D;
					itvNum = fabs(diff)/realPrecision + 1;
					if (itvNum < intvCapacity_sz){
						if (diff < 0) itvNum = -itvNum;
						type[index] = (int) (itvNum/2) + intvRadius;
						*cur_data_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
						//ganrantee comporession error against the case of machine-epsilon
						if(fabs(curData - *cur_data_pos)>tmp_realPrecision){	
							type[index] = 0;
							*cur_data_pos = curData;	
							unpredictable_data[unpredictable_count ++] = curData;
						}					
					}
					else{
						type[index] = 0;
						*cur_data_pos = curData;
						unpredictable_data[unpredictable_count ++] = curData;
					}
					index ++;
					cur_data_pos ++;
				}
				total_unpred += unpredictable_count;
				unpredictable_data += unpredictable_count;
				*blockwise_unpred_count_pos = unpredictable_count;
			}// end SZ
			blockwise_unpred_count_pos ++;
			type += block_size;
			indicator_pos ++;
		}// end i
T
tickduan 已提交
1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
	}	
	free(pred_buffer);
	int stateNum = 2*quantization_intervals;
	HuffmanTree* huffmanTree = createHuffmanTree(stateNum);

	size_t nodeCount = 0;
	init(huffmanTree, result_type, num_blocks*max_num_block_elements);
	size_t i = 0;
	for (i = 0; i < huffmanTree->stateNum; i++)
		if (huffmanTree->code[i]) nodeCount++; 
	nodeCount = nodeCount*2-1;

	unsigned char *treeBytes;
	unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);

	unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
	// total size 										metadata		  # elements     real precision		intervals	nodeCount		huffman 	 	block index 						unpredicatable count						mean 					 	unpred size 				elements
T
tickduan 已提交
1846
	unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + 5*treeByteSize +4*num_blocks*sizeof(int) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(float) + total_unpred * sizeof(float) + num_elements * sizeof(int), 1);
T
tickduan 已提交
1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877
	unsigned char * result_pos = result;
	initRandomAccessBytes(result_pos);
	
	result_pos += meta_data_offset;
	
	sizeToBytes(result_pos,num_elements); //SZ_SIZE_TYPE: 4 or 8
	result_pos += exe_params->SZ_SIZE_TYPE;

	intToBytes_bigEndian(result_pos, block_size);
	result_pos += sizeof(int);
	doubleToBytes(result_pos, realPrecision);
	result_pos += sizeof(double);
	intToBytes_bigEndian(result_pos, quantization_intervals);
	result_pos += sizeof(int);
	intToBytes_bigEndian(result_pos, treeByteSize);
	result_pos += sizeof(int);
	intToBytes_bigEndian(result_pos, nodeCount);
	result_pos += sizeof(int);
	memcpy(result_pos, treeBytes, treeByteSize);
	result_pos += treeByteSize;
	free(treeBytes);

	memcpy(result_pos, &use_mean, sizeof(unsigned char));
	result_pos += sizeof(unsigned char);
	memcpy(result_pos, &mean, sizeof(float));
	result_pos += sizeof(float);
	size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
	result_pos += indicator_size;
	
	//convert the lead/mid/resi to byte stream
	if(reg_count > 0){
T
tickduan 已提交
1878
		for(int e=0; e<2; e++){
T
tickduan 已提交
1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
			int stateNum = 2*coeff_intvCapacity_sz;
			HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
			size_t nodeCount = 0;
			init(huffmanTree, coeff_type[e], reg_count);
			size_t i = 0;
			for (i = 0; i < huffmanTree->stateNum; i++)
				if (huffmanTree->code[i]) nodeCount++; 
			nodeCount = nodeCount*2-1;
			unsigned char *treeBytes;
			unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
			doubleToBytes(result_pos, precision[e]);
			result_pos += sizeof(double);
			intToBytes_bigEndian(result_pos, coeff_intvRadius);
			result_pos += sizeof(int);
			intToBytes_bigEndian(result_pos, treeByteSize);
			result_pos += sizeof(int);
			intToBytes_bigEndian(result_pos, nodeCount);
			result_pos += sizeof(int);
			memcpy(result_pos, treeBytes, treeByteSize);		
			result_pos += treeByteSize;
			free(treeBytes);
			size_t typeArray_size = 0;
			encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
			sizeToBytes(result_pos, typeArray_size);
			result_pos += sizeof(size_t) + typeArray_size;
			intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
			result_pos += sizeof(int);
			memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(float));
			result_pos += coeff_unpredictable_count[e]*sizeof(float);
			SZ_ReleaseHuffman(huffmanTree);
		}
	}
	free(coeff_result_type);
	free(coeff_unpredictable_data);
	
