1 | /** |
---|
2 | * @file sz_float_pwr.c |
---|
3 | * @author Sheng Di |
---|
4 | * @date Aug, 2016 |
---|
5 | * @brief SZ_Init, Compression and Decompression functions |
---|
6 | * This file contains the compression/decompression functions related to point-wise relative errors |
---|
7 | * (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory. |
---|
8 | * See COPYRIGHT in top-level directory. |
---|
9 | */ |
---|
10 | |
---|
11 | |
---|
12 | #include <stdio.h> |
---|
13 | #include <stdlib.h> |
---|
14 | #include <string.h> |
---|
15 | #include <unistd.h> |
---|
16 | #include <math.h> |
---|
17 | #include "sz.h" |
---|
18 | #include "CompressElement.h" |
---|
19 | #include "DynamicByteArray.h" |
---|
20 | #include "DynamicIntArray.h" |
---|
21 | #include "TightDataPointStorageF.h" |
---|
22 | #include "sz_float.h" |
---|
23 | #include "sz_float_pwr.h" |
---|
24 | #include "zlib.h" |
---|
25 | #include "rw.h" |
---|
26 | #include "utility.h" |
---|
27 | |
---|
28 | void compute_segment_precisions_float_1D(float *oriData, size_t dataLength, float* pwrErrBound, unsigned char* pwrErrBoundBytes, double globalPrecision) |
---|
29 | { |
---|
30 | size_t i = 0, j = 0, k = 0; |
---|
31 | float realPrecision = oriData[0]!=0?fabs(confparams_cpr->pw_relBoundRatio*oriData[0]):confparams_cpr->pw_relBoundRatio; |
---|
32 | float approxPrecision; |
---|
33 | unsigned char realPrecBytes[4]; |
---|
34 | float curPrecision; |
---|
35 | float curValue; |
---|
36 | float sum = 0; |
---|
37 | for(i=0;i<dataLength;i++) |
---|
38 | { |
---|
39 | curValue = oriData[i]; |
---|
40 | if(i%confparams_cpr->segment_size==0&&i>0) |
---|
41 | { |
---|
42 | //get two first bytes of the realPrecision |
---|
43 | if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE) |
---|
44 | { |
---|
45 | realPrecision = sum/confparams_cpr->segment_size; |
---|
46 | sum = 0; |
---|
47 | } |
---|
48 | realPrecision *= confparams_cpr->pw_relBoundRatio; |
---|
49 | |
---|
50 | if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL) |
---|
51 | realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; |
---|
52 | else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL) |
---|
53 | realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision; |
---|
54 | |
---|
55 | floatToBytes(realPrecBytes, realPrecision); |
---|
56 | realPrecBytes[2] = realPrecBytes[3] = 0; |
---|
57 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
58 | //put the realPrecision in float* pwrErBound |
---|
59 | pwrErrBound[j++] = approxPrecision; |
---|
60 | //put the two bytes in pwrErrBoundBytes |
---|
61 | pwrErrBoundBytes[k++] = realPrecBytes[0]; |
---|
62 | pwrErrBoundBytes[k++] = realPrecBytes[1]; |
---|
63 | |
---|
64 | realPrecision = fabs(curValue); |
---|
65 | } |
---|
66 | |
---|
67 | if(curValue!=0) |
---|
68 | { |
---|
69 | curPrecision = fabs(curValue); |
---|
70 | |
---|
71 | switch(confparams_cpr->pwr_type) |
---|
72 | { |
---|
73 | case SZ_PWR_MIN_TYPE: |
---|
74 | if(realPrecision>curPrecision) |
---|
75 | realPrecision = curPrecision; |
---|
76 | break; |
---|
77 | case SZ_PWR_AVG_TYPE: |
---|
78 | sum += curPrecision; |
---|
79 | break; |
---|
80 | case SZ_PWR_MAX_TYPE: |
---|
81 | if(realPrecision<curPrecision) |
---|
82 | realPrecision = curPrecision; |
---|
83 | break; |
---|
84 | } |
---|
85 | } |
---|
86 | } |
---|
87 | if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE) |
---|
88 | { |
---|
89 | int size = dataLength%confparams_cpr->segment_size==0?confparams_cpr->segment_size:dataLength%confparams_cpr->segment_size; |
---|
90 | realPrecision = sum/size; |
---|
91 | } |
---|
92 | if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL) |
---|
93 | realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; |
---|
94 | else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL) |
---|
95 | realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision; |
---|
96 | floatToBytes(realPrecBytes, realPrecision); |
---|
97 | realPrecBytes[2] = realPrecBytes[3] = 0; |
---|
98 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
99 | //put the realPrecision in float* pwrErBound |
---|
100 | pwrErrBound[j++] = approxPrecision; |
---|
101 | //put the two bytes in pwrErrBoundBytes |
---|
102 | pwrErrBoundBytes[k++] = realPrecBytes[0]; |
---|
103 | pwrErrBoundBytes[k++] = realPrecBytes[1]; |
---|
104 | } |
---|
105 | |
---|
106 | unsigned int optimize_intervals_float_1D_pwr(float *oriData, size_t dataLength, float* pwrErrBound) |
---|
107 | { |
---|
108 | size_t i = 0, j = 0; |
---|
109 | float realPrecision = pwrErrBound[j++]; |
---|
110 | unsigned long radiusIndex; |
---|
111 | float pred_value = 0, pred_err; |
---|
112 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
113 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
114 | int totalSampleSize = dataLength/confparams_cpr->sampleDistance; |
---|
115 | for(i=2;i<dataLength;i++) |
---|
116 | { |
---|
117 | if(i%confparams_cpr->segment_size==0) |
---|
118 | realPrecision = pwrErrBound[j++]; |
---|
119 | if(i%confparams_cpr->sampleDistance==0) |
---|
120 | { |
---|
121 | //pred_value = 2*oriData[i-1] - oriData[i-2]; |
---|
122 | pred_value = oriData[i-1]; |
---|
123 | pred_err = fabs(pred_value - oriData[i]); |
---|
124 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
125 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
126 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
127 | intervals[radiusIndex]++; |
---|
128 | } |
---|
129 | } |
---|
