1 | /** |
---|
2 | * @file sz_float.c |
---|
3 | * @author Sheng Di and Dingwen Tao |
---|
4 | * @date Aug, 2016 |
---|
5 | * @brief SZ_Init, Compression and Decompression functions |
---|
6 | * (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory. |
---|
7 | * See COPYRIGHT in top-level directory. |
---|
8 | */ |
---|
9 | |
---|
10 | |
---|
11 | #include <stdio.h> |
---|
12 | #include <stdlib.h> |
---|
13 | #include <string.h> |
---|
14 | #include <unistd.h> |
---|
15 | #include <math.h> |
---|
16 | #include "sz.h" |
---|
17 | #include "CompressElement.h" |
---|
18 | #include "DynamicByteArray.h" |
---|
19 | #include "DynamicIntArray.h" |
---|
20 | #include "TightDataPointStorageF.h" |
---|
21 | #include "sz_float.h" |
---|
22 | #include "sz_float_pwr.h" |
---|
23 | #include "szd_float.h" |
---|
24 | #include "szd_float_pwr.h" |
---|
25 | #include "zlib.h" |
---|
26 | #include "rw.h" |
---|
27 | #include "sz_float_ts.h" |
---|
28 | |
---|
29 | unsigned char* SZ_skip_compress_float(float* data, size_t dataLength, size_t* outSize) |
---|
30 | { |
---|
31 | *outSize = dataLength*sizeof(float); |
---|
32 | unsigned char* out = (unsigned char*)malloc(dataLength*sizeof(float)); |
---|
33 | memcpy(out, data, dataLength*sizeof(float)); |
---|
34 | return out; |
---|
35 | } |
---|
36 | unsigned int optimize_intervals_float_1D(float *oriData, size_t dataLength, double realPrecision) |
---|
37 | { |
---|
38 | size_t i = 0, radiusIndex; |
---|
39 | float pred_value = 0, pred_err; |
---|
40 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
41 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
42 | size_t totalSampleSize = dataLength/confparams_cpr->sampleDistance; |
---|
43 | for(i=2;i<dataLength;i++) |
---|
44 | { |
---|
45 | if(i%confparams_cpr->sampleDistance==0) |
---|
46 | { |
---|
47 | //pred_value = 2*oriData[i-1] - oriData[i-2]; |
---|
48 | pred_value = oriData[i-1]; |
---|
49 | pred_err = fabs(pred_value - oriData[i]); |
---|
50 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
51 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
52 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
53 | intervals[radiusIndex]++; |
---|
54 | } |
---|
55 | } |
---|
56 | //compute the appropriate number |
---|
57 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
58 | size_t sum = 0; |
---|
59 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
60 | { |
---|
61 | sum += intervals[i]; |
---|
62 | if(sum>targetCount) |
---|
63 | break; |
---|
64 | } |
---|
65 | if(i>=confparams_cpr->maxRangeRadius) |
---|
66 | i = confparams_cpr->maxRangeRadius-1; |
---|
67 | |
---|
68 | unsigned int accIntervals = 2*(i+1); |
---|
69 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
70 | |
---|
71 | if(powerOf2<32) |
---|
72 | powerOf2 = 32; |
---|
73 | |
---|
74 | free(intervals); |
---|
75 | //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2); |
---|
76 | return powerOf2; |
---|
77 | } |
---|
78 | |
---|
79 | unsigned int optimize_intervals_float_2D(float *oriData, size_t r1, size_t r2, double realPrecision) |
---|
80 | { |
---|
81 | size_t i,j, index; |
---|
82 | size_t radiusIndex; |
---|
83 | float pred_value = 0, pred_err; |
---|
84 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
85 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
86 | size_t totalSampleSize = (r1-1)*(r2-1)/confparams_cpr->sampleDistance; |
---|
87 | |
---|
88 | //float max = oriData[0]; |
---|
89 | //float min = oriData[0]; |
---|
90 | |
---|
91 | for(i=1;i<r1;i++) |
---|
92 | { |
---|
93 | for(j=1;j<r2;j++) |
---|
94 | { |
---|
95 | if((i+j)%confparams_cpr->sampleDistance==0) |
---|
96 | { |
---|
97 | index = i*r2+j; |
---|
98 | pred_value = oriData[index-1] + oriData[index-r2] - oriData[index-r2-1]; |
---|
99 | pred_err = fabs(pred_value - oriData[index]); |
---|
100 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
101 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
102 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
103 | intervals[radiusIndex]++; |
---|
104 | |
---|
105 | // if (max < oriData[index]) max = oriData[index]; |
---|
106 | // if (min > oriData[index]) min = oriData[index]; |
---|
107 | } |
---|
108 | } |
---|
109 | } |
---|
110 | //compute the appropriate number |
---|
111 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
112 | size_t sum = 0; |
---|
113 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
114 | { |
---|
115 | sum += intervals[i]; |
---|
116 | if(sum>targetCount) |
---|
117 | break; |
---|
118 | } |
---|
119 | if(i>=confparams_cpr->maxRangeRadius) |
---|
120 | i = confparams_cpr->maxRangeRadius-1; |
---|
121 | unsigned int accIntervals = 2*(i+1); |
---|
122 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
123 | |
---|
124 | if(powerOf2<32) |
---|
125 | powerOf2 = 32; |
---|
126 | |
---|
127 | // struct timeval costStart, costEnd; |
---|
128 | // double cost_est = 0; |
---|
129 | // |
---|
130 | // gettimeofday(&costStart, NULL); |
---|
131 | // |
---|
132 | // //compute estimate of bit-rate and distortion |
---|
133 | // double est_br = 0; |
---|
134 | // double est_psnr = 0; |
---|
135 | // double c1 = log2(targetCount)+1; |
---|
136 | // double c2 = -20.0*log10(realPrecision) + 20.0*log10(max-min) + 10.0*log10(3); |
---|
137 | // |
---|
138 | // for (i = 0; i < powerOf2/2; i++) |
---|
139 | // { |
---|
140 | // int count = intervals[i]; |
---|
141 | // if (count != 0) |
---|
142 | // est_br += count*log2(count); |
---|
143 | // est_psnr += count; |
---|
144 | // } |
---|
145 | // |
---|
146 | // //compute estimate of bit-rate |
---|
147 | // est_br -= c1*est_psnr; |
---|
148 | // est_br /= totalSampleSize; |
---|
149 | // est_br = -est_br; |
---|
150 | // |
---|
151 | // //compute estimate of psnr |
---|
152 | // est_psnr /= totalSampleSize; |
---|
153 | // printf ("sum of P(i) = %lf\n", est_psnr); |
---|
154 | // est_psnr = -10.0*log10(est_psnr); |
---|
155 | // est_psnr += c2; |
---|
156 | // |
---|
157 | // printf ("estimate bitrate = %.2f\n", est_br); |
---|
158 | // printf ("estimate psnr = %.2f\n",est_psnr); |
---|
159 | // |
---|
160 | // gettimeofday(&costEnd, NULL); |
---|
161 | // cost_est = ((costEnd.tv_sec*1000000+costEnd.tv_usec)-(costStart.tv_sec*1000000+costStart.tv_usec))/1000000.0; |
---|
162 | // |
---|
163 | // printf ("analysis time = %f\n", cost_est); |
---|
164 | |
---|
165 | free(intervals); |
---|
166 | //printf("confparams_cpr->maxRangeRadius = %d, accIntervals=%d, powerOf2=%d\n", confparams_cpr->maxRangeRadius, accIntervals, powerOf2); |
---|
167 | return powerOf2; |
---|
168 | } |
---|
169 | |
---|
170 | unsigned int optimize_intervals_float_3D(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision) |
---|
171 | { |
---|
172 | size_t i,j,k, index; |
---|
173 | size_t radiusIndex; |
---|
174 | size_t r23=r2*r3; |
---|
175 | float pred_value = 0, pred_err; |
---|
176 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
177 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
178 | size_t totalSampleSize = (r1-1)*(r2-1)*(r3-1)/confparams_cpr->sampleDistance; |
---|
179 | |
---|
180 | //float max = oriData[0]; |
---|
181 | //float min = oriData[0]; |
---|
182 | |
---|
183 | for(i=1;i<r1;i++) |
---|
184 | { |
---|
185 | for(j=1;j<r2;j++) |
---|
186 | { |
---|
187 | for(k=1;k<r3;k++) |
---|
188 | { |
---|
189 | if((i+j+k)%confparams_cpr->sampleDistance==0) |
---|
190 | { |
---|
191 | index = i*r23+j*r3+k; |
---|
192 | pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r23] |
---|
193 | - oriData[index-1-r23] - oriData[index-r3-1] - oriData[index-r3-r23] + oriData[index-r3-r23-1]; |
---|
194 | pred_err = fabs(pred_value - oriData[index]); |
---|
195 | radiusIndex = (pred_err/realPrecision+1)/2; |
---|
196 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
197 | { |
---|
198 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
199 | //printf("radiusIndex=%d\n", radiusIndex); |
---|
200 | } |
---|
201 | intervals[radiusIndex]++; |
---|
202 | |
---|
203 | // if (max < oriData[index]) max = oriData[index]; |
---|
204 | // if (min > oriData[index]) min = oriData[index]; |
---|
205 | } |
---|
206 | } |
---|
207 | } |
---|
208 | } |
---|
209 | //compute the appropriate number |
---|
210 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
211 | size_t sum = 0; |
---|
212 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
213 | { |
---|
214 | sum += intervals[i]; |
---|
215 | if(sum>targetCount) |
---|
216 | break; |
---|
217 | } |
---|
218 | if(i>=confparams_cpr->maxRangeRadius) |
---|
219 | i = confparams_cpr->maxRangeRadius-1; |
---|
220 | unsigned int accIntervals = 2*(i+1); |
---|
221 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
222 | |
---|
223 | if(powerOf2<32) |
---|
224 | powerOf2 = 32; |
---|
225 | |
---|
226 | // struct timeval costStart, costEnd; |
---|
227 | // double cost_est = 0; |
---|
228 | // |
---|
229 | // gettimeofday(&costStart, NULL); |
---|
230 | // |
---|
231 | // //compute estimate of bit-rate and distortion |
---|
232 | // double est_br = 0; |
---|
233 | // double est_psnr = 0; |
---|
234 | // double c1 = log2(targetCount)+1; |
---|
235 | // double c2 = -20.0*log10(realPrecision) + 20.0*log10(max-min) + 10.0*log10(3); |
---|
236 | // |
---|
237 | // for (i = 0; i < powerOf2/2; i++) |
---|
238 | // { |
---|
239 | // int count = intervals[i]; |
---|
240 | // if (count != 0) |
---|
241 | // est_br += count*log2(count); |
---|
242 | // est_psnr += count; |
---|
243 | // } |
---|
244 | // |
---|
245 | // //compute estimate of bit-rate |
---|
246 | // est_br -= c1*est_psnr; |
---|
247 | // est_br /= totalSampleSize; |
---|
248 | // est_br = -est_br; |
---|
249 | // |
---|
250 | // //compute estimate of psnr |
---|
251 | // est_psnr /= totalSampleSize; |
---|
252 | // printf ("sum of P(i) = %lf\n", est_psnr); |
---|
253 | // est_psnr = -10.0*log10(est_psnr); |
---|
254 | // est_psnr += c2; |
---|
255 | // |
---|
256 | // printf ("estimate bitrate = %.2f\n", est_br); |
---|
257 | // printf ("estimate psnr = %.2f\n",est_psnr); |
---|
258 | // |
---|
259 | // gettimeofday(&costEnd, NULL); |
---|
260 | // cost_est = ((costEnd.tv_sec*1000000+costEnd.tv_usec)-(costStart.tv_sec*1000000+costStart.tv_usec))/1000000.0; |
---|
261 | // |
---|
262 | // printf ("analysis time = %f\n", cost_est); |
---|
263 | |
---|
264 | free(intervals); |
---|
265 | //printf("targetCount=%d, sum=%d, totalSampleSize=%d, ratio=%f, accIntervals=%d, powerOf2=%d\n", targetCount, sum, totalSampleSize, (double)sum/(double)totalSampleSize, accIntervals, powerOf2); |
---|
266 | return powerOf2; |
---|
267 | } |
---|
268 | |
---|
269 | |
---|
270 | unsigned int optimize_intervals_float_4D(float *oriData, size_t r1, size_t r2, size_t r3, size_t r4, double realPrecision) |
---|
271 | { |
---|
272 | size_t i,j,k,l, index; |
---|
273 | size_t radiusIndex; |
---|
274 | size_t r234=r2*r3*r4; |
---|
275 | size_t r34=r3*r4; |
---|
276 | float pred_value = 0, pred_err; |
---|
277 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
278 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
279 | size_t totalSampleSize = (r1-1)*(r2-1)*(r3-1)*(r4-1)/confparams_cpr->sampleDistance; |
---|
280 | for(i=1;i<r1;i++) |
---|
281 | { |
---|
282 | for(j=1;j<r2;j++) |
---|
283 | { |
---|
284 | for(k=1;k<r3;k++) |
---|
285 | { |
---|
286 | for (l=1;l<r4;l++) |
---|
287 | { |
---|
288 | if((i+j+k+l)%confparams_cpr->sampleDistance==0) |
---|
289 | { |
---|
290 | index = i*r234+j*r34+k*r4+l; |
---|
291 | pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r34] |
---|
292 | - oriData[index-1-r34] - oriData[index-r4-1] - oriData[index-r4-r34] + oriData[index-r4-r34-1]; |
---|
293 | pred_err = fabs(pred_value - oriData[index]); |
---|
294 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
295 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
296 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
297 | intervals[radiusIndex]++; |
---|
298 | } |
---|
299 | } |
---|
300 | } |
---|
301 | } |
---|
302 | } |
---|
303 | //compute the appropriate number |
---|
304 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
305 | size_t sum = 0; |
---|
306 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
307 | { |
---|
308 | sum += intervals[i]; |
---|
309 | if(sum>targetCount) |
---|
310 | break; |
---|
311 | } |
---|
312 | if(i>=confparams_cpr->maxRangeRadius) |
---|
313 | i = confparams_cpr->maxRangeRadius-1; |
---|
314 | |
---|
315 | unsigned int accIntervals = 2*(i+1); |
---|
316 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
317 | |
---|
318 | if(powerOf2<32) |
---|
319 | powerOf2 = 32; |
---|
320 | |
---|
321 | free(intervals); |
---|
322 | return powerOf2; |
---|
323 | } |
---|
324 | |
---|
325 | TightDataPointStorageF* SZ_compress_float_1D_MDQ(float *oriData, |
---|
326 | size_t dataLength, double realPrecision, float valueRangeSize, float medianValue_f) |
---|
327 | { |
---|
328 | #ifdef HAVE_TIMECMPR |
---|
329 | float* decData = NULL; |
---|
330 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
331 | decData = (float*)(multisteps->hist_data); |
---|
332 | #endif |
---|
333 | |
---|
334 | unsigned int quantization_intervals; |
---|
335 | if(exe_params->optQuantMode==1) |
---|
336 | quantization_intervals = optimize_intervals_float_1D_opt(oriData, dataLength, realPrecision); |
---|
337 | else |
---|
338 | quantization_intervals = exe_params->intvCapacity; |
---|
339 | updateQuantizationInfo(quantization_intervals); |
---|
340 | |
---|
341 | size_t i; |
---|
342 | int reqLength; |
---|
343 | float medianValue = medianValue_f; |
---|
344 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
345 | |
---|
346 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
347 | |
---|
348 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
349 | |
---|
350 | float* spaceFillingValue = oriData; // |
---|
351 | |
---|
352 | DynamicIntArray *exactLeadNumArray; |
---|
353 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
354 | |
---|
355 | DynamicByteArray *exactMidByteArray; |
---|
356 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
357 | |
---|
358 | DynamicIntArray *resiBitArray; |
---|
359 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
360 | |
---|
361 | unsigned char preDataBytes[4]; |
---|
362 | intToBytes_bigEndian(preDataBytes, 0); |
---|
363 | |
---|
364 | int reqBytesLength = reqLength/8; |
---|
365 | int resiBitsLength = reqLength%8; |
---|
366 | float last3CmprsData[3] = {0}; |
---|
367 | |
---|
368 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
369 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
370 | |
---|
371 | //add the first data |
---|
372 | type[0] = 0; |
---|
373 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
374 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
375 | memcpy(preDataBytes,vce->curBytes,4); |
---|
376 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
377 | listAdd_float(last3CmprsData, vce->data); |
---|
378 | #ifdef HAVE_TIMECMPR |
---|
379 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
380 | decData[0] = vce->data; |
---|
381 | #endif |
---|
382 | |
---|
383 | //add the second data |
---|
384 | type[1] = 0; |
---|
385 | compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
386 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
387 | memcpy(preDataBytes,vce->curBytes,4); |
---|
388 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
389 | listAdd_float(last3CmprsData, vce->data); |
---|
390 | #ifdef HAVE_TIMECMPR |
---|
391 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
392 | decData[1] = vce->data; |
---|
393 | #endif |
---|
394 | int state; |
---|
395 | double checkRadius; |
---|
396 | float curData; |
---|
397 | float pred; |
---|
398 | float predAbsErr; |
---|
399 | checkRadius = (exe_params->intvCapacity-1)*realPrecision; |
---|
400 | double interval = 2*realPrecision; |
---|
401 | |
---|
402 | for(i=2;i<dataLength;i++) |
---|
403 | { |
---|
404 | curData = spaceFillingValue[i]; |
---|
405 | //pred = 2*last3CmprsData[0] - last3CmprsData[1]; |
---|
406 | pred = last3CmprsData[0]; |
---|
407 | predAbsErr = fabs(curData - pred); |
---|
408 | if(predAbsErr<=checkRadius) |
---|
409 | { |
---|
410 | state = (predAbsErr/realPrecision+1)/2; |
---|
411 | if(curData>=pred) |
---|
412 | { |
---|
413 | type[i] = exe_params->intvRadius+state; |
---|
414 | pred = pred + state*interval; |
---|
415 | } |
---|
416 | else //curData<pred |
---|
417 | { |
---|
418 | type[i] = exe_params->intvRadius-state; |
---|
419 | pred = pred - state*interval; |
---|
420 | } |
---|
421 | |
---|
422 | //double-check the prediction error in case of machine-epsilon impact |
---|
423 | if(fabs(curData-pred)>realPrecision) |
---|
424 | { |
---|
425 | type[i] = 0; |
---|
426 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
427 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
428 | memcpy(preDataBytes,vce->curBytes,4); |
---|
429 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
430 | |
---|
431 | listAdd_float(last3CmprsData, vce->data); |
---|
432 | #ifdef HAVE_TIMECMPR |
---|
433 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
434 | decData[i] = vce->data; |
---|
435 | #endif |
---|
436 | } |
---|
437 | else |
---|
438 | { |
---|
439 | listAdd_float(last3CmprsData, pred); |
---|
440 | #ifdef HAVE_TIMECMPR |
---|
441 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
442 | decData[i] = pred; |
---|
443 | #endif |
---|
444 | } |
---|
445 | continue; |
---|
446 | } |
---|
447 | |
---|
448 | //unpredictable data processing |
---|
449 | type[i] = 0; |
---|
450 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
451 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
452 | memcpy(preDataBytes,vce->curBytes,4); |
---|
453 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
454 | |
---|
455 | listAdd_float(last3CmprsData, vce->data); |
---|
456 | #ifdef HAVE_TIMECMPR |
---|
457 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
458 | decData[i] = vce->data; |
---|
459 | #endif |
---|
460 | |
---|
461 | }//end of for |
---|
462 | |
---|
463 | // char* expSegmentsInBytes; |
---|
464 | // int expSegmentsInBytes_size = convertESCToBytes(esc, &expSegmentsInBytes); |
---|
465 | size_t exactDataNum = exactLeadNumArray->size; |
---|
466 | |
---|
467 | TightDataPointStorageF* tdps; |
---|
468 | |
---|
469 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
470 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
471 | exactLeadNumArray->array, |
---|
472 | resiBitArray->array, resiBitArray->size, |
---|
473 | resiBitsLength, |
---|
474 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
475 | |
---|
476 | //sdi:Debug |
---|
477 | /* int sum =0; |
---|
478 | for(i=0;i<dataLength;i++) |
---|
479 | if(type[i]==0) sum++; |
---|
480 | printf("opt_quantizations=%d, exactDataNum=%d, sum=%d\n",quantization_intervals, exactDataNum, sum);*/ |
---|
481 | |
---|
482 | //free memory |
---|
483 | free_DIA(exactLeadNumArray); |
---|
484 | free_DIA(resiBitArray); |
---|
485 | free(type); |
---|
486 | free(vce); |
---|
487 | free(lce); |
---|
488 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
489 | |
---|
490 | return tdps; |
---|
491 | } |
---|
492 | |
---|
493 | void SZ_compress_args_float_StoreOriData(float* oriData, size_t dataLength, TightDataPointStorageF* tdps, |
---|
494 | unsigned char** newByteData, size_t *outSize) |
---|
495 | { |
---|
496 | int floatSize=sizeof(float); |
---|
497 | size_t k = 0, i; |
---|
498 | tdps->isLossless = 1; |
---|
499 | size_t totalByteLength = 3 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 1 + floatSize*dataLength; |
---|
500 | *newByteData = (unsigned char*)malloc(totalByteLength); |
---|
501 | |
---|
502 | unsigned char dsLengthBytes[8]; |
---|
503 | for (i = 0; i < 3; i++)//3 |
---|
504 | (*newByteData)[k++] = versionNumber[i]; |
---|
505 | |
---|
506 | if(exe_params->SZ_SIZE_TYPE==4)//1 |
---|
507 | (*newByteData)[k++] = 16; //00010000 |
---|
508 | else |
---|
509 | (*newByteData)[k++] = 80; //01010000: 01000000 indicates the SZ_SIZE_TYPE=8 |
---|
510 | |
---|
511 | convertSZParamsToBytes(confparams_cpr, &((*newByteData)[k])); |