	//record the number of unpredictable data and also store them
	memcpy(result_pos, &total_unpred, sizeof(size_t));
	result_pos += sizeof(size_t);
	// record blockwise unpred data
	size_t compressed_blockwise_unpred_count_size;
	unsigned char * compressed_bw_unpred_count = SZ_compress_args(SZ_INT32, blockwise_unpred_count, &compressed_blockwise_unpred_count_size, ABS, 0.5, 0, 0, 0, 0, 0, 0, num_blocks);
	memcpy(result_pos, &compressed_blockwise_unpred_count_size, sizeof(size_t));
	result_pos += sizeof(size_t);
	memcpy(result_pos, compressed_bw_unpred_count, compressed_blockwise_unpred_count_size);
	result_pos += compressed_blockwise_unpred_count_size;
	free(blockwise_unpred_count);
	free(compressed_bw_unpred_count);
	memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(float));
	result_pos += total_unpred * sizeof(float);

	free(reg_params);
	free(indicator);
	free(result_unpredictable_data);
	// encode type array by block
	type = result_type;
	size_t total_type_array_size = 0;
	unsigned char * type_array_buffer = (unsigned char *) malloc(num_blocks*max_num_block_elements*sizeof(int));
	unsigned short * type_array_block_size = (unsigned short *) malloc(num_blocks*sizeof(unsigned short));
	unsigned char * type_array_buffer_pos = type_array_buffer;
	unsigned short * type_array_block_size_pos = type_array_block_size;
T
tickduan 已提交
1939

T
tickduan 已提交
1940
	for(size_t i=0; i<num_x; i++){
T
tickduan 已提交
1941 1942 1943 1944 1945 1946 1947
		size_t typeArray_size = 0;
		encode(huffmanTree, type, max_num_block_elements, type_array_buffer_pos, &typeArray_size);
		total_type_array_size += typeArray_size;
		*type_array_block_size_pos = typeArray_size;
		type_array_buffer_pos += typeArray_size;
		type += max_num_block_elements;
		type_array_block_size_pos ++;
T
tickduan 已提交
1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969
	}
	size_t compressed_type_array_block_size;
	unsigned char * compressed_type_array_block = SZ_compress_args(SZ_UINT16, type_array_block_size, &compressed_type_array_block_size, ABS, 0.5, 0, 0, 0, 0, 0, 0, num_blocks);
	memcpy(result_pos, &compressed_type_array_block_size, sizeof(size_t));
	result_pos += sizeof(size_t);
	memcpy(result_pos, compressed_type_array_block, compressed_type_array_block_size);
	result_pos += compressed_type_array_block_size;
	memcpy(result_pos, type_array_buffer, total_type_array_size);
	result_pos += total_type_array_size;
	// size_t typeArray_size = 0;
	// encode(huffmanTree, result_type, num_blocks*max_num_block_elements, result_pos, &typeArray_size);
	// result_pos += typeArray_size;

	free(compressed_type_array_block);
	free(type_array_buffer);
	free(type_array_block_size);
	size_t totalEncodeSize = result_pos - result;
	free(result_type);
	SZ_ReleaseHuffman(huffmanTree);
	*comp_size = totalEncodeSize;
	return result;
}