130 | //compute the appropriate number |
---|
131 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
132 | size_t sum = 0; |
---|
133 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
134 | { |
---|
135 | sum += intervals[i]; |
---|
136 | if(sum>targetCount) |
---|
137 | break; |
---|
138 | } |
---|
139 | if(i>=confparams_cpr->maxRangeRadius) |
---|
140 | i = confparams_cpr->maxRangeRadius-1; |
---|
141 | |
---|
142 | unsigned int accIntervals = 2*(i+1); |
---|
143 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
144 | |
---|
145 | if(powerOf2<32) |
---|
146 | powerOf2 = 32; |
---|
147 | |
---|
148 | free(intervals); |
---|
149 | //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2); |
---|
150 | return powerOf2; |
---|
151 | } |
---|
152 | |
---|
153 | void compute_segment_precisions_float_2D(float *oriData, float* pwrErrBound, |
---|
154 | size_t r1, size_t r2, size_t R2, size_t edgeSize, unsigned char* pwrErrBoundBytes, float Min, float Max, double globalPrecision) |
---|
155 | { |
---|
156 | size_t i = 0, j = 0, k = 0, p = 0, index = 0, J = 0; //I=-1,J=-1 if they are needed |
---|
157 | float realPrecision; |
---|
158 | float approxPrecision; |
---|
159 | unsigned char realPrecBytes[4]; |
---|
160 | float curValue, curAbsValue; |
---|
161 | float* statAbsValues = (float*)malloc(R2*sizeof(float)); |
---|
162 | |
---|
163 | float max = fabs(Min)<fabs(Max)?fabs(Max):fabs(Min); //get the max abs value. |
---|
164 | float min = fabs(Min)<fabs(Max)?fabs(Min):fabs(Max); |
---|
165 | for(i=0;i<R2;i++) |
---|
166 | { |
---|
167 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
168 | statAbsValues[i] = max; |
---|
169 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
170 | statAbsValues[i] = min; |
---|
171 | else |
---|
172 | statAbsValues[i] = 0; //for SZ_PWR_AVG_TYPE |
---|
173 | } |
---|
174 | for(i=0;i<r1;i++) |
---|
175 | { |
---|
176 | for(j=0;j<r2;j++) |
---|
177 | { |
---|
178 | index = i*r2+j; |
---|
179 | curValue = oriData[index]; |
---|
180 | if(((i%edgeSize==edgeSize-1 || i==r1-1) &&j%edgeSize==0&&j>0) || (i%edgeSize==0&&j==0&&i>0)) |
---|
181 | { |
---|
182 | if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE) |
---|
183 | { |
---|
184 | int a = edgeSize, b = edgeSize; |
---|
185 | if(j==0) |
---|
186 | { |
---|
187 | if(r2%edgeSize==0) |
---|
188 | b = edgeSize; |
---|
189 | else |
---|
190 | b = r2%edgeSize; |
---|
191 | } |
---|
192 | if(i==r1-1) |
---|
193 | { |
---|
194 | if(r1%edgeSize==0) |
---|
195 | a = edgeSize; |
---|
196 | else |
---|
197 | a = r1%edgeSize; |
---|
198 | } |
---|
199 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J]/(a*b); |
---|
200 | } |
---|
201 | else |
---|
202 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J]; |
---|
203 | |
---|
204 | if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL) |
---|
205 | realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; |
---|
206 | else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL) |
---|
207 | realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision; |
---|
208 | |
---|
209 | floatToBytes(realPrecBytes, realPrecision); |
---|
210 | realPrecBytes[2] = realPrecBytes[3] = 0; |
---|
211 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
212 | //put the realPrecision in float* pwrErBound |
---|
213 | pwrErrBound[p++] = approxPrecision; |
---|
214 | //put the two bytes in pwrErrBoundBytes |
---|
215 | pwrErrBoundBytes[k++] = realPrecBytes[0]; |
---|
216 | pwrErrBoundBytes[k++] = realPrecBytes[1]; |
---|
217 | |
---|
218 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
219 | statAbsValues[J] = max; |
---|
220 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
221 | statAbsValues[J] = min; |
---|
222 | else |
---|
223 | statAbsValues[J] = 0; //for SZ_PWR_AVG_TYPE |
---|
224 | } |
---|
225 | if(j==0) |
---|
226 | J = 0; |
---|
227 | else if(j%edgeSize==0) |
---|
228 | J++; |
---|
229 | if(curValue!=0) |
---|
230 | { |
---|
231 | curAbsValue = fabs(curValue); |
---|
232 | |
---|
233 | switch(confparams_cpr->pwr_type) |
---|
234 | { |
---|
235 | case SZ_PWR_MIN_TYPE: |
---|
236 | if(statAbsValues[J]>curAbsValue) |
---|
237 | statAbsValues[J] = curAbsValue; |
---|
238 | break; |
---|
239 | case SZ_PWR_AVG_TYPE: |
---|
240 | statAbsValues[J] += curAbsValue; |
---|
241 | break; |
---|
242 | case SZ_PWR_MAX_TYPE: |
---|
243 | if(statAbsValues[J]<curAbsValue) |
---|
244 | statAbsValues[J] = curAbsValue; |
---|
245 | break; |
---|
246 | } |
---|
247 | } |
---|
248 | } |
---|
249 | } |
---|
250 | |
---|
251 | if(confparams_cpr->pwr_type==SZ_PWR_AVG_TYPE) |
---|
252 | { |
---|
253 | int a = edgeSize, b = edgeSize; |
---|
254 | if(r2%edgeSize==0) |
---|
255 | b = edgeSize; |
---|
256 | else |
---|
257 | b = r2%edgeSize; |
---|
258 | if(r1%edgeSize==0) |
---|
259 | a = edgeSize; |
---|
260 | else |
---|
261 | a = r1%edgeSize; |
---|
262 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J]/(a*b); |
---|
263 | } |
---|
264 | else |
---|
265 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J]; |
---|
266 | |
---|
267 | if(confparams_cpr->errorBoundMode==ABS_AND_PW_REL||confparams_cpr->errorBoundMode==REL_AND_PW_REL) |
---|
268 | realPrecision = realPrecision<globalPrecision?realPrecision:globalPrecision; |
---|
269 | else if(confparams_cpr->errorBoundMode==ABS_OR_PW_REL||confparams_cpr->errorBoundMode==REL_OR_PW_REL) |
---|
270 | realPrecision = realPrecision<globalPrecision?globalPrecision:realPrecision; |
---|
271 | |
---|