---|
512 | k = k + MetaDataByteLength; |
---|
513 | |
---|
514 | sizeToBytes(dsLengthBytes,dataLength); //SZ_SIZE_TYPE: 4 or 8 |
---|
515 | for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++) |
---|
516 | (*newByteData)[k++] = dsLengthBytes[i]; |
---|
517 | |
---|
518 | if(sysEndianType==BIG_ENDIAN_SYSTEM) |
---|
519 | memcpy((*newByteData)+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE, oriData, dataLength*floatSize); |
---|
520 | else |
---|
521 | { |
---|
522 | unsigned char* p = (*newByteData)+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE; |
---|
523 | for(i=0;i<dataLength;i++,p+=floatSize) |
---|
524 | floatToBytes(p, oriData[i]); |
---|
525 | } |
---|
526 | *outSize = totalByteLength; |
---|
527 | } |
---|
528 | |
---|
529 | char SZ_compress_args_float_NoCkRngeNoGzip_1D(unsigned char** newByteData, float *oriData, |
---|
530 | size_t dataLength, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f) |
---|
531 | { |
---|
532 | char compressionType = 0; |
---|
533 | TightDataPointStorageF* tdps = NULL; |
---|
534 | |
---|
535 | #ifdef HAVE_TIMECMPR |
---|
536 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
537 | { |
---|
538 | int timestep = sz_tsc->currentStep; |
---|
539 | if(timestep % confparams_cpr->snapshotCmprStep != 0) |
---|
540 | { |
---|
541 | tdps = SZ_compress_float_1D_MDQ_ts(oriData, dataLength, multisteps, realPrecision, valueRangeSize, medianValue_f); |
---|
542 | compressionType = 1; //time-series based compression |
---|
543 | } |
---|
544 | else |
---|
545 | { |
---|
546 | tdps = SZ_compress_float_1D_MDQ(oriData, dataLength, realPrecision, valueRangeSize, medianValue_f); |
---|
547 | compressionType = 0; //snapshot-based compression |
---|
548 | multisteps->lastSnapshotStep = timestep; |
---|
549 | } |
---|
550 | } |
---|
551 | else |
---|
552 | #endif |
---|
553 | tdps = SZ_compress_float_1D_MDQ(oriData, dataLength, realPrecision, valueRangeSize, medianValue_f); |
---|
554 | |
---|
555 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
556 | |
---|
557 | if(*outSize>dataLength*sizeof(float)) |
---|
558 | SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize); |
---|
559 | |
---|
560 | free_TightDataPointStorageF(tdps); |
---|
561 | return compressionType; |
---|
562 | } |
---|
563 | |
---|
564 | TightDataPointStorageF* SZ_compress_float_2D_MDQ(float *oriData, size_t r1, size_t r2, double realPrecision, float valueRangeSize, float medianValue_f) |
---|
565 | { |
---|
566 | #ifdef HAVE_TIMECMPR |
---|
567 | float* decData = NULL; |
---|
568 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
569 | decData = (float*)(multisteps->hist_data); |
---|
570 | #endif |
---|
571 | |
---|
572 | unsigned int quantization_intervals; |
---|
573 | if(exe_params->optQuantMode==1) |
---|
574 | { |
---|
575 | quantization_intervals = optimize_intervals_float_2D_opt(oriData, r1, r2, realPrecision); |
---|
576 | updateQuantizationInfo(quantization_intervals); |
---|
577 | } |
---|
578 | else |
---|
579 | quantization_intervals = exe_params->intvCapacity; |
---|
580 | size_t i,j; |
---|
581 | int reqLength; |
---|
582 | float pred1D, pred2D; |
---|
583 | float diff = 0.0; |
---|
584 | double itvNum = 0; |
---|
585 | float *P0, *P1; |
---|
586 | |
---|
587 | size_t dataLength = r1*r2; |
---|
588 | |
---|
589 | P0 = (float*)malloc(r2*sizeof(float)); |
---|
590 | memset(P0, 0, r2*sizeof(float)); |
---|
591 | P1 = (float*)malloc(r2*sizeof(float)); |
---|
592 | memset(P1, 0, r2*sizeof(float)); |
---|
593 | |
---|
594 | float medianValue = medianValue_f; |
---|
595 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
596 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
597 | |
---|
598 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
599 | //type[dataLength]=0; |
---|
600 | |
---|
601 | float* spaceFillingValue = oriData; // |
---|
602 | |
---|
603 | DynamicIntArray *exactLeadNumArray; |
---|
604 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
605 | |
---|
606 | DynamicByteArray *exactMidByteArray; |
---|
607 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
608 | |
---|
609 | DynamicIntArray *resiBitArray; |
---|
610 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
611 | |
---|
612 | type[0] = 0; |
---|
613 | unsigned char preDataBytes[4]; |
---|
614 | intToBytes_bigEndian(preDataBytes, 0); |
---|
615 | |
---|
616 | int reqBytesLength = reqLength/8; |
---|
617 | int resiBitsLength = reqLength%8; |
---|
618 | |
---|
619 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
620 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
621 | |
---|
622 | /* Process Row-0 data 0*/ |
---|
623 | type[0] = 0; |
---|
624 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
625 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
626 | memcpy(preDataBytes,vce->curBytes,4); |
---|
627 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
628 | P1[0] = vce->data; |
---|
629 | #ifdef HAVE_TIMECMPR |
---|
630 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
631 | decData[0] = vce->data; |
---|
632 | #endif |
---|
633 | |
---|
634 | float curData; |
---|
635 | |
---|
636 | /* Process Row-0 data 1*/ |
---|
637 | pred1D = P1[0]; |
---|
638 | curData = spaceFillingValue[1]; |
---|
639 | diff = curData - pred1D; |
---|
640 | |
---|
641 | itvNum = fabs(diff)/realPrecision + 1; |
---|
642 | |
---|
643 | if (itvNum < exe_params->intvCapacity) |
---|
644 | { |
---|
645 | if (diff < 0) itvNum = -itvNum; |
---|
646 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
647 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
648 | |
---|
649 | //ganrantee comporession error against the case of machine-epsilon |
---|
650 | if(fabs(spaceFillingValue[1]-P1[1])>realPrecision) |
---|
651 | { |
---|
652 | type[1] = 0; |
---|
653 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
654 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
655 | memcpy(preDataBytes,vce->curBytes,4); |
---|
656 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
657 | |
---|
658 | P1[1] = vce->data; |
---|
659 | } |
---|
660 | } |
---|
661 | else |
---|
662 | { |
---|
663 | type[1] = 0; |
---|
664 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
665 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
666 | memcpy(preDataBytes,vce->curBytes,4); |
---|
667 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
668 | P1[1] = vce->data; |
---|
669 | } |
---|
670 | #ifdef HAVE_TIMECMPR |
---|
671 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
672 | decData[1] = P1[1]; |
---|
673 | #endif |
---|
674 | |
---|
675 | /* Process Row-0 data 2 --> data r2-1 */ |
---|
676 | for (j = 2; j < r2; j++) |
---|
677 | { |
---|
678 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
679 | curData = spaceFillingValue[j]; |
---|
680 | diff = curData - pred1D; |
---|
681 | |
---|
682 | itvNum = fabs(diff)/realPrecision + 1; |
---|
683 | |
---|
684 | if (itvNum < exe_params->intvCapacity) |
---|
685 | { |
---|
686 | if (diff < 0) itvNum = -itvNum; |
---|
687 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
688 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
689 | |
---|
690 | //ganrantee comporession error against the case of machine-epsilon |
---|
691 | if(fabs(curData-P1[j])>realPrecision) |
---|
692 | { |
---|
693 | type[j] = 0; |
---|
694 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
695 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
696 | memcpy(preDataBytes,vce->curBytes,4); |
---|
697 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
698 | |
---|
699 | P1[j] = vce->data; |
---|
700 | } |
---|
701 | } |
---|
702 | else |
---|
703 | { |
---|
704 | type[j] = 0; |
---|
705 | compressSingleFloatValue(vce,curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
706 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
707 | memcpy(preDataBytes,vce->curBytes,4); |
---|
708 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
709 | P1[j] = vce->data; |
---|
710 | } |
---|
711 | #ifdef HAVE_TIMECMPR |
---|
712 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
713 | decData[j] = P1[j]; |
---|
714 | #endif |
---|
715 | } |
---|
716 | |
---|
717 | /* Process Row-1 --> Row-r1-1 */ |
---|
718 | size_t index; |
---|
719 | for (i = 1; i < r1; i++) |
---|
720 | { |
---|
721 | /* Process row-i data 0 */ |
---|
722 | index = i*r2; |
---|
723 | pred1D = P1[0]; |
---|
724 | curData = spaceFillingValue[index]; |
---|
725 | diff = curData - pred1D; |
---|
726 | |
---|
727 | itvNum = fabs(diff)/realPrecision + 1; |
---|
728 | |
---|
729 | if (itvNum < exe_params->intvCapacity) |
---|
730 | { |
---|
731 | if (diff < 0) itvNum = -itvNum; |
---|
732 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
733 | P0[0] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
734 | |
---|
735 | //ganrantee comporession error against the case of machine-epsilon |
---|
736 | if(fabs(curData-P0[0])>realPrecision) |
---|
737 | { |
---|
738 | type[index] = 0; |
---|
739 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
740 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
741 | memcpy(preDataBytes,vce->curBytes,4); |
---|
742 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
743 | |
---|
744 | P0[0] = vce->data; |
---|
745 | } |
---|
746 | } |
---|
747 | else |
---|
748 | { |
---|
749 | type[index] = 0; |
---|
750 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
751 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
752 | memcpy(preDataBytes,vce->curBytes,4); |
---|
753 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
754 | P0[0] = vce->data; |
---|
755 | } |
---|
756 | #ifdef HAVE_TIMECMPR |
---|
757 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
758 | decData[index] = P0[0]; |
---|
759 | #endif |
---|
760 | |
---|
761 | /* Process row-i data 1 --> r2-1*/ |
---|
762 | for (j = 1; j < r2; j++) |
---|
763 | { |
---|
764 | index = i*r2+j; |
---|
765 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
766 | |
---|
767 | curData = spaceFillingValue[index]; |
---|
768 | diff = curData - pred2D; |
---|
769 | |
---|
770 | itvNum = fabs(diff)/realPrecision + 1; |
---|
771 | |
---|
772 | if (itvNum < exe_params->intvCapacity) |
---|
773 | { |
---|
774 | if (diff < 0) itvNum = -itvNum; |
---|
775 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
776 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
777 | |
---|
778 | //ganrantee comporession error against the case of machine-epsilon |
---|
779 | if(fabs(curData-P0[j])>realPrecision) |
---|
780 | { |
---|
781 | type[index] = 0; |
---|
782 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
783 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
784 | memcpy(preDataBytes,vce->curBytes,4); |
---|
785 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
786 | |
---|
787 | P0[j] = vce->data; |
---|
788 | } |
---|
789 | } |
---|
790 | else |
---|
791 | { |
---|
792 | type[index] = 0; |
---|
793 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
794 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
795 | memcpy(preDataBytes,vce->curBytes,4); |
---|
796 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
797 | P0[j] = vce->data; |
---|
798 | } |
---|
799 | #ifdef HAVE_TIMECMPR |
---|
800 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
801 | decData[index] = P0[j]; |
---|
802 | #endif |
---|
803 | } |
---|
804 | |
---|
805 | float *Pt; |
---|
806 | Pt = P1; |
---|
807 | P1 = P0; |
---|
808 | P0 = Pt; |
---|
809 | } |
---|
810 | |
---|
811 | if(r2!=1) |
---|
812 | free(P0); |
---|
813 | free(P1); |
---|
814 | size_t exactDataNum = exactLeadNumArray->size; |
---|
815 | |
---|
816 | TightDataPointStorageF* tdps; |
---|
817 | |
---|
818 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
819 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
820 | exactLeadNumArray->array, |
---|
821 | resiBitArray->array, resiBitArray->size, |
---|
822 | resiBitsLength, |
---|
823 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
824 | |
---|
825 | // printf("exactDataNum=%d, expSegmentsInBytes_size=%d, exactMidByteArray->size=%d\n", |
---|
826 | // exactDataNum, expSegmentsInBytes_size, exactMidByteArray->size); |
---|
827 | |
---|
828 | // for(i = 3800;i<3844;i++) |
---|
829 | // printf("exactLeadNumArray->array[%d]=%d\n",i,exactLeadNumArray->array[i]); |
---|
830 | |
---|
831 | //free memory |
---|
832 | free_DIA(exactLeadNumArray); |
---|
833 | free_DIA(resiBitArray); |
---|
834 | free(type); |
---|
835 | free(vce); |
---|
836 | free(lce); |
---|
837 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
838 | |
---|
839 | return tdps; |
---|
840 | } |
---|
841 | |
---|
842 | /** |
---|
843 | * |
---|
844 | * Note: @r1 is high dimension |
---|
845 | * @r2 is low dimension |
---|
846 | * */ |
---|
847 | char SZ_compress_args_float_NoCkRngeNoGzip_2D(unsigned char** newByteData, float *oriData, size_t r1, size_t r2, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f) |
---|
848 | { |
---|
849 | size_t dataLength = r1*r2; |
---|
850 | char compressionType = 0; |
---|
851 | TightDataPointStorageF* tdps = NULL; |
---|
852 | |
---|
853 | #ifdef HAVE_TIMECMPR |
---|
854 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
855 | { |
---|
856 | int timestep = sz_tsc->currentStep; |
---|
857 | if(timestep % confparams_cpr->snapshotCmprStep != 0) |
---|
858 | { |
---|
859 | tdps = SZ_compress_float_1D_MDQ_ts(oriData, dataLength, multisteps, realPrecision, valueRangeSize, medianValue_f); |
---|
860 | compressionType = 1; //time-series based compression |
---|
861 | } |
---|
862 | else |
---|
863 | { |
---|
864 | tdps = SZ_compress_float_2D_MDQ(oriData, r1, r2, realPrecision, valueRangeSize, medianValue_f); |
---|
865 | compressionType = 0; //snapshot-based compression |
---|
866 | multisteps->lastSnapshotStep = timestep; |
---|
867 | } |
---|
868 | } |
---|
869 | else |
---|
870 | #endif |
---|
871 | tdps = SZ_compress_float_2D_MDQ(oriData, r1, r2, realPrecision, valueRangeSize, medianValue_f); |
---|
872 | |
---|
873 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
874 | |
---|
875 | if(*outSize>dataLength*sizeof(float)) |
---|
876 | SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
877 | |
---|
878 | free_TightDataPointStorageF(tdps); |
---|
879 | |
---|
880 | return compressionType; |
---|
881 | } |
---|
882 | |
---|
883 | TightDataPointStorageF* SZ_compress_float_3D_MDQ(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, float valueRangeSize, float medianValue_f) |
---|
884 | { |
---|
885 | #ifdef HAVE_TIMECMPR |
---|
886 | float* decData = NULL; |
---|
887 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
888 | decData = (float*)(multisteps->hist_data); |
---|
889 | #endif |
---|
890 | |
---|
891 | unsigned int quantization_intervals; |
---|
892 | if(exe_params->optQuantMode==1) |
---|
893 | { |
---|
894 | quantization_intervals = optimize_intervals_float_3D_opt(oriData, r1, r2, r3, realPrecision); |
---|
895 | updateQuantizationInfo(quantization_intervals); |
---|
896 | } |
---|
897 | else |
---|
898 | quantization_intervals = exe_params->intvCapacity; |
---|
899 | size_t i,j,k; |
---|
900 | int reqLength; |
---|
901 | float pred1D, pred2D, pred3D; |
---|
902 | float diff = 0.0; |
---|
903 | double itvNum = 0; |
---|
904 | float *P0, *P1; |
---|
905 | |
---|
906 | size_t dataLength = r1*r2*r3; |
---|
907 | size_t r23 = r2*r3; |
---|
908 | P0 = (float*)malloc(r23*sizeof(float)); |
---|
909 | P1 = (float*)malloc(r23*sizeof(float)); |
---|
910 | |
---|
911 | float medianValue = medianValue_f; |
---|
912 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
913 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
914 | |
---|
915 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
916 | |
---|
917 | float* spaceFillingValue = oriData; // |
---|
918 | |
---|
919 | DynamicIntArray *exactLeadNumArray; |
---|
920 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
921 | |
---|
922 | DynamicByteArray *exactMidByteArray; |
---|
923 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
924 | |
---|
925 | DynamicIntArray *resiBitArray; |
---|
926 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
927 | |
---|
928 | unsigned char preDataBytes[4]; |
---|
929 | intToBytes_bigEndian(preDataBytes, 0); |
---|
930 | |
---|
931 | int reqBytesLength = reqLength/8; |
---|
932 | int resiBitsLength = reqLength%8; |
---|
933 | |
---|
934 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
935 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
936 | |
---|
937 | |
---|
938 | /////////////////////////// Process layer-0 /////////////////////////// |
---|
939 | /* Process Row-0 data 0*/ |
---|
940 | type[0] = 0; |
---|
941 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
942 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
943 | memcpy(preDataBytes,vce->curBytes,4); |
---|
944 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
945 | P1[0] = vce->data; |
---|
946 | #ifdef HAVE_TIMECMPR |
---|
947 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
948 | decData[0] = P1[0]; |
---|
949 | #endif |
---|
950 | |
---|
951 | float curData; |
---|
952 | |
---|
953 | /* Process Row-0 data 1*/ |
---|
954 | pred1D = P1[0]; |
---|
955 | curData = spaceFillingValue[1]; |
---|
956 | diff = curData - pred1D; |
---|
957 | |
---|
958 | itvNum = fabs(diff)/realPrecision + 1; |
---|
959 | |
---|
960 | if (itvNum < exe_params->intvCapacity) |
---|
961 | { |
---|
962 | if (diff < 0) itvNum = -itvNum; |
---|
963 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
964 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
965 | |
---|
966 | //ganrantee comporession error against the case of machine-epsilon |
---|
967 | if(fabs(curData-P1[1])>realPrecision) |
---|
968 | { |
---|
969 | type[1] = 0; |
---|
970 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
971 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
972 | memcpy(preDataBytes,vce->curBytes,4); |
---|
973 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
974 | |
---|
975 | P1[1] = vce->data; |
---|
976 | } |
---|
977 | } |
---|
978 | else |
---|
979 | { |
---|
980 | type[1] = 0; |
---|
981 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
982 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
983 | memcpy(preDataBytes,vce->curBytes,4); |
---|
984 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
985 | P1[1] = vce->data; |
---|
986 | } |
---|
987 | #ifdef HAVE_TIMECMPR |
---|
988 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
989 | decData[1] = P1[1]; |
---|
990 | #endif |
---|
991 | |
---|
992 | /* Process Row-0 data 2 --> data r3-1 */ |
---|
993 | for (j = 2; j < r3; j++) |
---|
994 | { |
---|
995 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
996 | curData = spaceFillingValue[j]; |
---|
997 | diff = curData - pred1D; |
---|
998 | |
---|
999 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1000 | |
---|
1001 | if (itvNum < exe_params->intvCapacity) |
---|
1002 | { |
---|
1003 | if (diff < 0) itvNum = -itvNum; |
---|
1004 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1005 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
1006 | |
---|
1007 | //ganrantee comporession error against the case of machine-epsilon |
---|
1008 | if(fabs(curData-P1[j])>realPrecision) |
---|
1009 | { |
---|
1010 | type[j] = 0; |
---|
1011 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1012 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1013 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1014 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1015 | |
---|
1016 | P1[j] = vce->data; |
---|
1017 | } |
---|
1018 | } |
---|
1019 | else |
---|
1020 | { |
---|
1021 | type[j] = 0; |
---|
1022 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1023 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1024 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1025 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1026 | P1[j] = vce->data; |
---|
1027 | } |
---|
1028 | #ifdef HAVE_TIMECMPR |
---|
1029 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1030 | decData[j] = P1[j]; |
---|
1031 | #endif |
---|
1032 | } |
---|
1033 | |
---|
1034 | /* Process Row-1 --> Row-r2-1 */ |
---|
1035 | size_t index; |
---|
1036 | for (i = 1; i < r2; i++) |
---|
1037 | { |
---|
1038 | /* Process row-i data 0 */ |
---|
1039 | index = i*r3; |
---|
1040 | pred1D = P1[index-r3]; |
---|
1041 | curData = spaceFillingValue[index]; |
---|
1042 | diff = curData - pred1D; |
---|
1043 | |
---|
1044 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1045 | |
---|
1046 | if (itvNum < exe_params->intvCapacity) |
---|
1047 | { |
---|
1048 | if (diff < 0) itvNum = -itvNum; |
---|
1049 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1050 | P1[index] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1051 | |
---|
1052 | //ganrantee comporession error against the case of machine-epsilon |
---|
1053 | if(fabs(curData-P1[index])>realPrecision) |
---|
1054 | { |
---|
1055 | type[index] = 0; |
---|
1056 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1057 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1058 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1059 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1060 | |
---|
1061 | P1[index] = vce->data; |
---|
1062 | } |
---|
1063 | } |
---|
1064 | else |
---|
1065 | { |
---|
1066 | type[index] = 0; |
---|
1067 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1068 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1069 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1070 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1071 | P1[index] = vce->data; |
---|
1072 | } |
---|
1073 | #ifdef HAVE_TIMECMPR |
---|
1074 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1075 | decData[index] = P1[index]; |
---|
1076 | #endif |
---|
1077 | |
---|
1078 | /* Process row-i data 1 --> data r3-1*/ |
---|
1079 | for (j = 1; j < r3; j++) |
---|
1080 | { |
---|
1081 | index = i*r3+j; |
---|
1082 | pred2D = P1[index-1] + P1[index-r3] - P1[index-r3-1]; |
---|
1083 | |
---|
1084 | curData = spaceFillingValue[index]; |
---|
1085 | diff = curData - pred2D; |
---|
1086 | |
---|
1087 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1088 | |
---|
1089 | if (itvNum < exe_params->intvCapacity) |
---|
1090 | { |
---|
1091 | if (diff < 0) itvNum = -itvNum; |
---|
1092 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1093 | P1[index] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1094 | |
---|
1095 | //ganrantee comporession error against the case of machine-epsilon |
---|
1096 | if(fabs(curData-P1[index])>realPrecision) |
---|
1097 | { |
---|
1098 | type[index] = 0; |
---|
1099 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1100 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1101 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1102 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1103 | |
---|
1104 | P1[index] = vce->data; |
---|
1105 | } |
---|
1106 | } |
---|
1107 | else |
---|
1108 | { |
---|
1109 | type[index] = 0; |
---|
1110 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1111 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1112 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1113 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1114 | P1[index] = vce->data; |
---|
1115 | } |
---|
1116 | #ifdef HAVE_TIMECMPR |
---|
1117 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1118 | decData[index] = P1[index]; |
---|
1119 | #endif |
---|
1120 | } |
---|
1121 | } |
---|
1122 | |
---|
1123 | |
---|
1124 | /////////////////////////// Process layer-1 --> layer-r1-1 /////////////////////////// |
---|
1125 | |
---|
1126 | for (k = 1; k < r1; k++) |
---|
1127 | { |
---|
1128 | /* Process Row-0 data 0*/ |
---|
1129 | index = k*r23; |
---|
1130 | pred1D = P1[0]; |
---|
1131 | curData = spaceFillingValue[index]; |
---|
1132 | diff = curData - pred1D; |
---|
1133 | |
---|
1134 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1135 | |
---|
1136 | if (itvNum < exe_params->intvCapacity) |
---|
1137 | { |
---|
1138 | if (diff < 0) itvNum = -itvNum; |
---|
1139 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1140 | P0[0] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1141 | |
---|
1142 | //ganrantee comporession error against the case of machine-epsilon |
---|
1143 | if(fabs(curData-P0[0])>realPrecision) |
---|
1144 | { |
---|
1145 | type[index] = 0; |
---|
1146 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1147 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1148 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1149 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1150 | |
---|
1151 | P0[0] = vce->data; |
---|
1152 | } |
---|
1153 | } |
---|
1154 | else |
---|
1155 | { |
---|
1156 | type[index] = 0; |
---|
1157 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1158 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1159 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1160 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1161 | P0[0] = vce->data; |
---|
1162 | } |
---|
1163 | #ifdef HAVE_TIMECMPR |
---|
1164 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1165 | decData[index] = P0[0]; |
---|
1166 | #endif |
---|
1167 | |
---|
1168 | /* Process Row-0 data 1 --> data r3-1 */ |
---|
1169 | for (j = 1; j < r3; j++) |
---|
1170 | { |
---|
1171 | //index = k*r2*r3+j; |
---|
1172 | index ++; |
---|
1173 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
1174 | curData = spaceFillingValue[index]; |
---|
1175 | diff = spaceFillingValue[index] - pred2D; |
---|
1176 | |
---|
1177 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1178 | |
---|
1179 | if (itvNum < exe_params->intvCapacity) |
---|
1180 | { |
---|
1181 | if (diff < 0) itvNum = -itvNum; |
---|
1182 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1183 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1184 | //ganrantee comporession error against the case of machine-epsilon |
---|
1185 | if(fabs(curData-P0[j])>realPrecision) |
---|
1186 | { |
---|
1187 | type[index] = 0; |
---|
1188 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1189 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1190 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1191 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1192 | |
---|
1193 | P0[j] = vce->data; |
---|
1194 | } |
---|
1195 | } |
---|
1196 | else |
---|
1197 | { |
---|
1198 | type[index] = 0; |
---|
1199 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1200 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1201 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1202 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1203 | P0[j] = vce->data; |
---|
1204 | } |
---|
1205 | #ifdef HAVE_TIMECMPR |
---|
1206 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1207 | decData[index] = P0[j]; |
---|
1208 | #endif |
---|
1209 | } |
---|
1210 | |
---|
1211 | /* Process Row-1 --> Row-r2-1 */ |
---|
1212 | size_t index2D; |
---|
1213 | for (i = 1; i < r2; i++) |
---|
1214 | { |
---|
1215 | /* Process Row-i data 0 */ |
---|
1216 | index = k*r23 + i*r3; |
---|
1217 | index2D = i*r3; |
---|
1218 | pred2D = P0[index2D-r3] + P1[index2D] - P1[index2D-r3]; |
---|
1219 | curData = spaceFillingValue[index]; |
---|
1220 | diff = spaceFillingValue[index] - pred2D; |
---|
1221 | |
---|
1222 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1223 | |
---|
1224 | if (itvNum < exe_params->intvCapacity) |
---|
1225 | { |
---|
1226 | if (diff < 0) itvNum = -itvNum; |
---|
1227 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1228 | P0[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1229 | //ganrantee comporession error against the case of machine-epsilon |
---|
1230 | if(fabs(curData-P0[index2D])>realPrecision) |
---|
1231 | { |
---|
1232 | type[index] = 0; |
---|
1233 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1234 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1235 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1236 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1237 | |
---|
1238 | P0[index2D] = vce->data; |
---|
1239 | } |
---|
1240 | } |
---|
1241 | else |
---|
1242 | { |
---|
1243 | type[index] = 0; |
---|
1244 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1245 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1246 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1247 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1248 | P0[index2D] = vce->data; |
---|
1249 | } |
---|
1250 | #ifdef HAVE_TIMECMPR |
---|
1251 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1252 | decData[index] = P0[index2D]; |
---|
1253 | #endif |
---|
1254 | |
---|
1255 | /* Process Row-i data 1 --> data r3-1 */ |
---|
1256 | for (j = 1; j < r3; j++) |
---|
1257 | { |
---|
1258 | // if(k==63&&i==43&&j==27) |
---|
1259 | // printf("i=%d\n", i); |
---|
1260 | //index = k*r2*r3 + i*r3 + j; |
---|
1261 | index ++; |
---|
1262 | index2D = i*r3 + j; |
---|
1263 | pred3D = P0[index2D-1] + P0[index2D-r3]+ P1[index2D] - P0[index2D-r3-1] - P1[index2D-r3] - P1[index2D-1] + P1[index2D-r3-1]; |
---|
1264 | curData = spaceFillingValue[index]; |
---|
1265 | diff = curData - pred3D; |
---|
1266 | |
---|
1267 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1268 | |
---|
1269 | if (itvNum < exe_params->intvCapacity) |
---|
1270 | { |
---|
1271 | if (diff < 0) itvNum = -itvNum; |
---|
1272 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1273 | P0[index2D] = pred3D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1274 | |
---|
1275 | //ganrantee comporession error against the case of machine-epsilon |
---|
1276 | if(fabs(curData-P0[index2D])>realPrecision) |
---|
1277 | { |
---|
1278 | type[index] = 0; |
---|
1279 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1280 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1281 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1282 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1283 | |
---|
1284 | P0[index2D] = vce->data; |
---|
1285 | } |
---|
1286 | } |
---|
1287 | else |
---|
1288 | { |
---|
1289 | type[index] = 0; |
---|
1290 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1291 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1292 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1293 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1294 | P0[index2D] = vce->data; |
---|
1295 | } |
---|
1296 | #ifdef HAVE_TIMECMPR |
---|
1297 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1298 | decData[index] = P0[index2D]; |
---|
1299 | #endif |
---|
1300 | } |
---|
1301 | } |
---|
1302 | |
---|
1303 | float *Pt; |
---|
1304 | Pt = P1; |
---|
1305 | P1 = P0; |
---|
1306 | P0 = Pt; |
---|
1307 | } |
---|
1308 | if(r23!=1) |
---|
1309 | free(P0); |
---|
1310 | free(P1); |
---|
1311 | size_t exactDataNum = exactLeadNumArray->size; |
---|
1312 | |
---|
1313 | TightDataPointStorageF* tdps; |
---|
1314 | |
---|
1315 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
1316 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
1317 | exactLeadNumArray->array, |
---|
1318 | resiBitArray->array, resiBitArray->size, |
---|
1319 | resiBitsLength, |
---|
1320 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
1321 | |
---|
1322 | //sdi:Debug |
---|
1323 | /* int sum =0; |
---|
1324 | for(i=0;i<dataLength;i++) |
---|
1325 | if(type[i]==0) sum++; |
---|
1326 | printf("opt_quantizations=%d, exactDataNum=%d, sum=%d\n",quantization_intervals, exactDataNum, sum);*/ |
---|
1327 | |
---|
1328 | |
---|
1329 | // printf("exactDataNum=%d, expSegmentsInBytes_size=%d, exactMidByteArray->size=%d\n", |
---|
1330 | // exactDataNum, expSegmentsInBytes_size, exactMidByteArray->size); |
---|
1331 | |
---|
1332 | //free memory |
---|
1333 | free_DIA(exactLeadNumArray); |
---|
1334 | free_DIA(resiBitArray); |
---|
1335 | free(type); |
---|
1336 | free(vce); |
---|
1337 | free(lce); |
---|
1338 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
1339 | |
---|
1340 | return tdps; |
---|
1341 | } |
---|
1342 | |
---|
1343 | char SZ_compress_args_float_NoCkRngeNoGzip_3D(unsigned char** newByteData, float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f) |
---|
1344 | { |
---|
1345 | size_t dataLength = r1*r2*r3; |
---|
1346 | char compressionType = 0; |
---|
1347 | TightDataPointStorageF* tdps = NULL; |
---|
1348 | |
---|
1349 | #ifdef HAVE_TIMECMPR |
---|
1350 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1351 | { |
---|
1352 | int timestep = sz_tsc->currentStep; |
---|
1353 | if(timestep % confparams_cpr->snapshotCmprStep != 0) |
---|
1354 | { |
---|
1355 | tdps = SZ_compress_float_1D_MDQ_ts(oriData, dataLength, multisteps, realPrecision, valueRangeSize, medianValue_f); |
---|
1356 | compressionType = 1; //time-series based compression |
---|
1357 | } |
---|
1358 | else |
---|
1359 | { |
---|
1360 | tdps = SZ_compress_float_3D_MDQ(oriData, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f); |
---|
1361 | compressionType = 0; //snapshot-based compression |
---|
1362 | multisteps->lastSnapshotStep = timestep; |
---|
1363 | } |
---|
1364 | } |
---|
1365 | else |
---|
1366 | #endif |
---|
1367 | tdps = SZ_compress_float_3D_MDQ(oriData, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f); |
---|
1368 | |
---|
1369 | |
---|
1370 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1371 | |
---|
1372 | if(*outSize>dataLength*sizeof(float)) |
---|
1373 | SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
1374 | |
---|
1375 | free_TightDataPointStorageF(tdps); |
---|
1376 | |
---|
1377 | return compressionType; |
---|
1378 | } |
---|
1379 | |
---|
1380 | |
---|
1381 | TightDataPointStorageF* SZ_compress_float_4D_MDQ(float *oriData, size_t r1, size_t r2, size_t r3, size_t r4, double realPrecision, float valueRangeSize, float medianValue_f) |
---|
1382 | { |
---|
1383 | unsigned int quantization_intervals; |
---|
1384 | if(exe_params->optQuantMode==1) |
---|
1385 | { |
---|
1386 | quantization_intervals = optimize_intervals_float_4D(oriData, r1, r2, r3, r4, realPrecision); |
---|
1387 | updateQuantizationInfo(quantization_intervals); |
---|
1388 | } |
---|
1389 | else |
---|
1390 | quantization_intervals = exe_params->intvCapacity; |
---|
1391 | |
---|
1392 | size_t i,j,k; |
---|
1393 | int reqLength; |
---|
1394 | float pred1D, pred2D, pred3D; |
---|
1395 | float diff = 0.0; |
---|
1396 | double itvNum = 0; |
---|
1397 | float *P0, *P1; |
---|
1398 | |
---|
1399 | size_t dataLength = r1*r2*r3*r4; |
---|
1400 | |
---|
1401 | size_t r234 = r2*r3*r4; |
---|
1402 | size_t r34 = r3*r4; |
---|
1403 | |
---|
1404 | P0 = (float*)malloc(r34*sizeof(float)); |
---|
1405 | P1 = (float*)malloc(r34*sizeof(float)); |
---|
1406 | |
---|
1407 | float medianValue = medianValue_f; |
---|
1408 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
1409 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1410 | |
---|
1411 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
1412 | |
---|
1413 | float* spaceFillingValue = oriData; // |
---|
1414 | |
---|
1415 | DynamicIntArray *exactLeadNumArray; |
---|
1416 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
1417 | |
---|
1418 | DynamicByteArray *exactMidByteArray; |
---|
1419 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
1420 | |
---|
1421 | DynamicIntArray *resiBitArray; |
---|
1422 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
1423 | |
---|
1424 | unsigned char preDataBytes[4]; |
---|
1425 | intToBytes_bigEndian(preDataBytes, 0); |
---|
1426 | |
---|
1427 | int reqBytesLength = reqLength/8; |
---|
1428 | int resiBitsLength = reqLength%8; |
---|
1429 | |
---|
1430 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
1431 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
1432 | |
---|
1433 | |
---|
1434 | size_t l; |
---|
1435 | for (l = 0; l < r1; l++) |
---|
1436 | { |
---|
1437 | |
---|
1438 | /////////////////////////// Process layer-0 /////////////////////////// |
---|
1439 | /* Process Row-0 data 0*/ |
---|
1440 | size_t index = l*r234; |
---|
1441 | size_t index2D = 0; |
---|
1442 | |
---|
1443 | type[index] = 0; |
---|
1444 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1445 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1446 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1447 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1448 | P1[index2D] = vce->data; |
---|
1449 | |
---|
1450 | /* Process Row-0 data 1*/ |
---|
1451 | index = l*r234+1; |
---|
1452 | index2D = 1; |
---|
1453 | |
---|
1454 | pred1D = P1[index2D-1]; |
---|
1455 | diff = spaceFillingValue[index] - pred1D; |
---|
1456 | |
---|
1457 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1458 | |
---|
1459 | if (itvNum < exe_params->intvCapacity) |
---|
1460 | { |
---|
1461 | if (diff < 0) itvNum = -itvNum; |
---|
1462 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1463 | P1[index2D] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1464 | } |
---|
1465 | else |
---|
1466 | { |
---|
1467 | type[index] = 0; |
---|
1468 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1469 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1470 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1471 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1472 | P1[index2D] = vce->data; |
---|
1473 | } |
---|
1474 | |
---|
1475 | /* Process Row-0 data 2 --> data r4-1 */ |
---|
1476 | for (j = 2; j < r4; j++) |
---|
1477 | { |
---|
1478 | index = l*r234+j; |
---|
1479 | index2D = j; |
---|
1480 | |
---|
1481 | pred1D = 2*P1[index2D-1] - P1[index2D-2]; |
---|
1482 | diff = spaceFillingValue[index] - pred1D; |
---|
1483 | |
---|
1484 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1485 | |
---|
1486 | if (itvNum < exe_params->intvCapacity) |
---|
1487 | { |
---|
1488 | if (diff < 0) itvNum = -itvNum; |
---|
1489 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1490 | P1[index2D] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1491 | } |
---|
1492 | else |
---|
1493 | { |
---|
1494 | type[index] = 0; |
---|
1495 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1496 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1497 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1498 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1499 | P1[index2D] = vce->data; |
---|
1500 | } |
---|
1501 | } |
---|
1502 | |
---|
1503 | /* Process Row-1 --> Row-r3-1 */ |
---|
1504 | for (i = 1; i < r3; i++) |
---|
1505 | { |
---|
1506 | /* Process row-i data 0 */ |
---|
1507 | index = l*r234+i*r4; |
---|
1508 | index2D = i*r4; |
---|
1509 | |
---|
1510 | pred1D = P1[index2D-r4]; |
---|
1511 | diff = spaceFillingValue[index] - pred1D; |
---|
1512 | |
---|
1513 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1514 | |
---|
1515 | if (itvNum < exe_params->intvCapacity) |
---|
1516 | { |
---|
1517 | if (diff < 0) itvNum = -itvNum; |
---|
1518 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1519 | P1[index2D] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1520 | } |
---|
1521 | else |
---|
1522 | { |
---|
1523 | type[index] = 0; |
---|
1524 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1525 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1526 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1527 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1528 | P1[index2D] = vce->data; |
---|
1529 | } |
---|
1530 | |
---|
1531 | /* Process row-i data 1 --> data r4-1*/ |
---|
1532 | for (j = 1; j < r4; j++) |
---|
1533 | { |
---|
1534 | index = l*r234+i*r4+j; |
---|
1535 | index2D = i*r4+j; |
---|
1536 | |
---|
1537 | pred2D = P1[index2D-1] + P1[index2D-r4] - P1[index2D-r4-1]; |
---|
1538 | |
---|
1539 | diff = spaceFillingValue[index] - pred2D; |
---|
1540 | |
---|
1541 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1542 | |
---|
1543 | if (itvNum < exe_params->intvCapacity) |
---|
1544 | { |
---|
1545 | if (diff < 0) itvNum = -itvNum; |
---|
1546 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1547 | P1[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1548 | } |
---|
1549 | else |
---|
1550 | { |
---|
1551 | type[index] = 0; |
---|