272 | floatToBytes(realPrecBytes, realPrecision); |
---|
273 | realPrecBytes[2] = realPrecBytes[3] = 0; |
---|
274 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
275 | //put the realPrecision in float* pwrErBound |
---|
276 | pwrErrBound[p++] = approxPrecision; |
---|
277 | //put the two bytes in pwrErrBoundBytes |
---|
278 | pwrErrBoundBytes[k++] = realPrecBytes[0]; |
---|
279 | pwrErrBoundBytes[k++] = realPrecBytes[1]; |
---|
280 | |
---|
281 | free(statAbsValues); |
---|
282 | } |
---|
283 | |
---|
284 | unsigned int optimize_intervals_float_2D_pwr(float *oriData, size_t r1, size_t r2, size_t R2, size_t edgeSize, float* pwrErrBound) |
---|
285 | { |
---|
286 | size_t i = 0,j = 0, index, I=0, J=0; |
---|
287 | float realPrecision = pwrErrBound[0]; |
---|
288 | unsigned long radiusIndex; |
---|
289 | float pred_value = 0, pred_err; |
---|
290 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
291 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
292 | size_t totalSampleSize = (r1-1)*(r2-1)/confparams_cpr->sampleDistance; |
---|
293 | size_t ir2; |
---|
294 | for(i=1;i<r1;i++) |
---|
295 | { |
---|
296 | ir2 = i*r2; |
---|
297 | if(i%edgeSize==0) |
---|
298 | { |
---|
299 | I++; |
---|
300 | J = 0; |
---|
301 | } |
---|
302 | for(j=1;j<r2;j++) |
---|
303 | { |
---|
304 | index = ir2+j; |
---|
305 | if(j%edgeSize==0) |
---|
306 | J++; |
---|
307 | |
---|
308 | if((i+j)%confparams_cpr->sampleDistance==0) |
---|
309 | { |
---|
310 | realPrecision = pwrErrBound[I*R2+J]; |
---|
311 | pred_value = oriData[index-1] + oriData[index-r2] - oriData[index-r2-1]; |
---|
312 | pred_err = fabs(pred_value - oriData[index]); |
---|
313 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
314 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
315 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
316 | intervals[radiusIndex]++; |
---|
317 | } |
---|
318 | } |
---|
319 | } |
---|
320 | //compute the appropriate number |
---|
321 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
322 | size_t sum = 0; |
---|
323 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
324 | { |
---|
325 | sum += intervals[i]; |
---|
326 | if(sum>targetCount) |
---|
327 | break; |
---|
328 | } |
---|
329 | if(i>=confparams_cpr->maxRangeRadius) |
---|
330 | i = confparams_cpr->maxRangeRadius-1; |
---|
331 | unsigned int accIntervals = 2*(i+1); |
---|
332 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
333 | |
---|
334 | if(powerOf2<32) |
---|
335 | powerOf2 = 32; |
---|
336 | |
---|
337 | free(intervals); |
---|
338 | //printf("confparams_cpr->maxRangeRadius = %d, accIntervals=%d, powerOf2=%d\n", confparams_cpr->maxRangeRadius, accIntervals, powerOf2); |
---|
339 | return powerOf2; |
---|
340 | } |
---|
341 | |
---|
342 | void compute_segment_precisions_float_3D(float *oriData, float* pwrErrBound, |
---|
343 | size_t r1, size_t r2, size_t r3, size_t R2, size_t R3, size_t edgeSize, unsigned char* pwrErrBoundBytes, float Min, float Max, double globalPrecision) |
---|
344 | { |
---|
345 | size_t i = 0, j = 0, k = 0, p = 0, q = 0, index = 0, J = 0, K = 0; //I=-1,J=-1 if they are needed |
---|
346 | size_t r23 = r2*r3, ir, jr; |
---|
347 | float realPrecision; |
---|
348 | float approxPrecision; |
---|
349 | unsigned char realPrecBytes[4]; |
---|
350 | float curValue, curAbsValue; |
---|
351 | |
---|
352 | float** statAbsValues = create2DArray_float(R2, R3); |
---|
353 | float max = fabs(Min)<fabs(Max)?fabs(Max):fabs(Min); //get the max abs value. |
---|
354 | float min = fabs(Min)<fabs(Max)?fabs(Min):fabs(Max); |
---|
355 | |
---|
356 | for(i=0;i<R2;i++) |
---|
357 | for(j=0;j<R3;j++) |
---|
358 | { |
---|
359 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
360 | statAbsValues[i][j] = max; |
---|
361 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
362 | statAbsValues[i][j] = min; |
---|
363 | else |
---|
364 | statAbsValues[i][j] = 0; |
---|
365 | } |
---|
366 | for(i=0;i<r1;i++) |
---|
367 | { |
---|
368 | ir = i*r23; |
---|
369 | if(i%edgeSize==0&&i>0) |
---|
370 | { |
---|
371 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J][K]; |
---|
372 | floatToBytes(realPrecBytes, realPrecision); |
---|
373 | memset(&realPrecBytes[2], 0, 2); |
---|
374 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
375 | //put the realPrecision in float* pwrErBound |
---|
376 | pwrErrBound[p++] = approxPrecision; |
---|
377 | //put the two bytes in pwrErrBoundBytes |
---|
378 | //printf("q=%d, i=%d, j=%d, k=%d\n",q,i,j,k); |
---|
379 | pwrErrBoundBytes[q++] = realPrecBytes[0]; |
---|
380 | pwrErrBoundBytes[q++] = realPrecBytes[1]; |
---|
381 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
382 | statAbsValues[J][K] = max; |
---|
383 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
384 | statAbsValues[J][K] = min; |
---|
385 | |
---|
386 | } |
---|
387 | for(j=0;j<r2;j++) |
---|
388 | { |
---|
389 | jr = j*r3; |
---|
390 | if((i%edgeSize==edgeSize-1 || i == r1-1)&&j%edgeSize==0&&j>0) |
---|
391 | { |
---|
392 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J][K]; |
---|
393 | floatToBytes(realPrecBytes, realPrecision); |
---|
394 | memset(&realPrecBytes[2], 0, 2); |
---|
395 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
396 | //put the realPrecision in float* pwrErBound |
---|
397 | pwrErrBound[p++] = approxPrecision; |
---|
398 | //put the two bytes in pwrErrBoundBytes |
---|
399 | //printf("q=%d, i=%d, j=%d, k=%d\n",q,i,j,k); |
---|
400 | pwrErrBoundBytes[q++] = realPrecBytes[0]; |
---|
401 | pwrErrBoundBytes[q++] = realPrecBytes[1]; |
---|
402 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
403 | statAbsValues[J][K] = max; |
---|
404 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
405 | statAbsValues[J][K] = min; |
---|
406 | } |
---|
407 | |
---|
408 | if(j==0) |
---|
409 | J = 0; |
---|
410 | else if(j%edgeSize==0) |
---|
411 | J++; |
---|
412 | |
---|
413 | for(k=0;k<r3;k++) |
---|
414 | { |
---|
415 | index = ir+jr+k; |
---|
416 | curValue = oriData[index]; |
---|
417 | if((i%edgeSize==edgeSize-1 || i == r1-1)&&(j%edgeSize==edgeSize-1||j==r2-1)&&k%edgeSize==0&&k>0) |
---|
418 | { |
---|
419 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J][K]; |
---|
420 | floatToBytes(realPrecBytes, realPrecision); |
---|
421 | memset(&realPrecBytes[2], 0, 2); |
---|
422 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
423 | //put the realPrecision in float* pwrErBound |
---|
424 | pwrErrBound[p++] = approxPrecision; |
---|
425 | //put the two bytes in pwrErrBoundBytes |
---|
426 | //printf("q=%d, i=%d, j=%d, k=%d\n",q,i,j,k); |
---|
427 | pwrErrBoundBytes[q++] = realPrecBytes[0]; |
---|
428 | pwrErrBoundBytes[q++] = realPrecBytes[1]; |
---|
429 | |
---|
430 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
431 | statAbsValues[J][K] = max; |
---|
432 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
433 | statAbsValues[J][K] = min; |
---|
434 | } |
---|
435 | |
---|
436 | if(k==0) |
---|
437 | K = 0; |
---|
438 | else if(k%edgeSize==0) |
---|
439 | K++; |
---|
440 | |
---|
441 | if(curValue!=0) |
---|
442 | { |
---|
443 | curAbsValue = fabs(curValue); |
---|
444 | if(confparams_cpr->pwr_type == SZ_PWR_MIN_TYPE) |
---|
445 | { |
---|
446 | if(statAbsValues[J][K]>curAbsValue) |
---|
447 | { |
---|
448 | statAbsValues[J][K] = curAbsValue; |
---|
449 | } |
---|
450 | } |
---|
451 | else if(confparams_cpr->pwr_type == SZ_PWR_MAX_TYPE) |
---|
452 | { |
---|
453 | if(statAbsValues[J][K]<curAbsValue) |
---|
454 | { |
---|
455 | statAbsValues[J][K] = curAbsValue; |
---|
456 | } |
---|
457 | } |
---|
458 | } |
---|
459 | } |
---|
460 | } |
---|
461 | } |
---|
462 | |
---|
463 | realPrecision = confparams_cpr->pw_relBoundRatio*statAbsValues[J][K]; |
---|
464 | floatToBytes(realPrecBytes, realPrecision); |
---|
465 | realPrecBytes[2] = realPrecBytes[3] = 0; |
---|
466 | approxPrecision = bytesToFloat(realPrecBytes); |
---|
467 | //put the realPrecision in float* pwrErBound |
---|
468 | pwrErrBound[p++] = approxPrecision; |
---|
469 | //put the two bytes in pwrErrBoundBytes |
---|
470 | pwrErrBoundBytes[q++] = realPrecBytes[0]; |
---|
471 | pwrErrBoundBytes[q++] = realPrecBytes[1]; |
---|
472 | |
---|
473 | free2DArray_float(statAbsValues, R2); |
---|
474 | } |
---|
475 | |
---|
476 | unsigned int optimize_intervals_float_3D_pwr(float *oriData, size_t r1, size_t r2, size_t r3, size_t R2, size_t R3, size_t edgeSize, float* pwrErrBound) |
---|
477 | { |
---|
478 | size_t i,j,k, ir,jr,index, I = 0,J=0,K=0; |
---|
479 | float realPrecision = pwrErrBound[0]; |
---|
480 | unsigned long radiusIndex; |
---|
481 | size_t r23=r2*r3; |
---|
482 | size_t R23 = R2*R3; |
---|
483 | float pred_value = 0, pred_err; |
---|
484 | int *intervals = (int*)malloc(confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
485 | memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(int)); |
---|
486 | size_t totalSampleSize = (r1-1)*(r2-1)*(r3-1)/confparams_cpr->sampleDistance; |
---|
487 | for(i=1;i<r1;i++) |
---|
488 | { |
---|
489 | ir = i*r23; |
---|
490 | if(i%edgeSize==0) |
---|
491 | { |
---|
492 | I++; |
---|
493 | J = 0; |
---|
494 | } |
---|
495 | for(j=1;j<r2;j++) |
---|
496 | { |
---|
497 | jr = j*r3; |
---|
498 | if(j%edgeSize==0) |
---|
499 | { |
---|
500 | J++; |
---|
501 | K = 0; |
---|
502 | } |
---|
503 | for(k=1;k<r3;k++) |
---|
504 | { |
---|
505 | index = ir+jr+k; |
---|
506 | if(k%edgeSize==0) |
---|
507 | K++; |
---|
508 | if((i+j+k)%confparams_cpr->sampleDistance==0) |
---|
509 | { |
---|
510 | realPrecision = pwrErrBound[I*R23+J*R2+K]; |
---|
511 | pred_value = oriData[index-1] + oriData[index-r3] + oriData[index-r23] |
---|
512 | - oriData[index-1-r23] - oriData[index-r3-1] - oriData[index-r3-r23] + oriData[index-r3-r23-1]; |
---|
513 | pred_err = fabs(pred_value - oriData[index]); |
---|
514 | radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2); |
---|
515 | if(radiusIndex>=confparams_cpr->maxRangeRadius) |
---|
516 | radiusIndex = confparams_cpr->maxRangeRadius - 1; |
---|
517 | intervals[radiusIndex]++; |
---|
518 | } |
---|
519 | } |
---|
520 | } |
---|
521 | } |
---|
522 | //compute the appropriate number |
---|
523 | size_t targetCount = totalSampleSize*confparams_cpr->predThreshold; |
---|
524 | size_t sum = 0; |
---|
525 | for(i=0;i<confparams_cpr->maxRangeRadius;i++) |
---|
526 | { |
---|
527 | sum += intervals[i]; |
---|
528 | if(sum>targetCount) |
---|
529 | break; |
---|
530 | } |
---|
531 | if(i>=confparams_cpr->maxRangeRadius) |
---|
532 | i = confparams_cpr->maxRangeRadius-1; |
---|
533 | unsigned int accIntervals = 2*(i+1); |
---|
534 | unsigned int powerOf2 = roundUpToPowerOf2(accIntervals); |
---|
535 | |
---|
536 | if(powerOf2<32) |
---|
537 | powerOf2 = 32; |
---|
538 | |
---|
539 | free(intervals); |
---|
540 | //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2); |
---|
541 | return powerOf2; |
---|
542 | } |
---|
543 | |
---|
544 | void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(unsigned char** newByteData, float *oriData, double globalPrecision, |