1552 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1553 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1554 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1555 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1556 | P1[index2D] = vce->data; |
---|
1557 | } |
---|
1558 | } |
---|
1559 | } |
---|
1560 | |
---|
1561 | |
---|
1562 | /////////////////////////// Process layer-1 --> layer-r2-1 /////////////////////////// |
---|
1563 | |
---|
1564 | for (k = 1; k < r2; k++) |
---|
1565 | { |
---|
1566 | /* Process Row-0 data 0*/ |
---|
1567 | index = l*r234+k*r34; |
---|
1568 | index2D = 0; |
---|
1569 | |
---|
1570 | pred1D = P1[index2D]; |
---|
1571 | diff = spaceFillingValue[index] - pred1D; |
---|
1572 | |
---|
1573 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1574 | |
---|
1575 | if (itvNum < exe_params->intvCapacity) |
---|
1576 | { |
---|
1577 | if (diff < 0) itvNum = -itvNum; |
---|
1578 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1579 | P0[index2D] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1580 | } |
---|
1581 | else |
---|
1582 | { |
---|
1583 | type[index] = 0; |
---|
1584 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1585 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1586 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1587 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1588 | P0[index2D] = vce->data; |
---|
1589 | } |
---|
1590 | |
---|
1591 | /* Process Row-0 data 1 --> data r4-1 */ |
---|
1592 | for (j = 1; j < r4; j++) |
---|
1593 | { |
---|
1594 | index = l*r234+k*r34+j; |
---|
1595 | index2D = j; |
---|
1596 | |
---|
1597 | pred2D = P0[index2D-1] + P1[index2D] - P1[index2D-1]; |
---|
1598 | diff = spaceFillingValue[index] - pred2D; |
---|
1599 | |
---|
1600 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1601 | |
---|
1602 | if (itvNum < exe_params->intvCapacity) |
---|
1603 | { |
---|
1604 | if (diff < 0) itvNum = -itvNum; |
---|
1605 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1606 | P0[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1607 | } |
---|
1608 | else |
---|
1609 | { |
---|
1610 | type[index] = 0; |
---|
1611 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1612 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1613 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1614 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1615 | P0[index2D] = vce->data; |
---|
1616 | } |
---|
1617 | } |
---|
1618 | |
---|
1619 | /* Process Row-1 --> Row-r3-1 */ |
---|
1620 | for (i = 1; i < r3; i++) |
---|
1621 | { |
---|
1622 | /* Process Row-i data 0 */ |
---|
1623 | index = l*r234+k*r34+i*r4; |
---|
1624 | index2D = i*r4; |
---|
1625 | |
---|
1626 | pred2D = P0[index2D-r4] + P1[index2D] - P1[index2D-r4]; |
---|
1627 | diff = spaceFillingValue[index] - pred2D; |
---|
1628 | |
---|
1629 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1630 | |
---|
1631 | if (itvNum < exe_params->intvCapacity) |
---|
1632 | { |
---|
1633 | if (diff < 0) itvNum = -itvNum; |
---|
1634 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1635 | P0[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1636 | } |
---|
1637 | else |
---|
1638 | { |
---|
1639 | type[index] = 0; |
---|
1640 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1641 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1642 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1643 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1644 | P0[index2D] = vce->data; |
---|
1645 | } |
---|
1646 | |
---|
1647 | /* Process Row-i data 1 --> data r4-1 */ |
---|
1648 | for (j = 1; j < r4; j++) |
---|
1649 | { |
---|
1650 | index = l*r234+k*r34+i*r4+j; |
---|
1651 | index2D = i*r4+j; |
---|
1652 | |
---|
1653 | pred3D = P0[index2D-1] + P0[index2D-r4]+ P1[index2D] - P0[index2D-r4-1] - P1[index2D-r4] - P1[index2D-1] + P1[index2D-r4-1]; |
---|
1654 | diff = spaceFillingValue[index] - pred3D; |
---|
1655 | |
---|
1656 | |
---|
1657 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1658 | |
---|
1659 | if (itvNum < exe_params->intvCapacity) |
---|
1660 | { |
---|
1661 | if (diff < 0) itvNum = -itvNum; |
---|
1662 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1663 | P0[index2D] = pred3D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1664 | } |
---|
1665 | else |
---|
1666 | { |
---|
1667 | type[index] = 0; |
---|
1668 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1669 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1670 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1671 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1672 | P0[index2D] = vce->data; |
---|
1673 | } |
---|
1674 | } |
---|
1675 | } |
---|
1676 | |
---|
1677 | float *Pt; |
---|
1678 | Pt = P1; |
---|
1679 | P1 = P0; |
---|
1680 | P0 = Pt; |
---|
1681 | } |
---|
1682 | } |
---|
1683 | |
---|
1684 | free(P0); |
---|
1685 | free(P1); |
---|
1686 | size_t exactDataNum = exactLeadNumArray->size; |
---|
1687 | |
---|
1688 | TightDataPointStorageF* tdps; |
---|
1689 | |
---|
1690 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
1691 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
1692 | exactLeadNumArray->array, |
---|
1693 | resiBitArray->array, resiBitArray->size, |
---|
1694 | resiBitsLength, |
---|
1695 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
1696 | |
---|
1697 | //free memory |
---|
1698 | free_DIA(exactLeadNumArray); |
---|
1699 | free_DIA(resiBitArray); |
---|
1700 | free(type); |
---|
1701 | free(vce); |
---|
1702 | free(lce); |
---|
1703 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
1704 | |
---|
1705 | return tdps; |
---|
1706 | } |
---|
1707 | |
---|
1708 | char SZ_compress_args_float_NoCkRngeNoGzip_4D(unsigned char** newByteData, float *oriData, size_t r1, size_t r2, size_t r3, size_t r4, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f) |
---|
1709 | { |
---|
1710 | TightDataPointStorageF* tdps = SZ_compress_float_4D_MDQ(oriData, r1, r2, r3, r4, realPrecision, valueRangeSize, medianValue_f); |
---|
1711 | |
---|
1712 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1713 | |
---|
1714 | int dataLength = r1*r2*r3*r4; |
---|
1715 | if(*outSize>dataLength*sizeof(float)) |
---|
1716 | SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
1717 | |
---|
1718 | free_TightDataPointStorageF(tdps); |
---|
1719 | |
---|
1720 | return 0; |
---|
1721 | } |
---|
1722 | |
---|
1723 | void SZ_compress_args_float_withinRange(unsigned char** newByteData, float *oriData, size_t dataLength, size_t *outSize) |
---|
1724 | { |
---|
1725 | TightDataPointStorageF* tdps = (TightDataPointStorageF*) malloc(sizeof(TightDataPointStorageF)); |
---|
1726 | tdps->rtypeArray = NULL; |
---|
1727 | tdps->typeArray = NULL; |
---|
1728 | tdps->leadNumArray = NULL; |
---|
1729 | tdps->residualMidBits = NULL; |
---|
1730 | |
---|
1731 | tdps->allSameData = 1; |
---|
1732 | tdps->dataSeriesLength = dataLength; |
---|
1733 | tdps->exactMidBytes = (unsigned char*)malloc(sizeof(unsigned char)*4); |
---|
1734 | tdps->pwrErrBoundBytes = NULL; |
---|
1735 | tdps->isLossless = 0; |
---|
1736 | float value = oriData[0]; |
---|
1737 | floatToBytes(tdps->exactMidBytes, value); |
---|
1738 | tdps->exactMidBytes_size = 4; |
---|
1739 | |
---|
1740 | size_t tmpOutSize; |
---|
1741 | //unsigned char *tmpByteData; |
---|
1742 | convertTDPStoFlatBytes_float(tdps, newByteData, &tmpOutSize); |
---|
1743 | |
---|
1744 | //*newByteData = (unsigned char*)malloc(sizeof(unsigned char)*12); //for floating-point data (1+3+4+4) |
---|
1745 | //memcpy(*newByteData, tmpByteData, 12); |
---|
1746 | *outSize = tmpOutSize; //8+SZ_SIZE_TYPE; //8==3+1+4(float_size) |
---|
1747 | free_TightDataPointStorageF(tdps); |
---|
1748 | } |
---|
1749 | |
---|
1750 | int SZ_compress_args_float_wRngeNoGzip(unsigned char** newByteData, float *oriData, |
---|
1751 | size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, size_t *outSize, |
---|
1752 | int errBoundMode, double absErr_Bound, double relBoundRatio, double pwrErrRatio) |
---|
1753 | { |
---|
1754 | int status = SZ_SCES; |
---|
1755 | size_t dataLength = computeDataLength(r5,r4,r3,r2,r1); |
---|
1756 | float valueRangeSize = 0, medianValue = 0; |
---|
1757 | |
---|
1758 | float min = computeRangeSize_float(oriData, dataLength, &valueRangeSize, &medianValue); |
---|
1759 | float max = min+valueRangeSize; |
---|
1760 | double realPrecision = getRealPrecision_float(valueRangeSize, errBoundMode, absErr_Bound, relBoundRatio, &status); |
---|
1761 | |
---|
1762 | if(valueRangeSize <= realPrecision) |
---|
1763 | { |
---|
1764 | SZ_compress_args_float_withinRange(newByteData, oriData, dataLength, outSize); |
---|
1765 | } |
---|
1766 | else |
---|
1767 | { |
---|
1768 | // SZ_compress_args_float_NoCkRngeNoGzip_2D(newByteData, oriData, r2, r1, realPrecision, outSize); |
---|
1769 | if(r5==0&&r4==0&&r3==0&&r2==0) |
---|
1770 | { |
---|
1771 | if(errBoundMode>=PW_REL) |
---|
1772 | { |
---|
1773 | //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(newByteData, oriData, realPrecision, r1, outSize, min, max); |
---|
1774 | SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(newByteData, oriData, r1, absErr_Bound, relBoundRatio, pwrErrRatio, valueRangeSize, medianValue, outSize); |
---|
1775 | } |
---|
1776 | else |
---|
1777 | SZ_compress_args_float_NoCkRngeNoGzip_1D(newByteData, oriData, r1, realPrecision, outSize, valueRangeSize, medianValue); |
---|
1778 | } |
---|
1779 | else if(r5==0&&r4==0&&r3==0) |
---|
1780 | { |
---|
1781 | if(errBoundMode>=PW_REL) |
---|
1782 | SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr(newByteData, oriData, realPrecision, r2, r1, outSize, min, max); |
---|
1783 | else |
---|
1784 | SZ_compress_args_float_NoCkRngeNoGzip_2D(newByteData, oriData, r2, r1, realPrecision, outSize, valueRangeSize, medianValue); |
---|
1785 | } |
---|
1786 | else if(r5==0&&r4==0) |
---|
1787 | { |
---|
1788 | if(errBoundMode>=PW_REL) |
---|
1789 | SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(newByteData, oriData, realPrecision, r3, r2, r1, outSize, min, max); |
---|
1790 | else |
---|
1791 | SZ_compress_args_float_NoCkRngeNoGzip_3D(newByteData, oriData, r3, r2, r1, realPrecision, outSize, valueRangeSize, medianValue); |
---|
1792 | } |
---|
1793 | else if(r5==0) |
---|
1794 | { |
---|
1795 | if(errBoundMode>=PW_REL) |
---|
1796 | SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(newByteData, oriData, realPrecision, r4*r3, r2, r1, outSize, min, max); |
---|
1797 | else |
---|
1798 | SZ_compress_args_float_NoCkRngeNoGzip_3D(newByteData, oriData, r4*r3, r2, r1, realPrecision, outSize, valueRangeSize, medianValue); |
---|
1799 | } |
---|
1800 | } |
---|
1801 | return status; |
---|
1802 | } |
---|
1803 | |
---|
1804 | int SZ_compress_args_float(unsigned char** newByteData, float *oriData, |
---|
1805 | size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, size_t *outSize, |
---|
1806 | int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRatio) |
---|
1807 | { |
---|
1808 | confparams_cpr->errorBoundMode = errBoundMode; |
---|
1809 | if(errBoundMode==PW_REL) |
---|
1810 | { |
---|
1811 | confparams_cpr->pw_relBoundRatio = pwRelBoundRatio; |
---|
1812 | //confparams_cpr->pwr_type = SZ_PWR_MIN_TYPE; |
---|
1813 | if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE && r3 != 0 ) |
---|
1814 | { |
---|
1815 | printf("Error: Current version doesn't support 3D data compression with point-wise relative error bound being based on pwrType=AVG\n"); |
---|
1816 | exit(0); |
---|
1817 | return SZ_NSCS; |
---|
1818 | } |
---|
1819 | } |
---|
1820 | int status = SZ_SCES; |
---|
1821 | size_t dataLength = computeDataLength(r5,r4,r3,r2,r1); |
---|
1822 | |
---|
1823 | if(dataLength <= MIN_NUM_OF_ELEMENTS) |
---|
1824 | { |
---|
1825 | *newByteData = SZ_skip_compress_float(oriData, dataLength, outSize); |
---|
1826 | return status; |
---|
1827 | } |
---|
1828 | |
---|
1829 | float valueRangeSize = 0, medianValue = 0; |
---|
1830 | |
---|
1831 | float min = computeRangeSize_float(oriData, dataLength, &valueRangeSize, &medianValue); |
---|
1832 | float max = min+valueRangeSize; |
---|
1833 | double realPrecision = 0; |
---|
1834 | |
---|
1835 | if(confparams_cpr->errorBoundMode==PSNR) |
---|
1836 | { |
---|
1837 | confparams_cpr->errorBoundMode = ABS; |
---|
1838 | realPrecision = confparams_cpr->absErrBound = computeABSErrBoundFromPSNR(confparams_cpr->psnr, (double)confparams_cpr->predThreshold, (double)valueRangeSize); |
---|
1839 | //printf("realPrecision=%lf\n", realPrecision); |
---|
1840 | } |
---|
1841 | else |
---|
1842 | realPrecision = getRealPrecision_float(valueRangeSize, errBoundMode, absErr_Bound, relBoundRatio, &status); |
---|
1843 | |
---|
1844 | if(valueRangeSize <= realPrecision) |
---|
1845 | { |
---|
1846 | SZ_compress_args_float_withinRange(newByteData, oriData, dataLength, outSize); |
---|
1847 | } |
---|
1848 | else |
---|
1849 | { |
---|
1850 | size_t tmpOutSize = 0; |
---|
1851 | unsigned char* tmpByteData; |
---|
1852 | |
---|
1853 | if (r2==0) |
---|
1854 | { |
---|
1855 | if(confparams_cpr->errorBoundMode>=PW_REL) |
---|
1856 | { |
---|
1857 | //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(&tmpByteData, oriData, realPrecision, r1, &tmpOutSize, min, max); |
---|
1858 | SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio, |
---|
1859 | valueRangeSize, medianValue, &tmpOutSize); |
---|
1860 | } |
---|
1861 | else |
---|
1862 | #ifdef HAVE_TIMECMPR |
---|
1863 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1864 | multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_1D(&tmpByteData, oriData, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1865 | else |
---|
1866 | #endif |
---|
1867 | SZ_compress_args_float_NoCkRngeNoGzip_1D(&tmpByteData, oriData, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1868 | } |
---|
1869 | else |
---|
1870 | if (r3==0) |
---|
1871 | { |
---|
1872 | if(confparams_cpr->errorBoundMode>=PW_REL) |
---|
1873 | SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr(&tmpByteData, oriData, realPrecision, r2, r1, &tmpOutSize, min, max); |
---|
1874 | else |
---|
1875 | #ifdef HAVE_TIMECMPR |
---|
1876 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1877 | multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1878 | else |
---|
1879 | #endif |
---|
1880 | SZ_compress_args_float_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1881 | } |
---|
1882 | else |
---|
1883 | if (r4==0) |
---|
1884 | { |
---|
1885 | if(confparams_cpr->errorBoundMode>=PW_REL) |
---|
1886 | SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r3, r2, r1, &tmpOutSize, min, max); |
---|
1887 | else |
---|
1888 | #ifdef HAVE_TIMECMPR |
---|
1889 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1890 | multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1891 | else |
---|
1892 | #endif |
---|
1893 | SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1894 | } |
---|
1895 | else |
---|
1896 | if (r5==0) |
---|
1897 | { |
---|
1898 | if(confparams_cpr->errorBoundMode>=PW_REL) |
---|
1899 | SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r4*r3, r2, r1, &tmpOutSize, min, max); |
---|
1900 | //ToDO |
---|
1901 | //SZ_compress_args_float_NoCkRngeNoGzip_4D_pwr(&tmpByteData, oriData, r4, r3, r2, r1, &tmpOutSize, min, max); |
---|
1902 | else |
---|
1903 | #ifdef HAVE_TIMECMPR |
---|
1904 | if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION) |
---|
1905 | multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1906 | else |
---|
1907 | #endif |
---|
1908 | SZ_compress_args_float_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue); |
---|
1909 | } |
---|
1910 | else |
---|
1911 | { |
---|
1912 | printf("Error: doesn't support 5 dimensions for now.\n"); |
---|
1913 | status = SZ_DERR; //dimension error |
---|
1914 | } |
---|
1915 | //Call Gzip to do the further compression. |
---|
1916 | if(confparams_cpr->szMode==SZ_BEST_SPEED) |
---|
1917 | { |
---|
1918 | *outSize = tmpOutSize; |
---|
1919 | *newByteData = tmpByteData; |
---|
1920 | } |
---|
1921 | else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION || confparams_cpr->szMode==SZ_TEMPORAL_COMPRESSION) |
---|
1922 | { |
---|
1923 | *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode); |
---|
1924 | free(tmpByteData); |
---|
1925 | } |
---|
1926 | else |
---|
1927 | { |
---|
1928 | printf("Error: Wrong setting of confparams_cpr->szMode in the float compression.\n"); |
---|
1929 | status = SZ_MERR; //mode error |
---|
1930 | } |
---|
1931 | } |
---|
1932 | |
---|
1933 | return status; |
---|
1934 | } |
---|
1935 | |
---|
1936 | |
---|
1937 | void computeReqLength_float(double realPrecision, short radExpo, int* reqLength, float* medianValue) |
---|
1938 | { |
---|
1939 | short reqExpo = getPrecisionReqLength_double(realPrecision); |
---|
1940 | *reqLength = 9+radExpo - reqExpo; //radExpo-reqExpo == reqMantiLength |
---|
1941 | if(*reqLength<9) |
---|
1942 | *reqLength = 9; |
---|
1943 | if(*reqLength>32) |
---|
1944 | { |
---|
1945 | *reqLength = 32; |
---|
1946 | *medianValue = 0; |
---|
1947 | } |
---|
1948 | } |
---|
1949 | |
---|
1950 | //TODO |
---|
1951 | int SZ_compress_args_float_subblock(unsigned char* compressedBytes, float *oriData, |
---|
1952 | size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, |
---|
1953 | size_t s5, size_t s4, size_t s3, size_t s2, size_t s1, |
---|
1954 | size_t e5, size_t e4, size_t e3, size_t e2, size_t e1, |
---|
1955 | size_t *outSize, int errBoundMode, double absErr_Bound, double relBoundRatio) |
---|
1956 | { |
---|
1957 | int status = SZ_SCES; |
---|
1958 | float valueRangeSize = 0, medianValue = 0; |
---|
1959 | computeRangeSize_float_subblock(oriData, &valueRangeSize, &medianValue, r5, r4, r3, r2, r1, s5, s4, s3, s2, s1, e5, e4, e3, e2, e1); |
---|
1960 | |
---|
1961 | double realPrecision = getRealPrecision_float(valueRangeSize, errBoundMode, absErr_Bound, relBoundRatio, &status); |
---|
1962 | |
---|
1963 | if(valueRangeSize <= realPrecision) |
---|
1964 | { |
---|
1965 | //TODO |
---|
1966 | //SZ_compress_args_float_withinRange_subblock(); |
---|
1967 | } |
---|
1968 | else |
---|
1969 | { |
---|
1970 | if (r2==0) |
---|
1971 | { |
---|
1972 | if(errBoundMode>=PW_REL) |
---|
1973 | { |
---|
1974 | //TODO |
---|
1975 | //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_subblock(); |
---|
1976 | printf ("Current subblock version does not support point-wise relative error bound.\n"); |
---|
1977 | } |
---|
1978 | else |
---|
1979 | SZ_compress_args_float_NoCkRnge_1D_subblock(compressedBytes, oriData, realPrecision, outSize, valueRangeSize, medianValue, r1, s1, e1); |
---|
1980 | } |
---|
1981 | else |
---|
1982 | if (r3==0) |
---|
1983 | { |
---|
1984 | //TODO |
---|
1985 | if(errBoundMode>=PW_REL) |
---|
1986 | { |
---|
1987 | //TODO |
---|
1988 | //SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr_subblock(); |
---|
1989 | printf ("Current subblock version does not support point-wise relative error bound.\n"); |
---|
1990 | } |
---|
1991 | else |
---|
1992 | SZ_compress_args_float_NoCkRnge_2D_subblock(compressedBytes, oriData, realPrecision, outSize, valueRangeSize, medianValue, r2, r1, s2, s1, e2, e1); |
---|
1993 | } |
---|
1994 | else |
---|
1995 | if (r4==0) |
---|
1996 | { |
---|
1997 | if(errBoundMode>=PW_REL) |
---|
1998 | { |
---|
1999 | //TODO |
---|
2000 | //SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_subblock(); |
---|
2001 | printf ("Current subblock version does not support point-wise relative error bound.\n"); |
---|
2002 | } |
---|
2003 | else |
---|
2004 | SZ_compress_args_float_NoCkRnge_3D_subblock(compressedBytes, oriData, realPrecision, outSize, valueRangeSize, medianValue, r3, r2, r1, s3, s2, s1, e3, e2, e1); |
---|
2005 | } |
---|
2006 | else |
---|
2007 | if (r5==0) |
---|
2008 | { |
---|
2009 | if(errBoundMode>=PW_REL) |
---|
2010 | { |
---|
2011 | //TODO |
---|
2012 | //SZ_compress_args_float_NoCkRngeNoGzip_4D_pwr_subblock(); |
---|
2013 | printf ("Current subblock version does not support point-wise relative error bound.\n"); |
---|
2014 | } |
---|
2015 | else |
---|