---|
545 | size_t dataLength, size_t *outSize, float min, float max) |
---|
546 | { |
---|
547 | size_t pwrLength = dataLength%confparams_cpr->segment_size==0?dataLength/confparams_cpr->segment_size:dataLength/confparams_cpr->segment_size+1; |
---|
548 | float* pwrErrBound = (float*)malloc(sizeof(float)*pwrLength); |
---|
549 | size_t pwrErrBoundBytes_size = sizeof(unsigned char)*pwrLength*2; |
---|
550 | unsigned char* pwrErrBoundBytes = (unsigned char*)malloc(pwrErrBoundBytes_size); |
---|
551 | |
---|
552 | compute_segment_precisions_float_1D(oriData, dataLength, pwrErrBound, pwrErrBoundBytes, globalPrecision); |
---|
553 | |
---|
554 | unsigned int quantization_intervals; |
---|
555 | if(exe_params->optQuantMode==1) |
---|
556 | { |
---|
557 | quantization_intervals = optimize_intervals_float_1D_pwr(oriData, dataLength, pwrErrBound); |
---|
558 | updateQuantizationInfo(quantization_intervals); |
---|
559 | } |
---|
560 | else |
---|
561 | quantization_intervals = exe_params->intvCapacity; |
---|
562 | size_t i = 0, j = 0; |
---|
563 | int reqLength; |
---|
564 | float realPrecision = pwrErrBound[j++]; |
---|
565 | float medianValue = 0; |
---|
566 | float radius = fabs(max)<fabs(min)?fabs(min):fabs(max); |
---|
567 | short radExpo = getExponent_float(radius); |
---|
568 | |
---|
569 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
570 | |
---|
571 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
572 | //type[dataLength]=0; |
---|
573 | |
---|
574 | float* spaceFillingValue = oriData; // |
---|
575 | |
---|
576 | DynamicByteArray *resiBitLengthArray; |
---|
577 | new_DBA(&resiBitLengthArray, DynArrayInitLen); |
---|
578 | |
---|
579 | DynamicIntArray *exactLeadNumArray; |
---|
580 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
581 | |
---|
582 | DynamicByteArray *exactMidByteArray; |
---|
583 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
584 | |
---|
585 | DynamicIntArray *resiBitArray; |
---|
586 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
587 | |
---|
588 | type[0] = 0; |
---|
589 | |
---|
590 | unsigned char preDataBytes[4] = {0}; |
---|
591 | intToBytes_bigEndian(preDataBytes, 0); |
---|
592 | |
---|
593 | int reqBytesLength = reqLength/8; |
---|
594 | int resiBitsLength = reqLength%8; |
---|
595 | float last3CmprsData[3] = {0}; |
---|
596 | |
---|
597 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
598 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
599 | |
---|
600 | //add the first data |
---|
601 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
602 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
603 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
604 | memcpy(preDataBytes,vce->curBytes,4); |
---|
605 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
606 | listAdd_float(last3CmprsData, vce->data); |
---|
607 | //printf("%.30G\n",last3CmprsData[0]); |
---|
608 | |
---|
609 | //add the second data |
---|
610 | type[1] = 0; |
---|
611 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
612 | compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
613 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
614 | memcpy(preDataBytes,vce->curBytes,4); |
---|
615 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
616 | listAdd_float(last3CmprsData, vce->data); |
---|
617 | //printf("%.30G\n",last3CmprsData[0]); |
---|
618 | |
---|
619 | int state; |
---|
620 | double checkRadius; |
---|
621 | float curData; |
---|
622 | float pred; |
---|
623 | double predAbsErr; |
---|
624 | checkRadius = (exe_params->intvCapacity-1)*realPrecision; |
---|
625 | double interval = 2*realPrecision; |
---|
626 | int updateReqLength = 0; //a marker: 1 means already updated |
---|
627 | |
---|
628 | for(i=2;i<dataLength;i++) |
---|
629 | { |
---|
630 | curData = spaceFillingValue[i]; |
---|
631 | if(i%confparams_cpr->segment_size==0) |
---|
632 | { |
---|
633 | realPrecision = pwrErrBound[j++]; |
---|
634 | checkRadius = (exe_params->intvCapacity-1)*realPrecision; |
---|
635 | interval = 2*realPrecision; |
---|
636 | updateReqLength = 0; |
---|
637 | } |
---|
638 | //pred = 2*last3CmprsData[0] - last3CmprsData[1]; |
---|
639 | pred = last3CmprsData[0]; |
---|
640 | predAbsErr = fabs(curData - pred); |
---|
641 | if(predAbsErr<checkRadius) |
---|
642 | { |
---|
643 | state = (predAbsErr/realPrecision+1)/2; |
---|
644 | if(curData>=pred) |
---|
645 | { |
---|
646 | type[i] = exe_params->intvRadius+state; |
---|
647 | pred = pred + state*interval; |
---|
648 | } |
---|
649 | else //curData<pred |
---|
650 | { |
---|
651 | type[i] = exe_params->intvRadius-state; |
---|
652 | pred = pred - state*interval; |
---|
653 | } |
---|
654 | listAdd_float(last3CmprsData, pred); |
---|
655 | continue; |
---|
656 | } |
---|
657 | |
---|
658 | //unpredictable data processing |
---|
659 | if(updateReqLength==0) |
---|
660 | { |
---|
661 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
662 | reqBytesLength = reqLength/8; |
---|
663 | resiBitsLength = reqLength%8; |
---|
664 | updateReqLength = 1; |
---|
665 | } |
---|
666 | |
---|
667 | type[i] = 0; |
---|
668 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
669 | |
---|
670 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
671 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
672 | memcpy(preDataBytes,vce->curBytes,4); |
---|
673 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
674 | |
---|
675 | listAdd_float(last3CmprsData, vce->data); |
---|
676 | }//end of for |
---|
677 | |
---|
678 | // char* expSegmentsInBytes; |
---|
679 | // int expSegmentsInBytes_size = convertESCToBytes(esc, &expSegmentsInBytes); |
---|
680 | int exactDataNum = exactLeadNumArray->size; |
---|
681 | |
---|
682 | TightDataPointStorageF* tdps; |
---|
683 | |
---|
684 | new_TightDataPointStorageF2(&tdps, dataLength, exactDataNum, |
---|
685 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
686 | exactLeadNumArray->array, |
---|
687 | resiBitArray->array, resiBitArray->size, |
---|
688 | resiBitLengthArray->array, resiBitLengthArray->size, |
---|
689 | realPrecision, medianValue, (char)reqLength, quantization_intervals, pwrErrBoundBytes, pwrErrBoundBytes_size, radExpo); |
---|
690 | |
---|
691 | //sdi:Debug |
---|
692 | /* int sum =0; |
---|
693 | for(i=0;i<dataLength;i++) |
---|
694 | if(type[i]==0) sum++; |
---|
695 | printf("opt_quantizations=%d, exactDataNum=%d, sum=%d\n",quantization_intervals, exactDataNum, sum); |
---|
696 | */ |
---|
697 | // writeUShortData(type, dataLength, "compressStateBytes.sb"); |
---|
698 | // unsigned short type_[dataLength]; |
---|
699 | // SZ_Reset(); |
---|
700 | // decode_withTree(tdps->typeArray, tdps->typeArray_size, type_); |
---|
701 | // printf("tdps->typeArray_size=%d\n", tdps->typeArray_size); |
---|
702 | |
---|
703 | //free memory |
---|
704 | free_DBA(resiBitLengthArray); |
---|
705 | free_DIA(exactLeadNumArray); |
---|
706 | free_DIA(resiBitArray); |
---|
707 | free(type); |
---|
708 | |
---|
709 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
710 | |
---|
711 | int floatSize=sizeof(float); |
---|
712 | if(*outSize>dataLength*floatSize) |
---|
713 | { |
---|
714 | size_t k = 0, i; |
---|
715 | tdps->isLossless = 1; |
---|
716 | size_t totalByteLength = 3 + exe_params->SZ_SIZE_TYPE + 1 + floatSize*dataLength; |
---|
717 | *newByteData = (unsigned char*)malloc(totalByteLength); |
---|
718 | |
---|
719 | unsigned char dsLengthBytes[exe_params->SZ_SIZE_TYPE]; |
---|
720 | intToBytes_bigEndian(dsLengthBytes, dataLength);//4 |
---|
721 | for (i = 0; i < 3; i++)//3 |
---|
722 | (*newByteData)[k++] = versionNumber[i]; |
---|
723 | |
---|
724 | if(exe_params->SZ_SIZE_TYPE==4) |
---|
725 | { |
---|
726 | (*newByteData)[k++] = 16; //=00010000 |
---|
727 | } |
---|
728 | else |
---|
729 | { |
---|
730 | (*newByteData)[k++] = 80; |
---|
731 | } |
---|
732 | for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++)//4 or 8 |
---|
733 | (*newByteData)[k++] = dsLengthBytes[i]; |
---|
734 | |
---|
735 | |
---|
736 | if(sysEndianType==BIG_ENDIAN_SYSTEM) |
---|
737 | memcpy((*newByteData)+4+exe_params->SZ_SIZE_TYPE, oriData, dataLength*floatSize); |
---|
738 | else |
---|
739 | { |
---|
740 | unsigned char* p = (*newByteData)+4+exe_params->SZ_SIZE_TYPE; |
---|
741 | for(i=0;i<dataLength;i++,p+=floatSize) |
---|
742 | floatToBytes(p, oriData[i]); |
---|
743 | } |
---|
744 | *outSize = totalByteLength; |
---|
745 | } |
---|
746 | |
---|
747 | free(pwrErrBound); |
---|
748 | |
---|
749 | free(vce); |
---|
750 | free(lce); |
---|
751 | free_TightDataPointStorageF(tdps); |
---|
752 | free(exactMidByteArray); |
---|
753 | } |
---|
754 | |
---|
755 | void SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr(unsigned char** newByteData, float *oriData, double globalPrecision, size_t r1, size_t r2, |
---|
756 | size_t *outSize, float min, float max) |
---|
757 | { |
---|
758 | size_t dataLength=r1*r2; |
---|
759 | int blockEdgeSize = computeBlockEdgeSize_2D(confparams_cpr->segment_size); |
---|
760 | size_t R1 = 1+(r1-1)/blockEdgeSize; |
---|
761 | size_t R2 = 1+(r2-1)/blockEdgeSize; |
---|
762 | float* pwrErrBound = (float*)malloc(sizeof(float)*R1*R2); |
---|
763 | size_t pwrErrBoundBytes_size = sizeof(unsigned char)*R1*R2*2; |
---|
764 | unsigned char* pwrErrBoundBytes = (unsigned char*)malloc(pwrErrBoundBytes_size); |
---|
765 | |
---|
766 | compute_segment_precisions_float_2D(oriData, pwrErrBound, r1, r2, R2, blockEdgeSize, pwrErrBoundBytes, min, max, globalPrecision); |
---|
767 | |
---|
768 | unsigned int quantization_intervals; |
---|
769 | if(exe_params->optQuantMode==1) |
---|
770 | { |
---|
771 | quantization_intervals = optimize_intervals_float_2D_pwr(oriData, r1, r2, R2, blockEdgeSize, pwrErrBound); |
---|
772 | updateQuantizationInfo(quantization_intervals); |
---|
773 | } |
---|
774 | else |
---|
775 | quantization_intervals = exe_params->intvCapacity; |
---|
776 | //printf("quantization_intervals=%d\n",quantization_intervals); |
---|
777 | |
---|
778 | size_t i=0,j=0,I=0,J=0; |
---|
779 | int reqLength; |
---|
780 | float realPrecision = pwrErrBound[I*R2+J]; |
---|
781 | float pred1D, pred2D; |
---|
782 | float diff = 0.0; |
---|
783 | double itvNum = 0; |
---|
784 | float *P0, *P1; |
---|
785 | |
---|
786 | P0 = (float*)malloc(r2*sizeof(float)); |
---|
787 | memset(P0, 0, r2*sizeof(float)); |
---|
788 | P1 = (float*)malloc(r2*sizeof(float)); |
---|
789 | memset(P1, 0, r2*sizeof(float)); |
---|
790 | |
---|
791 | float medianValue = 0; |
---|
792 | float radius = fabs(max)<fabs(min)?fabs(min):fabs(max); |
---|
793 | short radExpo = getExponent_float(radius); |
---|
794 | int updateReqLength = 1; |
---|
795 | |
---|
796 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
797 | |
---|
798 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
799 | //type[dataLength]=0; |
---|
800 | |
---|
801 | float* spaceFillingValue = oriData; // |
---|
802 | |
---|
803 | DynamicByteArray *resiBitLengthArray; |