2016 | SZ_compress_args_float_NoCkRnge_4D_subblock(compressedBytes, oriData, realPrecision, outSize, valueRangeSize, medianValue, r4, r3, r2, r1, s4, s3, s2, s1, e4, e3, e2, e1); |
---|
2017 | } |
---|
2018 | else |
---|
2019 | { |
---|
2020 | printf("Error: doesn't support 5 dimensions for now.\n"); |
---|
2021 | status = SZ_DERR; //dimension error |
---|
2022 | } |
---|
2023 | } |
---|
2024 | return status; |
---|
2025 | } |
---|
2026 | |
---|
2027 | void SZ_compress_args_float_NoCkRnge_1D_subblock(unsigned char* compressedBytes, float *oriData, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f, |
---|
2028 | size_t r1, size_t s1, size_t e1) |
---|
2029 | { |
---|
2030 | TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ_subblock(oriData, realPrecision, valueRangeSize, medianValue_f, r1, s1, e1); |
---|
2031 | |
---|
2032 | if (confparams_cpr->szMode==SZ_BEST_SPEED) |
---|
2033 | convertTDPStoFlatBytes_float_args(tdps, compressedBytes, outSize); |
---|
2034 | else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION) |
---|
2035 | { |
---|
2036 | unsigned char *tmpCompBytes; |
---|
2037 | size_t tmpOutSize; |
---|
2038 | convertTDPStoFlatBytes_float(tdps, &tmpCompBytes, &tmpOutSize); |
---|
2039 | *outSize = zlib_compress3(tmpCompBytes, tmpOutSize, compressedBytes, confparams_cpr->gzipMode); |
---|
2040 | free(tmpCompBytes); |
---|
2041 | } |
---|
2042 | else |
---|
2043 | { |
---|
2044 | printf ("Error: Wrong setting of confparams_cpr->szMode in the double compression.\n"); |
---|
2045 | } |
---|
2046 | |
---|
2047 | //TODO |
---|
2048 | // if(*outSize>dataLength*sizeof(float)) |
---|
2049 | // SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
2050 | |
---|
2051 | free_TightDataPointStorageF(tdps); |
---|
2052 | } |
---|
2053 | |
---|
2054 | void SZ_compress_args_float_NoCkRnge_2D_subblock(unsigned char* compressedBytes, float *oriData, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f, |
---|
2055 | size_t r2, size_t r1, size_t s2, size_t s1, size_t e2, size_t e1) |
---|
2056 | { |
---|
2057 | TightDataPointStorageF* tdps = SZ_compress_float_2D_MDQ_subblock(oriData, realPrecision, valueRangeSize, medianValue_f, r2, r1, s2, s1, e2, e1); |
---|
2058 | |
---|
2059 | if (confparams_cpr->szMode==SZ_BEST_SPEED) |
---|
2060 | convertTDPStoFlatBytes_float_args(tdps, compressedBytes, outSize); |
---|
2061 | else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION) |
---|
2062 | { |
---|
2063 | unsigned char *tmpCompBytes; |
---|
2064 | size_t tmpOutSize; |
---|
2065 | convertTDPStoFlatBytes_float(tdps, &tmpCompBytes, &tmpOutSize); |
---|
2066 | *outSize = zlib_compress3(tmpCompBytes, tmpOutSize, compressedBytes, confparams_cpr->gzipMode); |
---|
2067 | free(tmpCompBytes); |
---|
2068 | } |
---|
2069 | else |
---|
2070 | { |
---|
2071 | printf ("Error: Wrong setting of confparams_cpr->szMode in the double compression.\n"); |
---|
2072 | } |
---|
2073 | |
---|
2074 | //TODO |
---|
2075 | // if(*outSize>dataLength*sizeof(float)) |
---|
2076 | // SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
2077 | |
---|
2078 | free_TightDataPointStorageF(tdps); |
---|
2079 | } |
---|
2080 | |
---|
2081 | void SZ_compress_args_float_NoCkRnge_3D_subblock(unsigned char* compressedBytes, float *oriData, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f, |
---|
2082 | size_t r3, size_t r2, size_t r1, size_t s3, size_t s2, size_t s1, size_t e3, size_t e2, size_t e1) |
---|
2083 | { |
---|
2084 | TightDataPointStorageF* tdps = SZ_compress_float_3D_MDQ_subblock(oriData, realPrecision, valueRangeSize, medianValue_f, r3, r2, r1, s3, s2, s1, e3, e2, e1); |
---|
2085 | |
---|
2086 | if (confparams_cpr->szMode==SZ_BEST_SPEED) |
---|
2087 | convertTDPStoFlatBytes_float_args(tdps, compressedBytes, outSize); |
---|
2088 | else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION) |
---|
2089 | { |
---|
2090 | unsigned char *tmpCompBytes; |
---|
2091 | size_t tmpOutSize; |
---|
2092 | convertTDPStoFlatBytes_float(tdps, &tmpCompBytes, &tmpOutSize); |
---|
2093 | *outSize = zlib_compress3(tmpCompBytes, tmpOutSize, compressedBytes, confparams_cpr->gzipMode); |
---|
2094 | free(tmpCompBytes); |
---|
2095 | } |
---|
2096 | else |
---|
2097 | { |
---|
2098 | printf ("Error: Wrong setting of confparams_cpr->szMode in the double compression.\n"); |
---|
2099 | } |
---|
2100 | |
---|
2101 | //TODO |
---|
2102 | // if(*outSize>dataLength*sizeof(float)) |
---|
2103 | // SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
2104 | |
---|
2105 | free_TightDataPointStorageF(tdps); |
---|
2106 | } |
---|
2107 | |
---|
2108 | void SZ_compress_args_float_NoCkRnge_4D_subblock(unsigned char* compressedBytes, float *oriData, double realPrecision, size_t *outSize, float valueRangeSize, float medianValue_f, |
---|
2109 | size_t r4, size_t r3, size_t r2, size_t r1, size_t s4, size_t s3, size_t s2, size_t s1, size_t e4, size_t e3, size_t e2, size_t e1) |
---|
2110 | { |
---|
2111 | TightDataPointStorageF* tdps = SZ_compress_float_4D_MDQ_subblock(oriData, realPrecision, valueRangeSize, medianValue_f, r4, r3, r2, r1, s4, s3, s2, s1, e4, e3, e2, e1); |
---|
2112 | |
---|
2113 | if (confparams_cpr->szMode==SZ_BEST_SPEED) |
---|
2114 | convertTDPStoFlatBytes_float_args(tdps, compressedBytes, outSize); |
---|
2115 | else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION) |
---|
2116 | { |
---|
2117 | unsigned char *tmpCompBytes; |
---|
2118 | size_t tmpOutSize; |
---|
2119 | convertTDPStoFlatBytes_float(tdps, &tmpCompBytes, &tmpOutSize); |
---|
2120 | *outSize = zlib_compress3(tmpCompBytes, tmpOutSize, compressedBytes, confparams_cpr->gzipMode); |
---|
2121 | free(tmpCompBytes); |
---|
2122 | } |
---|
2123 | else |
---|
2124 | { |
---|
2125 | printf ("Error: Wrong setting of confparams_cpr->szMode in the double compression.\n"); |
---|
2126 | } |
---|
2127 | |
---|
2128 | //TODO |
---|
2129 | // if(*outSize>dataLength*sizeof(float)) |
---|
2130 | // SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize); |
---|
2131 | |
---|
2132 | free_TightDataPointStorageF(tdps); |
---|
2133 | |
---|
2134 | } |
---|
2135 | |
---|
2136 | unsigned int optimize_intervals_float_1D_subblock(float *oriData, double realPrecision, size_t r1, size_t s1, size_t e1) |
---|
2137 | { |
---|
2138 | size_t dataLength = e1 - s1 + 1; |
---|
2139 | oriData = oriData + s1; |
---|
2140 | |
---|
2141 | size_t i = 0; |
---|
2142 | unsigned long radiusIndex; |
---|
2143 | float pred_value = 0, pred_err; |
---|
2144 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2145 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2146 | size_t totalSampleSize = dataLength/confparams_cpr->sampleDistance; |
---|
2147 | for(i=2;i<dataLength;i++) |
---|
2148 | { |
---|
2149 | if(i%confparams_cpr->sampleDistance==0) |
---|
2150 | { |
---|
2151 | pred_value = 2*oriData[i-1] - oriData[i-2]; |
---|
2152 | //pred_value = oriData[i-1]; |
---|
2153 | pred_err = fabs(pred_value - oriData[i]); |
---|
2154 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
2155 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
2156 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
2157 | intervals[radiusIndex]++; |
---|
2158 | } |
---|
2159 | } |
---|
2160 | //compute the appropriate number |
---|
2161 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
2162 | size_t sum = 0; |
---|
2163 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
2164 | { |
---|
2165 | sum += intervals[i]; |
---|
2166 | if(sum>targetCount) |
---|
2167 | break; |
---|
2168 | } |
---|
2169 | if(i>=confparams_cpr->maxRangeRadius) |
---|
2170 | i = confparams_cpr->maxRangeRadius-1; |
---|
2171 | |
---|
2172 | unsigned int accIntervals = 2*(i+1); |
---|
2173 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
2174 | |
---|
2175 | if(powerOf2<32) |
---|
2176 | powerOf2 = 32; |
---|
2177 | |
---|
2178 | free(intervals); |
---|
2179 | //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2); |
---|
2180 | return powerOf2; |
---|
2181 | } |
---|
2182 | |
---|
2183 | unsigned int optimize_intervals_float_2D_subblock(float *oriData, double realPrecision, size_t r1, size_t r2, size_t s1, size_t s2, size_t e1, size_t e2) |
---|
2184 | { |
---|
2185 | size_t R1 = e1 - s1 + 1; |
---|
2186 | size_t R2 = e2 - s2 + 1; |
---|
2187 | |
---|
2188 | size_t i,j, index; |
---|
2189 | unsigned long radiusIndex; |
---|
2190 | float pred_value = 0, pred_err; |
---|
2191 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2192 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2193 | size_t totalSampleSize = R1*R2/confparams_cpr->sampleDistance; |
---|
2194 | for(i=s1+1;i<=e1;i++) |
---|
2195 | { |
---|
2196 | for(j=s2+1;j<=e2;j++) |
---|
2197 | { |
---|
2198 | if((i+j)%confparams_cpr->sampleDistance==0) |
---|
2199 | { |
---|
2200 | index = i*r2+j; |
---|
2201 | pred_value = oriData[index-1] + oriData[index-r2] - oriData[index-r2-1]; |
---|
2202 | pred_err = fabs(pred_value - oriData[index]); |
---|
2203 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
2204 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
2205 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
2206 | intervals[radiusIndex]++; |
---|
2207 | } |
---|
2208 | } |
---|
2209 | } |
---|
2210 | //compute the appropriate number |
---|
2211 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
2212 | size_t sum = 0; |
---|
2213 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
2214 | { |
---|
2215 | sum += intervals[i]; |
---|
2216 | if(sum>targetCount) |
---|
2217 | break; |
---|
2218 | } |
---|
2219 | if(i>=confparams_cpr->maxRangeRadius) |
---|
2220 | i = confparams_cpr->maxRangeRadius-1; |
---|
2221 | unsigned int accIntervals = 2*(i+1); |
---|
2222 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
2223 | |
---|
2224 | if(powerOf2<32) |
---|
2225 | powerOf2 = 32; |
---|
2226 | |
---|
2227 | free(intervals); |
---|
2228 | //printf("confparams_cpr->maxRangeRadius = %d, accIntervals=%d, powerOf2=%d\n", confparams_cpr->maxRangeRadius, accIntervals, powerOf2); |
---|
2229 | return powerOf2; |
---|
2230 | } |
---|
2231 | |
---|
2232 | unsigned int optimize_intervals_float_3D_subblock(float *oriData, double realPrecision, size_t r1, size_t r2, size_t r3, size_t s1, size_t s2, size_t s3, size_t e1, size_t e2, size_t e3) |
---|
2233 | { |
---|
2234 | size_t R1 = e1 - s1 + 1; |
---|
2235 | size_t R2 = e2 - s2 + 1; |
---|
2236 | size_t R3 = e3 - s3 + 1; |
---|
2237 | |
---|
2238 | size_t r23 = r2*r3; |
---|
2239 | |
---|
2240 | size_t i,j,k, index; |
---|
2241 | unsigned long radiusIndex; |
---|
2242 | float pred_value = 0, pred_err; |
---|
2243 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2244 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2245 | size_t totalSampleSize = R1*R2*R3/confparams_cpr->sampleDistance; |
---|
2246 | for(i=s1+1;i<=e1;i++) |
---|
2247 | { |
---|
2248 | for(j=s2+1;j<=e2;j++) |
---|
2249 | { |
---|
2250 | for(k=s3+1;k<=e3;k++) |
---|
2251 | { |
---|
2252 | if((i+j+k)%confparams_cpr->sampleDistance==0) |
---|
2253 | { |
---|
2254 | index = i*r23+j*r3+k; |
---|
2255 | pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r23] |
---|
2256 | - oriData[index-1-r23] - oriData[index-r3-1] - oriData[index-r3-r23] + oriData[index-r3-r23-1]; |
---|
2257 | pred_err = fabs(pred_value - oriData[index]); |
---|
2258 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
2259 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
2260 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
2261 | intervals[radiusIndex]++; |
---|
2262 | } |
---|
2263 | } |
---|
2264 | } |
---|
2265 | } |
---|
2266 | //compute the appropriate number |
---|
2267 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
2268 | size_t sum = 0; |
---|
2269 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
2270 | { |
---|
2271 | sum += intervals[i]; |
---|
2272 | if(sum>targetCount) |
---|
2273 | break; |
---|
2274 | } |
---|
2275 | if(i>=confparams_cpr->maxRangeRadius) |
---|
2276 | i = confparams_cpr->maxRangeRadius-1; |
---|
2277 | unsigned int accIntervals = 2*(i+1); |
---|
2278 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
2279 | |
---|
2280 | if(powerOf2<32) |
---|
2281 | powerOf2 = 32; |
---|
2282 | |
---|
2283 | free(intervals); |
---|
2284 | return powerOf2; |
---|
2285 | } |
---|
2286 | |
---|
2287 | unsigned int optimize_intervals_float_4D_subblock(float *oriData, double realPrecision, |
---|
2288 | size_t r1, size_t r2, size_t r3, size_t r4, size_t s1, size_t s2, size_t s3, size_t s4, size_t e1, size_t e2, size_t e3, size_t e4) |
---|
2289 | { |
---|
2290 | size_t R1 = e1 - s1 + 1; |
---|
2291 | size_t R2 = e2 - s2 + 1; |
---|
2292 | size_t R3 = e3 - s3 + 1; |
---|
2293 | size_t R4 = e4 - s4 + 1; |
---|
2294 | |
---|
2295 | size_t r34 = r3*r4; |
---|
2296 | size_t r234 = r2*r3*r4; |
---|
2297 | |
---|
2298 | size_t i,j,k,l, index; |
---|
2299 | unsigned long radiusIndex; |
---|
2300 | float pred_value = 0, pred_err; |
---|
2301 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2302 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
2303 | size_t totalSampleSize = R1*R2*R3*R4/confparams_cpr->sampleDistance; |
---|
2304 | for(i=s1+1;i<=e1;i++) |
---|
2305 | { |
---|
2306 | for(j=s2+1;j<=e2;j++) |
---|
2307 | { |
---|
2308 | for(k=s3+1;k<=e3;k++) |
---|
2309 | { |
---|
2310 | for (l=s4+1;l<=e4;l++) |
---|
2311 | { |
---|
2312 | if((i+j+k+l)%confparams_cpr->sampleDistance==0) |
---|
2313 | { |
---|
2314 | index = i*r234+j*r34+k*r4+l; |
---|
2315 | pred_value = oriData[index-1] + oriData[index-r4] + oriData[index-r34] |
---|
2316 | - oriData[index-1-r34] - oriData[index-r4-1] - oriData[index-r4-r34] + oriData[index-r4-r34-1]; |
---|
2317 | pred_err = fabs(pred_value - oriData[index]); |
---|
2318 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
2319 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
2320 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
2321 | intervals[radiusIndex]++; |
---|
2322 | } |
---|
2323 | } |
---|
2324 | } |
---|
2325 | } |
---|
2326 | } |
---|
2327 | //compute the appropriate number |
---|
2328 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
2329 | size_t sum = 0; |
---|
2330 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
2331 | { |
---|
2332 | sum += intervals[i]; |
---|
2333 | if(sum>targetCount) |
---|
2334 | break; |
---|
2335 | } |
---|
2336 | if(i>=confparams_cpr->maxRangeRadius) |
---|
2337 | i = confparams_cpr->maxRangeRadius-1; |
---|
2338 | |
---|
2339 | unsigned int accIntervals = 2*(i+1); |
---|
2340 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
2341 | |
---|
2342 | if(powerOf2<32) |
---|
2343 | powerOf2 = 32; |
---|
2344 | |
---|
2345 | free(intervals); |
---|
2346 | return powerOf2; |
---|
2347 | } |
---|
2348 | |
---|
2349 | TightDataPointStorageF* SZ_compress_float_1D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f, |
---|
2350 | size_t r1, size_t s1, size_t e1) |
---|
2351 | { |
---|
2352 | size_t dataLength = e1 - s1 + 1; |
---|
2353 | unsigned int quantization_intervals; |
---|
2354 | if(exe_params->optQuantMode==1) |
---|
2355 | quantization_intervals = optimize_intervals_float_1D_subblock(oriData, realPrecision, r1, s1, e1); |
---|
2356 | else |
---|
2357 | quantization_intervals = exe_params->intvCapacity; |
---|
2358 | updateQuantizationInfo(quantization_intervals); |
---|
2359 | |
---|
2360 | size_t i; |
---|
2361 | int reqLength; |
---|
2362 | float medianValue = medianValue_f; |
---|
2363 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
2364 | |
---|
2365 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
2366 | |
---|
2367 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
2368 | |
---|
2369 | float* spaceFillingValue = oriData + s1; |
---|
2370 | |
---|
2371 | DynamicIntArray *exactLeadNumArray; |
---|
2372 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
2373 | |
---|
2374 | DynamicByteArray *exactMidByteArray; |
---|
2375 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
2376 | |
---|
2377 | DynamicIntArray *resiBitArray; |
---|
2378 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
2379 | |
---|
2380 | type[0] = 0; |
---|
2381 | |
---|
2382 | unsigned char preDataBytes[4]; |
---|
2383 | intToBytes_bigEndian(preDataBytes, 0); |
---|
2384 | |
---|
2385 | int reqBytesLength = reqLength/8; |
---|
2386 | int resiBitsLength = reqLength%8; |
---|
2387 | float last3CmprsData[3] = {0}; |
---|
2388 | |
---|
2389 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
2390 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
2391 | |
---|
2392 | //add the first data |
---|
2393 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2394 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2395 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2396 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2397 | listAdd_float(last3CmprsData, vce->data); |
---|
2398 | |
---|
2399 | //add the second data |
---|
2400 | type[1] = 0; |
---|
2401 | compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2402 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2403 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2404 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2405 | listAdd_float(last3CmprsData, vce->data); |
---|
2406 | |
---|
2407 | int state; |
---|
2408 | double checkRadius; |
---|
2409 | float curData; |
---|
2410 | float pred; |
---|
2411 | float predAbsErr; |
---|
2412 | checkRadius = (exe_params->intvCapacity-1)*realPrecision; |
---|
2413 | double interval = 2*realPrecision; |
---|
2414 | |
---|
2415 | for(i=2;i<dataLength;i++) |
---|
2416 | { |
---|
2417 | curData = spaceFillingValue[i]; |
---|
2418 | pred = 2*last3CmprsData[0] - last3CmprsData[1]; |
---|
2419 | predAbsErr = fabs(curData - pred); |
---|
2420 | if(predAbsErr<=checkRadius) |
---|
2421 | { |
---|
2422 | state = (predAbsErr/realPrecision+1)/2; |
---|
2423 | if(curData>=pred) |
---|
2424 | { |
---|
2425 | type[i] = exe_params->intvRadius+state; |
---|
2426 | pred = pred + state*interval; |
---|
2427 | } |
---|
2428 | else |
---|
2429 | { |
---|
2430 | type[i] = exe_params->intvRadius-state; |
---|
2431 | pred = pred - state*interval; |
---|
2432 | } |
---|
2433 | |
---|
2434 | listAdd_float(last3CmprsData, pred); |
---|
2435 | continue; |
---|
2436 | } |
---|
2437 | |
---|
2438 | //unpredictable data processing |
---|
2439 | type[i] = 0; |
---|
2440 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2441 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2442 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2443 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2444 | |
---|
2445 | listAdd_float(last3CmprsData, vce->data); |
---|
2446 | } |
---|
2447 | |
---|
2448 | size_t exactDataNum = exactLeadNumArray->size; |
---|
2449 | |
---|
2450 | TightDataPointStorageF* tdps; |
---|
2451 | |
---|
2452 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
2453 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
2454 | exactLeadNumArray->array, |
---|
2455 | resiBitArray->array, resiBitArray->size, |
---|
2456 | resiBitsLength, |
---|
2457 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
2458 | |
---|
2459 | //free memory |
---|
2460 | free_DIA(exactLeadNumArray); |
---|
2461 | free_DIA(resiBitArray); |
---|
2462 | free(type); |
---|
2463 | free(vce); |
---|
2464 | free(lce); |
---|
2465 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
2466 | |
---|
2467 | return tdps; |
---|
2468 | } |
---|
2469 | |
---|
2470 | TightDataPointStorageF* SZ_compress_float_2D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f, |
---|
2471 | size_t r1, size_t r2, size_t s1, size_t s2, size_t e1, size_t e2) |
---|
2472 | { |
---|
2473 | unsigned int quantization_intervals; |
---|
2474 | if(exe_params->optQuantMode==1) |
---|
2475 | { |
---|
2476 | quantization_intervals = optimize_intervals_float_2D_subblock(oriData, realPrecision, r1, r2, s1, s2, e1, e2); |
---|
2477 | updateQuantizationInfo(quantization_intervals); |
---|
2478 | } |
---|
2479 | else |
---|
2480 | quantization_intervals = exe_params->intvCapacity; |
---|
2481 | |
---|