---|
804 | new_DBA(&resiBitLengthArray, DynArrayInitLen); |
---|
805 | |
---|
806 | DynamicIntArray *exactLeadNumArray; |
---|
807 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
808 | |
---|
809 | DynamicByteArray *exactMidByteArray; |
---|
810 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
811 | |
---|
812 | DynamicIntArray *resiBitArray; |
---|
813 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
814 | |
---|
815 | type[0] = 0; |
---|
816 | |
---|
817 | unsigned char preDataBytes[4]; |
---|
818 | intToBytes_bigEndian(preDataBytes, 0); |
---|
819 | |
---|
820 | int reqBytesLength = reqLength/8; |
---|
821 | int resiBitsLength = reqLength%8; |
---|
822 | |
---|
823 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
824 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
825 | |
---|
826 | /* Process Row-0 data 0*/ |
---|
827 | type[0] = 0; |
---|
828 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
829 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
830 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
831 | memcpy(preDataBytes,vce->curBytes,4); |
---|
832 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
833 | P1[0] = vce->data; |
---|
834 | |
---|
835 | /* Process Row-0 data 1*/ |
---|
836 | pred1D = P1[0]; |
---|
837 | diff = spaceFillingValue[1] - pred1D; |
---|
838 | |
---|
839 | itvNum = fabs(diff)/realPrecision + 1; |
---|
840 | |
---|
841 | if (itvNum < exe_params->intvCapacity) |
---|
842 | { |
---|
843 | if (diff < 0) itvNum = -itvNum; |
---|
844 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
845 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
846 | } |
---|
847 | else |
---|
848 | { |
---|
849 | type[1] = 0; |
---|
850 | |
---|
851 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
852 | compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
853 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
854 | memcpy(preDataBytes,vce->curBytes,4); |
---|
855 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
856 | P1[1] = vce->data; |
---|
857 | } |
---|
858 | |
---|
859 | /* Process Row-0 data 2 --> data r2-1 */ |
---|
860 | for (j = 2; j < r2; j++) |
---|
861 | { |
---|
862 | if(j%blockEdgeSize==0) |
---|
863 | { |
---|
864 | J++; |
---|
865 | realPrecision = pwrErrBound[I*R2+J]; |
---|
866 | updateReqLength = 0; |
---|
867 | } |
---|
868 | |
---|
869 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
870 | diff = spaceFillingValue[j] - pred1D; |
---|
871 | |
---|
872 | itvNum = fabs(diff)/realPrecision + 1; |
---|
873 | |
---|
874 | if (itvNum < exe_params->intvCapacity) |
---|
875 | { |
---|
876 | if (diff < 0) itvNum = -itvNum; |
---|
877 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
878 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
879 | } |
---|
880 | else |
---|
881 | { |
---|
882 | if(updateReqLength==0) |
---|
883 | { |
---|
884 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
885 | reqBytesLength = reqLength/8; |
---|
886 | resiBitsLength = reqLength%8; |
---|
887 | updateReqLength = 1; |
---|
888 | } |
---|
889 | |
---|
890 | type[j] = 0; |
---|
891 | |
---|
892 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
893 | compressSingleFloatValue(vce, spaceFillingValue[j], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
894 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
895 | memcpy(preDataBytes,vce->curBytes,4); |
---|
896 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
897 | P1[j] = vce->data; |
---|
898 | } |
---|
899 | } |
---|
900 | |
---|
901 | /* Process Row-1 --> Row-r1-1 */ |
---|
902 | size_t index; |
---|
903 | for (i = 1; i < r1; i++) |
---|
904 | { |
---|
905 | /* Process row-i data 0 */ |
---|
906 | index = i*r2; |
---|
907 | J = 0; |
---|
908 | if(i%blockEdgeSize==0) |
---|
909 | I++; |
---|
910 | realPrecision = pwrErrBound[I*R2+J]; //J==0 |
---|
911 | updateReqLength = 0; |
---|
912 | |
---|
913 | pred1D = P1[0]; |
---|
914 | diff = spaceFillingValue[index] - pred1D; |
---|
915 | |
---|
916 | itvNum = fabs(diff)/realPrecision + 1; |
---|
917 | |
---|
918 | if (itvNum < exe_params->intvCapacity) |
---|
919 | { |
---|
920 | if (diff < 0) itvNum = -itvNum; |
---|
921 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
922 | P0[0] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
923 | } |
---|
924 | else |
---|
925 | { |
---|
926 | if(updateReqLength==0) |
---|
927 | { |
---|
928 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
929 | reqBytesLength = reqLength/8; |
---|
930 | resiBitsLength = reqLength%8; |
---|
931 | updateReqLength = 1; |
---|
932 | } |
---|
933 | |
---|
934 | type[index] = 0; |
---|
935 | |
---|
936 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
937 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
938 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
939 | memcpy(preDataBytes,vce->curBytes,4); |
---|
940 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
941 | P0[0] = vce->data; |
---|
942 | } |
---|
943 | |
---|
944 | /* Process row-i data 1 --> r2-1*/ |
---|
945 | for (j = 1; j < r2; j++) |
---|
946 | { |
---|
947 | index = i*r2+j; |
---|
948 | |
---|
949 | if(j%blockEdgeSize==0) |
---|
950 | { |
---|
951 | J++; |
---|
952 | realPrecision = pwrErrBound[I*R2+J]; |
---|
953 | updateReqLength = 0; |
---|
954 | } |
---|
955 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
956 | |
---|
957 | diff = spaceFillingValue[index] - pred2D; |
---|
958 | |
---|
959 | itvNum = fabs(diff)/realPrecision + 1; |