2482 | size_t i,j; |
---|
2483 | int reqLength; |
---|
2484 | float pred1D, pred2D; |
---|
2485 | float diff = 0.0; |
---|
2486 | double itvNum = 0; |
---|
2487 | float *P0, *P1; |
---|
2488 | |
---|
2489 | size_t R1 = e1 - s1 + 1; |
---|
2490 | size_t R2 = e2 - s2 + 1; |
---|
2491 | size_t dataLength = R1*R2; |
---|
2492 | |
---|
2493 | P0 = (float*)malloc(R2*sizeof(float)); |
---|
2494 | memset(P0, 0, R2*sizeof(float)); |
---|
2495 | P1 = (float*)malloc(R2*sizeof(float)); |
---|
2496 | memset(P1, 0, R2*sizeof(float)); |
---|
2497 | |
---|
2498 | float medianValue = medianValue_f; |
---|
2499 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
2500 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
2501 | |
---|
2502 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
2503 | |
---|
2504 | float* spaceFillingValue = oriData; // |
---|
2505 | |
---|
2506 | DynamicIntArray *exactLeadNumArray; |
---|
2507 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
2508 | |
---|
2509 | DynamicByteArray *exactMidByteArray; |
---|
2510 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
2511 | |
---|
2512 | DynamicIntArray *resiBitArray; |
---|
2513 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
2514 | |
---|
2515 | unsigned char preDataBytes[4]; |
---|
2516 | intToBytes_bigEndian(preDataBytes, 0); |
---|
2517 | |
---|
2518 | int reqBytesLength = reqLength/8; |
---|
2519 | int resiBitsLength = reqLength%8; |
---|
2520 | |
---|
2521 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
2522 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
2523 | |
---|
2524 | /* Process Row-s1 data s2*/ |
---|
2525 | size_t gIndex; |
---|
2526 | size_t lIndex; |
---|
2527 | |
---|
2528 | gIndex = s1*r2+s2; |
---|
2529 | lIndex = 0; |
---|
2530 | |
---|
2531 | type[lIndex] = 0; |
---|
2532 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2533 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2534 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2535 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2536 | P1[0] = vce->data; |
---|
2537 | |
---|
2538 | /* Process Row-s1 data s2+1*/ |
---|
2539 | gIndex = s1*r2+(s2+1); |
---|
2540 | lIndex = 1; |
---|
2541 | |
---|
2542 | pred1D = P1[0]; |
---|
2543 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2544 | |
---|
2545 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2546 | |
---|
2547 | if (itvNum < exe_params->intvCapacity) |
---|
2548 | { |
---|
2549 | if (diff < 0) itvNum = -itvNum; |
---|
2550 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2551 | P1[1] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2552 | } |
---|
2553 | else |
---|
2554 | { |
---|
2555 | type[lIndex] = 0; |
---|
2556 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2557 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2558 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2559 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2560 | P1[1] = vce->data; |
---|
2561 | } |
---|
2562 | |
---|
2563 | /* Process Row-s1 data s2+2 --> data e2 */ |
---|
2564 | for (j = 2; j < R2; j++) |
---|
2565 | { |
---|
2566 | gIndex = s1*r2+(s2+j); |
---|
2567 | lIndex = j; |
---|
2568 | |
---|
2569 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
2570 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2571 | |
---|
2572 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2573 | |
---|
2574 | if (itvNum < exe_params->intvCapacity) |
---|
2575 | { |
---|
2576 | if (diff < 0) itvNum = -itvNum; |
---|
2577 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2578 | P1[j] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2579 | } |
---|
2580 | else |
---|
2581 | { |
---|
2582 | type[lIndex] = 0; |
---|
2583 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2584 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2585 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2586 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2587 | P1[j] = vce->data; |
---|
2588 | } |
---|
2589 | } |
---|
2590 | |
---|
2591 | /* Process Row-s1+1 --> Row-e1 */ |
---|
2592 | for (i = 1; i < R1; i++) |
---|
2593 | { |
---|
2594 | /* Process row-s1+i data s2 */ |
---|
2595 | gIndex = (s1+i)*r2+s2; |
---|
2596 | lIndex = i*R2; |
---|
2597 | |
---|
2598 | pred1D = P1[0]; |
---|
2599 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2600 | |
---|
2601 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2602 | |
---|
2603 | if (itvNum < exe_params->intvCapacity) |
---|
2604 | { |
---|
2605 | if (diff < 0) itvNum = -itvNum; |
---|
2606 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2607 | P0[0] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2608 | } |
---|
2609 | else |
---|
2610 | { |
---|
2611 | type[lIndex] = 0; |
---|
2612 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2613 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2614 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2615 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2616 | P0[0] = vce->data; |
---|
2617 | } |
---|
2618 | |
---|
2619 | /* Process row-s1+i data s2+1 --> e2 */ |
---|
2620 | for (j = 1; j < R2; j++) |
---|
2621 | { |
---|
2622 | gIndex = (s1+i)*r2+(s2+j); |
---|
2623 | lIndex = i*R2+j; |
---|
2624 | |
---|
2625 | // printf ("global index = %d, local index = %d\n", gIndex, lIndex); |
---|
2626 | |
---|
2627 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
2628 | |
---|
2629 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
2630 | |
---|
2631 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2632 | |
---|
2633 | if (itvNum < exe_params->intvCapacity) |
---|
2634 | { |
---|
2635 | if (diff < 0) itvNum = -itvNum; |
---|
2636 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2637 | P0[j] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2638 | } |
---|
2639 | else |
---|
2640 | { |
---|
2641 | type[lIndex] = 0; |
---|
2642 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2643 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2644 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2645 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2646 | P0[j] = vce->data; |
---|
2647 | } |
---|
2648 | } |
---|
2649 | |
---|
2650 | float *Pt; |
---|
2651 | Pt = P1; |
---|
2652 | P1 = P0; |
---|
2653 | P0 = Pt; |
---|
2654 | } |
---|
2655 | |
---|
2656 | free(P0); |
---|
2657 | free(P1); |
---|
2658 | size_t exactDataNum = exactLeadNumArray->size; |
---|
2659 | |
---|
2660 | TightDataPointStorageF* tdps; |
---|
2661 | |
---|
2662 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
2663 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
2664 | exactLeadNumArray->array, |
---|
2665 | resiBitArray->array, resiBitArray->size, |
---|
2666 | resiBitsLength, |
---|
2667 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
2668 | |
---|
2669 | //free memory |
---|
2670 | free_DIA(exactLeadNumArray); |
---|
2671 | free_DIA(resiBitArray); |
---|
2672 | free(type); |
---|
2673 | free(vce); |
---|
2674 | free(lce); |
---|
2675 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
2676 | |
---|
2677 | return tdps; |
---|
2678 | } |
---|
2679 | |
---|
2680 | TightDataPointStorageF* SZ_compress_float_3D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f, |
---|
2681 | size_t r1, size_t r2, size_t r3, size_t s1, size_t s2, size_t s3, size_t e1, size_t e2, size_t e3) |
---|
2682 | { |
---|
2683 | unsigned int quantization_intervals; |
---|
2684 | if(exe_params->optQuantMode==1) |
---|
2685 | { |
---|
2686 | quantization_intervals = optimize_intervals_float_3D_subblock(oriData, realPrecision, r1, r2, r3, s1, s2, s3, e1, e2, e3); |
---|
2687 | updateQuantizationInfo(quantization_intervals); |
---|
2688 | } |
---|
2689 | else |
---|
2690 | quantization_intervals = exe_params->intvCapacity; |
---|
2691 | |
---|
2692 | size_t i,j,k; |
---|
2693 | int reqLength; |
---|
2694 | float pred1D, pred2D, pred3D; |
---|
2695 | float diff = 0.0; |
---|
2696 | double itvNum = 0; |
---|
2697 | float *P0, *P1; |
---|
2698 | |
---|
2699 | size_t R1 = e1 - s1 + 1; |
---|
2700 | size_t R2 = e2 - s2 + 1; |
---|
2701 | size_t R3 = e3 - s3 + 1; |
---|
2702 | size_t dataLength = R1*R2*R3; |
---|
2703 | |
---|
2704 | size_t r23 = r2*r3; |
---|
2705 | size_t R23 = R2*R3; |
---|
2706 | |
---|
2707 | P0 = (float*)malloc(R23*sizeof(float)); |
---|
2708 | P1 = (float*)malloc(R23*sizeof(float)); |
---|
2709 | |
---|
2710 | float medianValue = medianValue_f; |
---|
2711 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
2712 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
2713 | |
---|
2714 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
2715 | //type[dataLength]=0; |
---|
2716 | |
---|
2717 | float* spaceFillingValue = oriData; // |
---|
2718 | |
---|
2719 | DynamicIntArray *exactLeadNumArray; |
---|
2720 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
2721 | |
---|
2722 | DynamicByteArray *exactMidByteArray; |
---|
2723 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
2724 | |
---|
2725 | DynamicIntArray *resiBitArray; |
---|
2726 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
2727 | |
---|
2728 | unsigned char preDataBytes[4]; |
---|
2729 | intToBytes_bigEndian(preDataBytes, 0); |
---|
2730 | |
---|
2731 | int reqBytesLength = reqLength/8; |
---|
2732 | int resiBitsLength = reqLength%8; |
---|
2733 | |
---|
2734 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
2735 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
2736 | |
---|
2737 | |
---|
2738 | /////////////////////////// Process layer-s1 /////////////////////////// |
---|
2739 | /* Process Row-s2 data s3*/ |
---|
2740 | size_t gIndex; //global index |
---|
2741 | size_t lIndex; //local index |
---|
2742 | size_t index2D; //local 2D index |
---|
2743 | |
---|
2744 | gIndex = s1*r23+s2*r3+s3; |
---|
2745 | lIndex = 0; |
---|
2746 | index2D = 0; |
---|
2747 | |
---|
2748 | type[lIndex] = 0; |
---|
2749 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2750 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2751 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2752 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2753 | P1[index2D] = vce->data; |
---|
2754 | |
---|
2755 | /* Process Row-s2 data s3+1*/ |
---|
2756 | gIndex = s1*r23+s2*r3+s3+1; |
---|
2757 | lIndex = 1; |
---|
2758 | index2D = 1; |
---|
2759 | |
---|
2760 | pred1D = P1[index2D-1]; |
---|
2761 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2762 | |
---|
2763 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2764 | |
---|
2765 | if (itvNum < exe_params->intvCapacity) |
---|
2766 | { |
---|
2767 | if (diff < 0) itvNum = -itvNum; |
---|
2768 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2769 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2770 | } |
---|
2771 | else |
---|
2772 | { |
---|
2773 | type[lIndex] = 0; |
---|
2774 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2775 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2776 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2777 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2778 | P1[index2D] = vce->data; |
---|
2779 | } |
---|
2780 | |
---|
2781 | /* Process Row-s2 data s3+2 --> data e3 */ |
---|
2782 | for (j = 2; j < R3; j++) |
---|
2783 | { |
---|
2784 | gIndex = s1*r23+s2*r3+s3+j; |
---|
2785 | lIndex = j; |
---|
2786 | index2D = j; |
---|
2787 | |
---|
2788 | pred1D = 2*P1[index2D-1] - P1[index2D-2]; |
---|
2789 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2790 | |
---|
2791 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2792 | |
---|
2793 | if (itvNum < exe_params->intvCapacity) |
---|
2794 | { |
---|
2795 | if (diff < 0) itvNum = -itvNum; |
---|
2796 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2797 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2798 | } |
---|
2799 | else |
---|
2800 | { |
---|
2801 | type[lIndex] = 0; |
---|
2802 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2803 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2804 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2805 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2806 | P1[index2D] = vce->data; |
---|
2807 | } |
---|
2808 | } |
---|
2809 | |
---|
2810 | /* Process Row-s2+1 --> Row-e2 */ |
---|
2811 | for (i = 1; i < R2; i++) |
---|
2812 | { |
---|
2813 | /* Process row-s2+i data s3 */ |
---|
2814 | gIndex = s1*r23+(s2+i)*r3+s3; |
---|
2815 | lIndex = i*R3; |
---|
2816 | index2D = i*R3; |
---|
2817 | |
---|
2818 | pred1D = P1[index2D-R3]; |
---|
2819 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2820 | |
---|
2821 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2822 | |
---|
2823 | if (itvNum < exe_params->intvCapacity) |
---|
2824 | { |
---|
2825 | if (diff < 0) itvNum = -itvNum; |
---|
2826 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2827 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2828 | } |
---|
2829 | else |
---|
2830 | { |
---|
2831 | type[lIndex] = 0; |
---|
2832 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2833 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2834 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2835 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2836 | P1[index2D] = vce->data; |
---|
2837 | } |
---|
2838 | |
---|
2839 | /* Process row-s2+i data s3+1 --> data e3*/ |
---|
2840 | for (j = 1; j < R3; j++) |
---|
2841 | { |
---|
2842 | gIndex = s1*r23+(s2+i)*r3+s3+j; |
---|
2843 | lIndex = i*R3+j; |
---|
2844 | index2D = i*R3+j; |
---|
2845 | |
---|
2846 | pred2D = P1[index2D-1] + P1[index2D-R3] - P1[index2D-R3-1]; |
---|
2847 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
2848 | |
---|
2849 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2850 | |
---|
2851 | if (itvNum < exe_params->intvCapacity) |
---|
2852 | { |
---|
2853 | if (diff < 0) itvNum = -itvNum; |
---|
2854 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2855 | P1[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2856 | } |
---|
2857 | else |
---|
2858 | { |
---|
2859 | type[lIndex] = 0; |
---|
2860 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2861 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2862 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2863 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2864 | P1[index2D] = vce->data; |
---|
2865 | } |
---|
2866 | } |
---|
2867 | } |
---|
2868 | |
---|
2869 | |
---|
2870 | /////////////////////////// Process layer-s1+1 --> layer-e1 /////////////////////////// |
---|
2871 | |
---|
2872 | for (k = 1; k < R1; k++) |
---|
2873 | { |
---|
2874 | /* Process Row-s2 data s3*/ |
---|
2875 | gIndex = (s1+k)*r23+s2*r3+s3; |
---|
2876 | lIndex = k*R23; |
---|
2877 | index2D = 0; |
---|
2878 | |
---|
2879 | pred1D = P1[index2D]; |
---|
2880 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
2881 | |
---|
2882 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2883 | |
---|
2884 | if (itvNum < exe_params->intvCapacity) |
---|
2885 | { |
---|
2886 | if (diff < 0) itvNum = -itvNum; |
---|
2887 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2888 | P0[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2889 | } |
---|
2890 | else |
---|
2891 | { |
---|
2892 | type[lIndex] = 0; |
---|
2893 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2894 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2895 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2896 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2897 | P0[index2D] = vce->data; |
---|
2898 | } |
---|
2899 | |
---|
2900 | /* Process Row-s2 data s3+1 --> data e3 */ |
---|
2901 | for (j = 1; j < R3; j++) |
---|
2902 | { |
---|
2903 | gIndex = (s1+k)*r23+s2*r3+s3+j; |
---|
2904 | lIndex = k*R23+j; |
---|
2905 | index2D = j; |
---|
2906 | |
---|
2907 | pred2D = P0[index2D-1] + P1[index2D] - P1[index2D-1]; |
---|
2908 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
2909 | |
---|
2910 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2911 | |
---|
2912 | if (itvNum < exe_params->intvCapacity) |
---|
2913 | { |
---|
2914 | if (diff < 0) itvNum = -itvNum; |
---|
2915 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2916 | P0[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2917 | } |
---|
2918 | else |
---|
2919 | { |
---|
2920 | type[lIndex] = 0; |
---|
2921 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2922 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2923 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2924 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2925 | P0[index2D] = vce->data; |
---|
2926 | } |
---|
2927 | } |
---|
2928 | |
---|
2929 | /* Process Row-s2+1 --> Row-e2 */ |
---|
2930 | for (i = 1; i < R2; i++) |
---|
2931 | { |
---|
2932 | /* Process Row-s2+i data s3 */ |
---|
2933 | gIndex = (s1+k)*r23+(s2+i)*r3+s3; |
---|
2934 | lIndex = k*R23+i*R3; |
---|
2935 | index2D = i*R3; |
---|
2936 | |
---|
2937 | pred2D = P0[index2D-R3] + P1[index2D] - P1[index2D-R3]; |
---|
2938 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
2939 | |
---|
2940 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2941 | |
---|
2942 | if (itvNum < exe_params->intvCapacity) |
---|
2943 | { |
---|
2944 | if (diff < 0) itvNum = -itvNum; |
---|
2945 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2946 | P0[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2947 | } |
---|
2948 | else |
---|
2949 | { |
---|
2950 | type[lIndex] = 0; |
---|
2951 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2952 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2953 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2954 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2955 | P0[index2D] = vce->data; |
---|
2956 | } |
---|
2957 | |
---|
2958 | /* Process Row-s2+i data s3+1 --> data e3 */ |
---|
2959 | for (j = 1; j < R3; j++) |
---|
2960 | { |
---|
2961 | gIndex = (s1+k)*r23+(s2+i)*r3+s3+j; |
---|
2962 | lIndex = k*R23+i*R3+j; |
---|
2963 | index2D = i*R3+j; |
---|
2964 | |
---|
2965 | // printf ("global index = %d, local index = %d\n", gIndex, lIndex); |
---|
2966 | |
---|
2967 | pred3D = P0[index2D-1] + P0[index2D-R3]+ P1[index2D] - P0[index2D-R3-1] - P1[index2D-R3] - P1[index2D-1] + P1[index2D-R3-1]; |
---|
2968 | diff = spaceFillingValue[gIndex] - pred3D; |
---|
2969 | |
---|
2970 | itvNum = fabs(diff)/realPrecision + 1; |
---|
2971 | |
---|
2972 | if (itvNum < exe_params->intvCapacity) |
---|
2973 | { |
---|
2974 | if (diff < 0) itvNum = -itvNum; |
---|
2975 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
2976 | P0[index2D] = pred3D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
2977 | } |
---|
2978 | else |
---|
2979 | { |
---|
2980 | type[lIndex] = 0; |
---|
2981 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
2982 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
2983 | memcpy(preDataBytes,vce->curBytes,4); |
---|
2984 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
2985 | P0[index2D] = vce->data; |
---|
2986 | } |
---|
2987 | } |
---|
2988 | } |
---|
2989 | |
---|
2990 | float *Pt; |
---|
2991 | Pt = P1; |
---|
2992 | P1 = P0; |
---|
2993 | P0 = Pt; |
---|
2994 | } |
---|
2995 | |
---|
2996 | free(P0); |
---|
2997 | free(P1); |
---|
2998 | size_t exactDataNum = exactLeadNumArray->size; |
---|
2999 | |
---|
3000 | TightDataPointStorageF* tdps; |
---|
3001 | |
---|
3002 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
3003 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
3004 | exactLeadNumArray->array, |
---|
3005 | resiBitArray->array, resiBitArray->size, |
---|
3006 | resiBitsLength, |
---|
3007 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
3008 | |
---|
3009 | //free memory |
---|
3010 | free_DIA(exactLeadNumArray); |
---|
3011 | free_DIA(resiBitArray); |
---|
3012 | free(type); |
---|
3013 | free(vce); |
---|
3014 | free(lce); |
---|
3015 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
3016 | |
---|
3017 | return tdps; |
---|
3018 | } |
---|
3019 | |
---|
3020 | TightDataPointStorageF* SZ_compress_float_4D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f, |
---|
3021 | size_t r1, size_t r2, size_t r3, size_t r4, size_t s1, size_t s2, size_t s3, size_t s4, size_t e1, size_t e2, size_t e3, size_t e4) |
---|
3022 | { |
---|
3023 | unsigned int quantization_intervals; |
---|
3024 | if(exe_params->optQuantMode==1) |
---|
3025 | { |
---|
3026 | quantization_intervals = optimize_intervals_float_4D_subblock(oriData, realPrecision, r1, r2, r3, r4, s1, s2, s3, s4, e1, e2, e3, e4); |
---|
3027 | updateQuantizationInfo(quantization_intervals); |
---|
3028 | } |
---|
3029 | else |
---|
3030 | quantization_intervals = exe_params->intvCapacity; |
---|
3031 | |
---|
3032 | size_t i,j,k; |
---|
3033 | int reqLength; |
---|
3034 | float pred1D, pred2D, pred3D; |
---|
3035 | float diff = 0.0; |
---|
3036 | double itvNum = 0; |
---|
3037 | float *P0, *P1; |
---|
3038 | |
---|
3039 | size_t R1 = e1 - s1 + 1; |
---|
3040 | size_t R2 = e2 - s2 + 1; |
---|
3041 | size_t R3 = e3 - s3 + 1; |
---|
3042 | size_t R4 = e4 - s4 + 1; |
---|
3043 | |
---|
3044 | size_t dataLength = R1*R2*R3*R4; |
---|
3045 | |
---|
3046 | size_t r34 = r3*r4; |
---|
3047 | size_t r234 = r2*r3*r4; |
---|
3048 | size_t R34 = R3*R4; |
---|
3049 | size_t R234 = R2*R3*R4; |
---|
3050 | |
---|
3051 | P0 = (float*)malloc(R34*sizeof(float)); |
---|
3052 | P1 = (float*)malloc(R34*sizeof(float)); |
---|
3053 | |
---|
3054 | float medianValue = medianValue_f; |
---|
3055 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
3056 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
3057 | |
---|
3058 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
3059 | |
---|
3060 | float* spaceFillingValue = oriData; // |
---|
3061 | |
---|
3062 | DynamicIntArray *exactLeadNumArray; |
---|
3063 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
3064 | |
---|
3065 | DynamicByteArray *exactMidByteArray; |
---|
3066 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
3067 | |
---|
3068 | DynamicIntArray *resiBitArray; |
---|
3069 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
3070 | |
---|
3071 | unsigned char preDataBytes[4]; |
---|
3072 | intToBytes_bigEndian(preDataBytes, 0); |
---|
3073 | |
---|
3074 | int reqBytesLength = reqLength/8; |
---|
3075 | int resiBitsLength = reqLength%8; |
---|
3076 | |
---|
3077 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
3078 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
3079 | |
---|
3080 | |
---|
3081 | size_t l; |
---|
3082 | for (l = 0; l < R1; l++) |
---|
3083 | { |
---|
3084 | |
---|
3085 | /////////////////////////// Process layer-s2 /////////////////////////// |
---|
3086 | /* Process Row-s3 data s4*/ |
---|
3087 | size_t gIndex; //global index |
---|
3088 | size_t lIndex; //local index |
---|
3089 | size_t index2D; //local 2D index |
---|
3090 | |
---|
3091 | gIndex = (s1+l)*r234+s2*r34+s3*r4+s4; |
---|
3092 | lIndex = l*R234; |
---|
3093 | index2D = 0; |
---|
3094 | |
---|
3095 | type[lIndex] = 0; |
---|
3096 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3097 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3098 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3099 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3100 | P1[index2D] = vce->data; |
---|
3101 | |
---|
3102 | /* Process Row-s3 data s4+1*/ |
---|
3103 | gIndex = (s1+l)*r234+s2*r34+s3*r4+s4+1; |
---|
3104 | lIndex = l*R234+1; |
---|
3105 | index2D = 1; |
---|
3106 | |
---|
3107 | pred1D = P1[index2D-1]; |
---|
3108 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
3109 | |
---|
3110 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3111 | |
---|
3112 | if (itvNum < exe_params->intvCapacity) |
---|
3113 | { |
---|
3114 | if (diff < 0) itvNum = -itvNum; |
---|
3115 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3116 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3117 | } |
---|
3118 | else |
---|
3119 | { |
---|
3120 | type[lIndex] = 0; |
---|
3121 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3122 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3123 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3124 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3125 | P1[index2D] = vce->data; |
---|
3126 | } |
---|
3127 | |
---|
3128 | /* Process Row-s3 data s4+2 --> data e4 */ |
---|
3129 | for (j = 2; j < R4; j++) |
---|
3130 | { |
---|
3131 | gIndex = (s1+l)*r234+s2*r34+s3*r4+s4+j; |
---|
3132 | lIndex = l*R234+j; |
---|
3133 | index2D = j; |
---|
3134 | |
---|
3135 | pred1D = 2*P1[index2D-1] - P1[index2D-2]; |
---|
3136 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
3137 | |
---|
3138 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3139 | |
---|
3140 | if (itvNum < exe_params->intvCapacity) |
---|
3141 | { |
---|
3142 | if (diff < 0) itvNum = -itvNum; |
---|
3143 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3144 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3145 | } |
---|
3146 | else |
---|
3147 | { |
---|
3148 | type[lIndex] = 0; |
---|
3149 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3150 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3151 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3152 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3153 | P1[index2D] = vce->data; |
---|
3154 | } |
---|
3155 | } |
---|
3156 | |
---|
3157 | /* Process Row-s3+1 --> Row-e3 */ |
---|
3158 | for (i = 1; i < R3; i++) |
---|
3159 | { |
---|
3160 | /* Process row-s2+i data s3 */ |
---|
3161 | gIndex = (s1+l)*r234+s2*r34+(s3+i)*r4+s4; |
---|
3162 | lIndex = l*R234+i*R4; |
---|
3163 | index2D = i*R4; |
---|
3164 | |
---|
3165 | pred1D = P1[index2D-R4]; |
---|
3166 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
3167 | |
---|
3168 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3169 | |
---|
3170 | if (itvNum < exe_params->intvCapacity) |
---|
3171 | { |
---|
3172 | if (diff < 0) itvNum = -itvNum; |
---|
3173 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3174 | P1[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3175 | } |
---|
3176 | else |
---|
3177 | { |
---|
3178 | type[lIndex] = 0; |
---|
3179 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3180 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3181 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3182 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3183 | P1[index2D] = vce->data; |
---|
3184 | } |
---|
3185 | |
---|
3186 | /* Process row-s3+i data s4+1 --> data e4*/ |
---|
3187 | for (j = 1; j < R4; j++) |
---|
3188 | { |
---|
3189 | gIndex = (s1+l)*r234+s2*r34+(s3+i)*r4+s4+j; |
---|
3190 | lIndex = l*R234+i*R4+j; |
---|
3191 | index2D = i*R4+j; |
---|
3192 | |
---|
3193 | pred2D = P1[index2D-1] + P1[index2D-R4] - P1[index2D-R4-1]; |
---|
3194 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
3195 | |
---|
3196 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3197 | |
---|
3198 | if (itvNum < exe_params->intvCapacity) |
---|
3199 | { |
---|
3200 | if (diff < 0) itvNum = -itvNum; |
---|
3201 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3202 | P1[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3203 | } |
---|
3204 | else |
---|
3205 | { |
---|
3206 | type[lIndex] = 0; |
---|
3207 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3208 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3209 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3210 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3211 | P1[index2D] = vce->data; |
---|
3212 | } |
---|
3213 | } |
---|
3214 | } |
---|
3215 | |
---|
3216 | |
---|
3217 | /////////////////////////// Process layer-s2+1 --> layer-e2 /////////////////////////// |
---|
3218 | |
---|
3219 | for (k = 1; k < R2; k++) |
---|
3220 | { |
---|
3221 | /* Process Row-s3 data s4*/ |
---|
3222 | gIndex = (s1+l)*r234+(s2+k)*r34+s3*r4+s4; |
---|
3223 | lIndex = l*R234+k*R34; |
---|
3224 | index2D = 0; |
---|
3225 | |
---|
3226 | pred1D = P1[index2D]; |
---|
3227 | diff = spaceFillingValue[gIndex] - pred1D; |
---|
3228 | |
---|
3229 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3230 | |
---|
3231 | if (itvNum < exe_params->intvCapacity) |
---|
3232 | { |
---|
3233 | if (diff < 0) itvNum = -itvNum; |
---|
3234 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3235 | P0[index2D] = pred1D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3236 | } |
---|
3237 | else |
---|
3238 | { |
---|
3239 | type[lIndex] = 0; |
---|
3240 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3241 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3242 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3243 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3244 | P0[index2D] = vce->data; |
---|
3245 | } |
---|
3246 | |
---|
3247 | /* Process Row-s3 data s4+1 --> data e4 */ |
---|
3248 | for (j = 1; j < R4; j++) |
---|
3249 | { |
---|
3250 | gIndex = (s1+l)*r234+(s2+k)*r34+s3*r4+s4+j; |
---|
3251 | lIndex = l*R234+k*R34+j; |
---|
3252 | index2D = j; |
---|
3253 | |
---|
3254 | pred2D = P0[index2D-1] + P1[index2D] - P1[index2D-1]; |
---|
3255 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
3256 | |
---|
3257 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3258 | |
---|
3259 | if (itvNum < exe_params->intvCapacity) |
---|
3260 | { |
---|
3261 | if (diff < 0) itvNum = -itvNum; |
---|
3262 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3263 | P0[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3264 | } |
---|
3265 | else |
---|
3266 | { |
---|
3267 | type[lIndex] = 0; |
---|
3268 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3269 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3270 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3271 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3272 | P0[index2D] = vce->data; |
---|
3273 | } |
---|
3274 | } |
---|
3275 | |
---|
3276 | /* Process Row-s3+1 --> Row-e3 */ |
---|
3277 | for (i = 1; i < R3; i++) |
---|
3278 | { |
---|
3279 | /* Process Row-s3+i data s4 */ |
---|
3280 | gIndex = (s1+l)*r234+(s2+k)*r34+(s3+i)*r4+s4; |
---|
3281 | lIndex = l*R234+k*R34+i*R4; |
---|
3282 | index2D = i*R4; |
---|
3283 | |
---|
3284 | pred2D = P0[index2D-R4] + P1[index2D] - P1[index2D-R4]; |
---|
3285 | diff = spaceFillingValue[gIndex] - pred2D; |
---|
3286 | |
---|
3287 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3288 | |
---|
3289 | if (itvNum < exe_params->intvCapacity) |
---|
3290 | { |
---|
3291 | if (diff < 0) itvNum = -itvNum; |
---|
3292 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3293 | P0[index2D] = pred2D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3294 | } |
---|
3295 | else |
---|
3296 | { |
---|
3297 | type[lIndex] = 0; |
---|
3298 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3299 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3300 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3301 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3302 | P0[index2D] = vce->data; |
---|
3303 | } |
---|
3304 | |
---|
3305 | /* Process Row-s3+i data s4+1 --> data e4 */ |
---|
3306 | for (j = 1; j < R4; j++) |
---|
3307 | { |
---|
3308 | gIndex = (s1+l)*r234+(s2+k)*r34+(s3+i)*r4+s4+j; |
---|
3309 | lIndex = l*R234+k*R34+i*R4+j; |
---|
3310 | index2D = i*R4+j; |
---|
3311 | |
---|
3312 | // printf ("global index = %d, local index = %d\n", gIndex, lIndex); |
---|
3313 | |
---|
3314 | pred3D = P0[index2D-1] + P0[index2D-R4]+ P1[index2D] - P0[index2D-R4-1] - P1[index2D-R4] - P1[index2D-1] + P1[index2D-R4-1]; |
---|
3315 | diff = spaceFillingValue[gIndex] - pred3D; |
---|
3316 | |
---|
3317 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3318 | |
---|
3319 | if (itvNum < exe_params->intvCapacity) |
---|
3320 | { |
---|
3321 | if (diff < 0) itvNum = -itvNum; |
---|
3322 | type[lIndex] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3323 | P0[index2D] = pred3D + 2 * (type[lIndex] - exe_params->intvRadius) * realPrecision; |
---|
3324 | } |
---|
3325 | else |
---|
3326 | { |
---|
3327 | type[lIndex] = 0; |
---|
3328 | compressSingleFloatValue(vce, spaceFillingValue[gIndex], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
3329 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
3330 | memcpy(preDataBytes,vce->curBytes,4); |
---|
3331 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
3332 | P0[index2D] = vce->data; |
---|
3333 | } |
---|
3334 | } |
---|
3335 | } |
---|
3336 | |
---|
3337 | float *Pt; |
---|
3338 | Pt = P1; |
---|
3339 | P1 = P0; |
---|
3340 | P0 = Pt; |
---|
3341 | } |
---|
3342 | |
---|
3343 | } |
---|
3344 | |
---|
3345 | free(P0); |
---|
3346 | free(P1); |
---|
3347 | size_t exactDataNum = exactLeadNumArray->size; |
---|
3348 | |
---|
3349 | TightDataPointStorageF* tdps; |
---|
3350 | |
---|
3351 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
3352 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
3353 | exactLeadNumArray->array, |
---|
3354 | resiBitArray->array, resiBitArray->size, |
---|
3355 | resiBitsLength, |
---|
3356 | realPrecision, medianValue, (char)reqLength, quantization_intervals, NULL, 0, 0); |
---|
3357 | |
---|
3358 | //free memory |
---|
3359 | free_DIA(exactLeadNumArray); |
---|
3360 | free_DIA(resiBitArray); |
---|
3361 | free(type); |
---|
3362 | free(vce); |
---|
3363 | free(lce); |
---|
3364 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
3365 | |
---|
3366 | return tdps; |
---|
3367 | } |
---|
3368 | |
---|
3369 | unsigned int optimize_intervals_float_3D_opt(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision) |
---|
3370 | { |
---|
3371 | size_t i; |
---|
3372 | size_t radiusIndex; |
---|
3373 | size_t r23=r2*r3; |
---|
3374 | float pred_value = 0, pred_err; |
---|
3375 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3376 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3377 | size_t totalSampleSize = 0;//(r1-1)*(r2-1)*(r3-1)/confparams_cpr->sampleDistance; |
---|
3378 | |
---|
3379 | size_t offset_count = confparams_cpr->sampleDistance - 2; // count r3 offset |
---|
3380 | size_t offset_count_2; |
---|
3381 | float * data_pos = oriData + r23 + r3 + offset_count; |
---|
3382 | size_t n1_count = 1, n2_count = 1; // count i,j sum |
---|
3383 | size_t len = r1 * r2 * r3; |
---|
3384 | while(data_pos - oriData < len){ |
---|
3385 | totalSampleSize++; |
---|
3386 | pred_value = data_pos[-1] + data_pos[-r3] + data_pos[-r23] - data_pos[-1-r23] - data_pos[-r3-1] - data_pos[-r3-r23] + data_pos[-r3-r23-1]; |
---|
3387 | pred_err = fabs(pred_value - *data_pos); |
---|
3388 | radiusIndex = (pred_err/realPrecision+1)/2; |
---|
3389 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
3390 | { |
---|
3391 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
3392 | //printf("radiusIndex=%d\n", radiusIndex); |
---|
3393 | } |
---|
3394 | intervals[radiusIndex]++; |
---|
3395 | // printf("TEST: %ld, i: %ld\tj: %ld\tk: %ld\n", data_pos - oriData); |
---|
3396 | // fflush(stdout); |
---|
3397 | offset_count += confparams_cpr->sampleDistance; |
---|
3398 | if(offset_count >= r3){ |
---|
3399 | n2_count ++; |
---|
3400 | if(n2_count == r2){ |
---|
3401 | n1_count ++; |
---|
3402 | n2_count = 1; |
---|
3403 | data_pos += r3; |
---|
3404 | } |
---|
3405 | offset_count_2 = (n1_count + n2_count) % confparams_cpr->sampleDistance; |
---|
3406 | data_pos += (r3 + confparams_cpr->sampleDistance - offset_count) + (confparams_cpr->sampleDistance - offset_count_2); |
---|
3407 | offset_count = (confparams_cpr->sampleDistance - offset_count_2); |
---|
3408 | if(offset_count == 0) offset_count ++; |
---|
3409 | } |
---|
3410 | else data_pos += confparams_cpr->sampleDistance; |
---|
3411 | } |
---|
3412 | // printf("sample_count: %ld\n", sample_count); |
---|
3413 | // fflush(stdout); |
---|
3414 | // if(*max_freq < 0.15) *max_freq *= 2; |
---|
3415 | //compute the appropriate number |
---|
3416 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
3417 | size_t sum = 0; |
---|
3418 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
3419 | { |
---|
3420 | sum += intervals[i]; |
---|
3421 | if(sum>targetCount) |
---|
3422 | break; |
---|
3423 | } |
---|
3424 | if(i>=confparams_cpr->maxRangeRadius) |
---|
3425 | i = confparams_cpr->maxRangeRadius-1; |
---|
3426 | unsigned int accIntervals = 2*(i+1); |
---|
3427 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
3428 | |
---|
3429 | if(powerOf2<32) |
---|
3430 | powerOf2 = 32; |
---|
3431 | free(intervals); |
---|
3432 | //printf("targetCount=%d, sum=%d, totalSampleSize=%d, ratio=%f, accIntervals=%d, powerOf2=%d\n", targetCount, sum, totalSampleSize, (double)sum/(double)totalSampleSize, accIntervals, powerOf2); |
---|
3433 | return powerOf2; |
---|
3434 | } |
---|
3435 | |
---|
3436 | size_t SZ_compress_float_3D_MDQ_RA_block(float * block_ori_data, float * mean, size_t dim_0, size_t dim_1, size_t dim_2, size_t block_dim_0, size_t block_dim_1, size_t block_dim_2, double realPrecision, float * P0, float * P1, int * type, float * unpredictable_data){ |
---|
3437 | |
---|
3438 | size_t dim0_offset = dim_1 * dim_2; |
---|
3439 | size_t dim1_offset = dim_2; |
---|
3440 | |
---|
3441 | // data_pos = block_ori_data; |
---|
3442 | // for(size_t i=0; i<block_dim_0; i++){ |
---|
3443 | // for(size_t j=0; j<block_dim_1; j++){ |
---|
3444 | // for(size_t k=0; k<block_dim_2; k++){ |
---|
3445 | // sum += *data_pos; |
---|
3446 | // data_pos ++; |
---|
3447 | // } |
---|
3448 | // data_pos += dim1_offset - block_dim_2; |
---|
3449 | // } |
---|
3450 | // data_pos += dim0_offset - block_dim_1 * dim1_offset; |
---|
3451 | // } |
---|
3452 | // size_t num_elements = block_dim_0 * block_dim_1 * block_dim_2; |
---|
3453 | // if(num_elements > 0) mean[0] = sum / num_elements; |
---|
3454 | // else mean[0] = 0.0; |
---|
3455 | mean[0] = block_ori_data[0]; |
---|
3456 | |
---|
3457 | size_t unpredictable_count = 0; |
---|
3458 | size_t r1, r2, r3; |
---|
3459 | r1 = block_dim_0; |
---|
3460 | r2 = block_dim_1; |
---|
3461 | r3 = block_dim_2; |
---|
3462 | |
---|
3463 | float * cur_data_pos = block_ori_data; |
---|
3464 | float curData; |
---|
3465 | float pred1D, pred2D, pred3D; |
---|
3466 | double itvNum; |
---|
3467 | double diff; |
---|
3468 | size_t i, j, k; |
---|
3469 | size_t r23 = r2*r3; |
---|
3470 | // Process Row-0 data 0 |
---|
3471 | pred1D = mean[0]; |
---|
3472 | curData = *cur_data_pos; |
---|
3473 | diff = curData - pred1D; |
---|
3474 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3475 | if (itvNum < exe_params->intvCapacity){ |
---|
3476 | if (diff < 0) itvNum = -itvNum; |
---|
3477 | type[0] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3478 | P1[0] = pred1D + 2 * (type[0] - exe_params->intvRadius) * realPrecision; |
---|
3479 | //ganrantee comporession error against the case of machine-epsilon |
---|
3480 | if(fabs(curData-P1[0])>realPrecision){ |
---|
3481 | type[0] = 0; |
---|
3482 | P1[0] = curData; |
---|
3483 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3484 | } |
---|
3485 | } |
---|
3486 | else{ |
---|
3487 | type[0] = 0; |
---|
3488 | P1[0] = curData; |
---|
3489 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3490 | } |
---|
3491 | |
---|
3492 | /* Process Row-0 data 1*/ |
---|
3493 | pred1D = P1[0]; |
---|
3494 | curData = cur_data_pos[1]; |
---|
3495 | diff = curData - pred1D; |
---|
3496 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3497 | if (itvNum < exe_params->intvCapacity){ |
---|
3498 | if (diff < 0) itvNum = -itvNum; |
---|
3499 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3500 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
3501 | //ganrantee comporession error against the case of machine-epsilon |
---|
3502 | if(fabs(curData-P1[1])>realPrecision){ |
---|
3503 | type[1] = 0; |
---|
3504 | P1[1] = curData; |
---|
3505 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3506 | } |
---|
3507 | } |
---|
3508 | else{ |
---|
3509 | type[1] = 0; |
---|
3510 | P1[1] = curData; |
---|
3511 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3512 | } |
---|
3513 | /* Process Row-0 data 2 --> data r3-1 */ |
---|
3514 | for (j = 2; j < r3; j++){ |
---|
3515 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
3516 | curData = cur_data_pos[j]; |
---|
3517 | diff = curData - pred1D; |
---|
3518 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3519 | if (itvNum < exe_params->intvCapacity){ |
---|
3520 | if (diff < 0) itvNum = -itvNum; |
---|
3521 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3522 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
3523 | //ganrantee comporession error against the case of machine-epsilon |
---|
3524 | if(fabs(curData-P1[j])>realPrecision){ |
---|
3525 | type[j] = 0; |
---|
3526 | P1[j] = curData; |
---|
3527 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3528 | } |
---|
3529 | } |
---|
3530 | else{ |
---|
3531 | type[j] = 0; |
---|
3532 | P1[j] = curData; |
---|
3533 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3534 | } |
---|
3535 | } |
---|
3536 | cur_data_pos += dim1_offset; |
---|
3537 | |
---|
3538 | /* Process Row-1 --> Row-r2-1 */ |
---|
3539 | size_t index; |
---|
3540 | for (i = 1; i < r2; i++) |
---|
3541 | { |
---|
3542 | /* Process row-i data 0 */ |
---|
3543 | index = i*r3; |
---|
3544 | pred1D = P1[index-r3]; |
---|
3545 | curData = *cur_data_pos; |
---|
3546 | diff = curData - pred1D; |
---|
3547 | |
---|
3548 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3549 | |
---|
3550 | if (itvNum < exe_params->intvCapacity) |
---|
3551 | { |
---|
3552 | if (diff < 0) itvNum = -itvNum; |
---|
3553 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3554 | P1[index] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3555 | |
---|
3556 | //ganrantee comporession error against the case of machine-epsilon |
---|
3557 | if(fabs(curData-P1[index])>realPrecision) |
---|
3558 | { |
---|
3559 | type[index] = 0; |
---|
3560 | P1[index] = curData; |
---|
3561 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3562 | } |
---|
3563 | } |
---|
3564 | else |
---|
3565 | { |
---|
3566 | type[index] = 0; |
---|
3567 | P1[index] = curData; |
---|
3568 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3569 | } |
---|
3570 | |
---|
3571 | /* Process row-i data 1 --> data r3-1*/ |
---|
3572 | for (j = 1; j < r3; j++) |
---|
3573 | { |
---|
3574 | index = i*r3+j; |
---|
3575 | pred2D = P1[index-1] + P1[index-r3] - P1[index-r3-1]; |
---|
3576 | |
---|
3577 | curData = cur_data_pos[j]; |
---|
3578 | diff = curData - pred2D; |
---|
3579 | |
---|
3580 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3581 | |
---|
3582 | if (itvNum < exe_params->intvCapacity) |
---|
3583 | { |
---|
3584 | if (diff < 0) itvNum = -itvNum; |
---|
3585 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3586 | P1[index] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3587 | |
---|
3588 | //ganrantee comporession error against the case of machine-epsilon |
---|
3589 | if(fabs(curData-P1[index])>realPrecision) |
---|
3590 | { |
---|
3591 | type[index] = 0; |
---|
3592 | P1[index] = curData; |
---|
3593 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3594 | } |
---|
3595 | } |
---|
3596 | else |
---|
3597 | { |
---|
3598 | type[index] = 0; |
---|
3599 | P1[index] = curData; |
---|
3600 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3601 | } |
---|
3602 | } |
---|
3603 | cur_data_pos += dim1_offset; |
---|
3604 | } |
---|
3605 | cur_data_pos += dim0_offset - r2 * dim1_offset; |
---|
3606 | |
---|
3607 | /////////////////////////// Process layer-1 --> layer-r1-1 /////////////////////////// |
---|
3608 | |
---|
3609 | for (k = 1; k < r1; k++) |
---|
3610 | { |
---|
3611 | /* Process Row-0 data 0*/ |
---|
3612 | index = k*r23; |
---|
3613 | pred1D = P1[0]; |
---|
3614 | curData = *cur_data_pos; |
---|
3615 | diff = curData - pred1D; |
---|
3616 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3617 | if (itvNum < exe_params->intvCapacity) |
---|
3618 | { |
---|
3619 | if (diff < 0) itvNum = -itvNum; |
---|
3620 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3621 | P0[0] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3622 | //ganrantee comporession error against the case of machine-epsilon |
---|
3623 | if(fabs(curData-P0[0])>realPrecision) |
---|
3624 | { |
---|
3625 | type[index] = 0; |
---|
3626 | P0[0] = curData; |
---|
3627 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3628 | } |
---|
3629 | } |
---|
3630 | else |
---|
3631 | { |
---|
3632 | type[index] = 0; |
---|
3633 | P0[0] = curData; |
---|
3634 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3635 | } |
---|
3636 | /* Process Row-0 data 1 --> data r3-1 */ |
---|
3637 | for (j = 1; j < r3; j++) |
---|
3638 | { |
---|
3639 | //index = k*r2*r3+j; |
---|
3640 | index ++; |
---|
3641 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
3642 | curData = cur_data_pos[j]; |
---|
3643 | diff = curData - pred2D; |
---|
3644 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3645 | if (itvNum < exe_params->intvCapacity) |
---|
3646 | { |
---|
3647 | if (diff < 0) itvNum = -itvNum; |
---|
3648 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3649 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3650 | //ganrantee comporession error against the case of machine-epsilon |
---|
3651 | if(fabs(curData-P0[j])>realPrecision) |
---|
3652 | { |
---|
3653 | type[index] = 0; |
---|
3654 | P0[j] = curData; |
---|
3655 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3656 | } |
---|
3657 | } |
---|
3658 | else |
---|
3659 | { |
---|
3660 | type[index] = 0; |
---|
3661 | P0[j] = curData; |
---|
3662 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3663 | } |
---|
3664 | } |
---|
3665 | |
---|
3666 | cur_data_pos += dim1_offset; |
---|
3667 | /* Process Row-1 --> Row-r2-1 */ |
---|
3668 | size_t index2D; |
---|
3669 | for (i = 1; i < r2; i++) |
---|
3670 | { |
---|
3671 | /* Process Row-i data 0 */ |
---|
3672 | index = k*r23 + i*r3; |
---|
3673 | index2D = i*r3; |
---|
3674 | pred2D = P0[index2D-r3] + P1[index2D] - P1[index2D-r3]; |
---|
3675 | curData = *cur_data_pos; |
---|
3676 | diff = curData - pred2D; |
---|
3677 | |
---|
3678 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3679 | |
---|
3680 | if (itvNum < exe_params->intvCapacity) |
---|
3681 | { |
---|
3682 | if (diff < 0) itvNum = -itvNum; |
---|
3683 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3684 | P0[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3685 | //ganrantee comporession error against the case of machine-epsilon |
---|
3686 | if(fabs(curData-P0[index2D])>realPrecision) |
---|
3687 | { |
---|
3688 | type[index] = 0; |
---|
3689 | P0[index2D] = curData; |
---|
3690 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3691 | } |
---|
3692 | } |
---|
3693 | else |
---|
3694 | { |
---|
3695 | type[index] = 0; |
---|
3696 | P0[index2D] = curData; |
---|
3697 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3698 | } |
---|
3699 | |
---|
3700 | /* Process Row-i data 1 --> data r3-1 */ |
---|
3701 | for (j = 1; j < r3; j++) |
---|
3702 | { |
---|
3703 | //index = k*r2*r3 + i*r3 + j; |
---|
3704 | index ++; |
---|
3705 | index2D = i*r3 + j; |
---|
3706 | pred3D = P0[index2D-1] + P0[index2D-r3]+ P1[index2D] - P0[index2D-r3-1] - P1[index2D-r3] - P1[index2D-1] + P1[index2D-r3-1]; |
---|
3707 | curData = cur_data_pos[j]; |
---|
3708 | diff = curData - pred3D; |
---|
3709 | |
---|
3710 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3711 | |
---|
3712 | if (itvNum < exe_params->intvCapacity) |
---|
3713 | { |
---|
3714 | if (diff < 0) itvNum = -itvNum; |
---|
3715 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3716 | P0[index2D] = pred3D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
3717 | |
---|
3718 | //ganrantee comporession error against the case of machine-epsilon |
---|
3719 | if(fabs(curData-P0[index2D])>realPrecision) |
---|
3720 | { |
---|
3721 | type[index] = 0; |
---|
3722 | P0[index2D] = curData; |
---|
3723 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3724 | } |
---|
3725 | } |
---|
3726 | else |
---|
3727 | { |
---|
3728 | type[index] = 0; |
---|
3729 | P0[index2D] = curData; |
---|
3730 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3731 | } |
---|
3732 | } |
---|
3733 | cur_data_pos += dim1_offset; |
---|
3734 | } |
---|
3735 | cur_data_pos += dim0_offset - r2 * dim1_offset; |
---|
3736 | float *Pt; |
---|
3737 | Pt = P1; |
---|
3738 | P1 = P0; |
---|
3739 | P0 = Pt; |
---|
3740 | } |
---|
3741 | |
---|
3742 | return unpredictable_count; |
---|
3743 | } |
---|
3744 | |
---|
3745 | unsigned int optimize_intervals_float_2D_opt(float *oriData, size_t r1, size_t r2, double realPrecision) |
---|
3746 | { |
---|
3747 | size_t i; |
---|
3748 | size_t radiusIndex; |
---|
3749 | float pred_value = 0, pred_err; |
---|
3750 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3751 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3752 | size_t totalSampleSize = 0;//(r1-1)*(r2-1)/confparams_cpr->sampleDistance; |
---|
3753 | |
---|
3754 | //float max = oriData[0]; |
---|
3755 | //float min = oriData[0]; |
---|
3756 | |
---|
3757 | size_t offset_count = confparams_cpr->sampleDistance - 1; // count r2 offset |
---|
3758 | size_t offset_count_2; |
---|
3759 | float * data_pos = oriData + r2 + offset_count; |
---|
3760 | size_t n1_count = 1; // count i sum |
---|
3761 | size_t len = r1 * r2; |
---|
3762 | while(data_pos - oriData < len){ |
---|
3763 | totalSampleSize++; |
---|
3764 | pred_value = data_pos[-1] + data_pos[-r2] - data_pos[-r2-1]; |
---|
3765 | pred_err = fabs(pred_value - *data_pos); |
---|
3766 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
3767 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
3768 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
3769 | intervals[radiusIndex]++; |
---|
3770 | |
---|
3771 | offset_count += confparams_cpr->sampleDistance; |
---|
3772 | if(offset_count >= r2){ |
---|
3773 | n1_count ++; |
---|
3774 | offset_count_2 = n1_count % confparams_cpr->sampleDistance; |
---|
3775 | data_pos += (r2 + confparams_cpr->sampleDistance - offset_count) + (confparams_cpr->sampleDistance - offset_count_2); |
---|
3776 | offset_count = (confparams_cpr->sampleDistance - offset_count_2); |
---|
3777 | if(offset_count == 0) offset_count ++; |
---|
3778 | } |
---|
3779 | else data_pos += confparams_cpr->sampleDistance; |
---|
3780 | } |
---|
3781 | |
---|
3782 | //compute the appropriate number |
---|
3783 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
3784 | size_t sum = 0; |
---|
3785 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
3786 | { |
---|
3787 | sum += intervals[i]; |
---|
3788 | if(sum>targetCount) |
---|
3789 | break; |
---|
3790 | } |
---|
3791 | if(i>=confparams_cpr->maxRangeRadius) |
---|
3792 | i = confparams_cpr->maxRangeRadius-1; |
---|
3793 | unsigned int accIntervals = 2*(i+1); |
---|
3794 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
3795 | |
---|
3796 | if(powerOf2<32) |
---|
3797 | powerOf2 = 32; |
---|
3798 | |
---|
3799 | free(intervals); |
---|
3800 | return powerOf2; |
---|
3801 | } |
---|
3802 | |
---|
3803 | unsigned int optimize_intervals_float_1D_opt(float *oriData, size_t dataLength, double realPrecision) |
---|
3804 | { |
---|
3805 | size_t i = 0, radiusIndex; |
---|
3806 | float pred_value = 0, pred_err; |
---|
3807 | size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3808 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t)); |
---|
3809 | size_t totalSampleSize = 0;//dataLength/confparams_cpr->sampleDistance; |
---|
3810 | |
---|
3811 | float * data_pos = oriData + 2; |
---|
3812 | while(data_pos - oriData < dataLength){ |
---|
3813 | totalSampleSize++; |
---|
3814 | //pred_value = 2*data_pos[-1] - data_pos[-2]; |
---|
3815 | pred_value = data_pos[-1]; |
---|
3816 | pred_err = fabs(pred_value - *data_pos); |
---|
3817 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
3818 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
3819 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
3820 | intervals[radiusIndex]++; |
---|
3821 | |
---|
3822 | data_pos += confparams_cpr->sampleDistance; |
---|
3823 | } |
---|
3824 | //compute the appropriate number |
---|
3825 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
3826 | size_t sum = 0; |
---|
3827 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
3828 | { |
---|
3829 | sum += intervals[i]; |
---|
3830 | if(sum>targetCount) |
---|
3831 | break; |
---|
3832 | } |
---|
3833 | if(i>=confparams_cpr->maxRangeRadius) |
---|
3834 | i = confparams_cpr->maxRangeRadius-1; |
---|
3835 | |
---|
3836 | unsigned int accIntervals = 2*(i+1); |
---|
3837 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
3838 | |
---|
3839 | if(powerOf2<32) |
---|
3840 | powerOf2 = 32; |
---|
3841 | |
---|
3842 | free(intervals); |
---|
3843 | //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2); |
---|
3844 | return powerOf2; |
---|
3845 | } |
---|
3846 | |
---|
3847 | 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){ |
---|
3848 | |
---|
3849 | mean[0] = block_ori_data[0]; |
---|
3850 | unsigned short unpredictable_count = 0; |
---|
3851 | |
---|
3852 | float curData; |
---|
3853 | double itvNum; |
---|
3854 | double diff; |
---|
3855 | float last_over_thres = mean[0]; |
---|
3856 | float pred1D; |
---|
3857 | size_t type_index = 0; |
---|
3858 | float * data_pos = block_ori_data; |
---|
3859 | for(size_t i=0; i<block_dim_0; i++){ |
---|
3860 | curData = *data_pos; |
---|
3861 | |
---|
3862 | pred1D = last_over_thres; |
---|
3863 | diff = curData - pred1D; |
---|
3864 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3865 | if (itvNum < exe_params->intvCapacity){ |
---|
3866 | if (diff < 0) itvNum = -itvNum; |
---|
3867 | type[type_index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3868 | last_over_thres = pred1D + 2 * (type[type_index] - exe_params->intvRadius) * realPrecision; |
---|
3869 | if(fabs(curData-last_over_thres)>realPrecision){ |
---|
3870 | type[type_index] = 0; |
---|
3871 | last_over_thres = curData; |
---|
3872 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3873 | } |
---|
3874 | |
---|
3875 | } |
---|
3876 | else{ |
---|
3877 | type[type_index] = 0; |
---|
3878 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3879 | last_over_thres = curData; |
---|
3880 | } |
---|
3881 | type_index ++; |
---|
3882 | data_pos ++; |
---|
3883 | } |
---|
3884 | return unpredictable_count; |
---|
3885 | |
---|
3886 | } |
---|
3887 | |
---|
3888 | size_t SZ_compress_float_2D_MDQ_RA_block(float * block_ori_data, float * mean, size_t dim_0, size_t dim_1, size_t block_dim_0, size_t block_dim_1, double realPrecision, float * P0, float * P1, int * type, float * unpredictable_data){ |
---|
3889 | |
---|
3890 | size_t dim0_offset = dim_1; |
---|
3891 | mean[0] = block_ori_data[0]; |
---|
3892 | |
---|
3893 | size_t unpredictable_count = 0; |
---|
3894 | size_t r1, r2; |
---|
3895 | r1 = block_dim_0; |
---|
3896 | r2 = block_dim_1; |
---|
3897 | |
---|
3898 | float * cur_data_pos = block_ori_data; |
---|
3899 | float curData; |
---|
3900 | float pred1D, pred2D; |
---|
3901 | double itvNum; |
---|
3902 | double diff; |
---|
3903 | size_t i, j; |
---|
3904 | /* Process Row-0 data 0*/ |
---|
3905 | curData = *cur_data_pos; |
---|
3906 | pred1D = mean[0]; |
---|
3907 | diff = curData - pred1D; |
---|
3908 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3909 | if (itvNum < exe_params->intvCapacity){ |
---|
3910 | if (diff < 0) itvNum = -itvNum; |
---|
3911 | type[0] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3912 | P1[0] = pred1D + 2 * (type[0] - exe_params->intvRadius) * realPrecision; |
---|
3913 | //ganrantee comporession error against the case of machine-epsilon |
---|
3914 | if(fabs(curData-P1[0])>realPrecision){ |
---|
3915 | type[0] = 0; |
---|
3916 | P1[0] = curData; |
---|
3917 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3918 | } |
---|
3919 | } |
---|
3920 | else{ |
---|
3921 | type[0] = 0; |
---|
3922 | P1[0] = curData; |
---|
3923 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3924 | } |
---|
3925 | |
---|
3926 | /* Process Row-0 data 1*/ |
---|
3927 | curData = cur_data_pos[1]; |
---|
3928 | pred1D = P1[0]; |
---|
3929 | diff = curData - pred1D; |
---|
3930 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3931 | if (itvNum < exe_params->intvCapacity){ |
---|
3932 | if (diff < 0) itvNum = -itvNum; |
---|
3933 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3934 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
3935 | //ganrantee comporession error against the case of machine-epsilon |
---|
3936 | if(fabs(curData-P1[1])>realPrecision){ |
---|
3937 | type[1] = 0; |
---|
3938 | P1[1] = curData; |
---|
3939 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3940 | } |
---|
3941 | } |
---|
3942 | else{ |
---|
3943 | type[1] = 0; |
---|
3944 | P1[1] = curData; |
---|
3945 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3946 | } |
---|
3947 | |
---|
3948 | /* Process Row-0 data 2 --> data r2-1 */ |
---|
3949 | for (j = 2; j < r2; j++) |
---|
3950 | { |
---|
3951 | curData = cur_data_pos[j]; |
---|
3952 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
3953 | diff = curData - pred1D; |
---|
3954 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3955 | if (itvNum < exe_params->intvCapacity){ |
---|
3956 | if (diff < 0) itvNum = -itvNum; |
---|
3957 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3958 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
3959 | //ganrantee comporession error against the case of machine-epsilon |
---|
3960 | if(fabs(curData-P1[j])>realPrecision){ |
---|
3961 | type[j] = 0; |
---|
3962 | P1[j] = curData; |
---|
3963 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3964 | } |
---|
3965 | } |
---|
3966 | else{ |
---|
3967 | type[j] = 0; |
---|
3968 | P1[j] = curData; |
---|
3969 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3970 | } |
---|
3971 | } |
---|
3972 | cur_data_pos += dim0_offset; |
---|
3973 | /* Process Row-1 --> Row-r1-1 */ |
---|
3974 | size_t index; |
---|
3975 | for (i = 1; i < r1; i++) |
---|
3976 | { |
---|
3977 | /* Process row-i data 0 */ |
---|
3978 | index = i*r2; |
---|
3979 | curData = *cur_data_pos; |
---|
3980 | pred1D = P1[0]; |
---|
3981 | diff = curData - pred1D; |
---|
3982 | itvNum = fabs(diff)/realPrecision + 1; |
---|
3983 | if (itvNum < exe_params->intvCapacity){ |
---|
3984 | if (diff < 0) itvNum = -itvNum; |
---|
3985 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
3986 | P0[0] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
3987 | //ganrantee comporession error against the case of machine-epsilon |
---|
3988 | if(fabs(curData-P0[0])>realPrecision){ |
---|
3989 | type[index] = 0; |
---|
3990 | P0[0] = curData; |
---|
3991 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3992 | } |
---|
3993 | } |
---|
3994 | else{ |
---|
3995 | type[index] = 0; |
---|
3996 | P0[0] = curData; |
---|
3997 | unpredictable_data[unpredictable_count ++] = curData; |
---|
3998 | } |
---|
3999 | |
---|
4000 | /* Process row-i data 1 --> r2-1*/ |
---|
4001 | for (j = 1; j < r2; j++) |
---|
4002 | { |
---|
4003 | index = i*r2+j; |
---|
4004 | curData = cur_data_pos[j]; |
---|
4005 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
4006 | diff = curData - pred2D; |
---|
4007 | itvNum = fabs(diff)/realPrecision + 1; |
---|
4008 | if (itvNum < exe_params->intvCapacity) |
---|
4009 | { |
---|
4010 | if (diff < 0) itvNum = -itvNum; |
---|
4011 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
4012 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
4013 | |
---|
4014 | //ganrantee comporession error against the case of machine-epsilon |
---|
4015 | if(fabs(curData-P0[j])>realPrecision) |
---|
4016 | { |
---|
4017 | type[index] = 0; |
---|
4018 | P0[j] = curData; |
---|
4019 | unpredictable_data[unpredictable_count ++] = curData; |
---|
4020 | } |
---|
4021 | } |
---|
4022 | else |
---|
4023 | { |
---|
4024 | type[index] = 0; |
---|
4025 | P0[j] = curData; |
---|
4026 | unpredictable_data[unpredictable_count ++] = curData; |
---|
4027 | } |
---|
4028 | } |
---|
4029 | cur_data_pos += dim0_offset; |
---|
4030 | |
---|
4031 | float *Pt; |
---|
4032 | Pt = P1; |
---|
4033 | P1 = P0; |
---|
4034 | P0 = Pt; |
---|
4035 | } |
---|
4036 | return unpredictable_count; |
---|
4037 | } |
---|
4038 | |
---|