---|
960 | |
---|
961 | if (itvNum < exe_params->intvCapacity) |
---|
962 | { |
---|
963 | if (diff < 0) itvNum = -itvNum; |
---|
964 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
965 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
966 | } |
---|
967 | else |
---|
968 | { |
---|
969 | if(updateReqLength==0) |
---|
970 | { |
---|
971 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
972 | reqBytesLength = reqLength/8; |
---|
973 | resiBitsLength = reqLength%8; |
---|
974 | updateReqLength = 1; |
---|
975 | } |
---|
976 | |
---|
977 | type[index] = 0; |
---|
978 | |
---|
979 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
980 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
981 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
982 | memcpy(preDataBytes,vce->curBytes,4); |
---|
983 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
984 | P0[j] = vce->data; |
---|
985 | } |
---|
986 | } |
---|
987 | |
---|
988 | float *Pt; |
---|
989 | Pt = P1; |
---|
990 | P1 = P0; |
---|
991 | P0 = Pt; |
---|
992 | } |
---|
993 | |
---|
994 | if(r2!=1) |
---|
995 | free(P0); |
---|
996 | free(P1); |
---|
997 | int exactDataNum = exactLeadNumArray->size; |
---|
998 | |
---|
999 | TightDataPointStorageF* tdps; |
---|
1000 | |
---|
1001 | new_TightDataPointStorageF2(&tdps, dataLength, exactDataNum, |
---|
1002 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
1003 | exactLeadNumArray->array, |
---|
1004 | resiBitArray->array, resiBitArray->size, |
---|
1005 | resiBitLengthArray->array, resiBitLengthArray->size, |
---|
1006 | realPrecision, medianValue, (char)reqLength, quantization_intervals, pwrErrBoundBytes, pwrErrBoundBytes_size, radExpo); |
---|
1007 | |
---|
1008 | //free memory |
---|
1009 | free_DBA(resiBitLengthArray); |
---|
1010 | free_DIA(exactLeadNumArray); |
---|
1011 | free_DIA(resiBitArray); |
---|
1012 | free(type); |
---|
1013 | |
---|
1014 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1015 | |
---|
1016 | free(pwrErrBound); |
---|
1017 | |
---|
1018 | free(vce); |
---|
1019 | free(lce); |
---|
1020 | free_TightDataPointStorageF(tdps); |
---|
1021 | free(exactMidByteArray); |
---|
1022 | } |
---|
1023 | |
---|
1024 | void SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(unsigned char** newByteData, float *oriData, double globalPrecision, |
---|
1025 | size_t r1, size_t r2, size_t r3, size_t *outSize, float min, float max) |
---|
1026 | { |
---|
1027 | size_t dataLength=r1*r2*r3; |
---|
1028 | |
---|
1029 | int blockEdgeSize = computeBlockEdgeSize_3D(confparams_cpr->segment_size); |
---|
1030 | size_t R1 = 1+(r1-1)/blockEdgeSize; |
---|
1031 | size_t R2 = 1+(r2-1)/blockEdgeSize; |
---|
1032 | size_t R3 = 1+(r3-1)/blockEdgeSize; |
---|
1033 | float* pwrErrBound = (float*)malloc(sizeof(float)*R1*R2*R3); |
---|
1034 | size_t pwrErrBoundBytes_size = sizeof(unsigned char)*R1*R2*R3*2; |
---|
1035 | unsigned char* pwrErrBoundBytes = (unsigned char*)malloc(pwrErrBoundBytes_size); |
---|
1036 | |
---|
1037 | compute_segment_precisions_float_3D(oriData, pwrErrBound, r1, r2, r3, R2, R3, blockEdgeSize, pwrErrBoundBytes, min, max, globalPrecision); |
---|
1038 | |
---|
1039 | unsigned int quantization_intervals; |
---|
1040 | if(exe_params->optQuantMode==1) |
---|
1041 | { |
---|
1042 | quantization_intervals = optimize_intervals_float_3D_pwr(oriData, r1, r2, r3, R2, R3, blockEdgeSize, pwrErrBound); |
---|
1043 | updateQuantizationInfo(quantization_intervals); |
---|
1044 | } |
---|
1045 | else |
---|
1046 | quantization_intervals = exe_params->intvCapacity; |
---|
1047 | size_t i=0,j=0,k=0, I = 0, J = 0, K = 0; |
---|
1048 | int reqLength; |
---|
1049 | float realPrecision = pwrErrBound[0]; |
---|
1050 | float pred1D, pred2D, pred3D; |
---|
1051 | float diff = 0.0; |
---|
1052 | double itvNum = 0; |
---|
1053 | float *P0, *P1; |
---|
1054 | |
---|
1055 | size_t r23 = r2*r3; |
---|
1056 | size_t R23 = R2*R3; |
---|
1057 | P0 = (float*)malloc(r23*sizeof(float)); |
---|
1058 | P1 = (float*)malloc(r23*sizeof(float)); |
---|
1059 | float radius = fabs(max)<fabs(min)?fabs(min):fabs(max); |
---|
1060 | float medianValue = 0; |
---|
1061 | short radExpo = getExponent_float(radius); |
---|
1062 | int updateReqLength = 0; |
---|
1063 | |
---|
1064 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1065 | |
---|
1066 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
1067 | //type[dataLength]=0;realPrecision |
---|
1068 | |
---|
1069 | float* spaceFillingValue = oriData; // |
---|
1070 | |
---|
1071 | DynamicByteArray *resiBitLengthArray; |
---|
1072 | new_DBA(&resiBitLengthArray, DynArrayInitLen); |
---|
1073 | |
---|
1074 | DynamicIntArray *exactLeadNumArray; |
---|
1075 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
1076 | |
---|
1077 | DynamicByteArray *exactMidByteArray; |
---|
1078 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
1079 | |
---|
1080 | DynamicIntArray *resiBitArray; |
---|
1081 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
1082 | |
---|
1083 | type[0] = 0; |
---|
1084 | |
---|
1085 | unsigned char preDataBytes[4]; |
---|
1086 | intToBytes_bigEndian(preDataBytes, 0); |
---|
1087 | |
---|
1088 | int reqBytesLength = reqLength/8; |
---|
1089 | int resiBitsLength = reqLength%8; |
---|
1090 | |
---|
1091 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
1092 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
1093 | |
---|
1094 | |
---|
1095 | /////////////////////////// Process layer-0 /////////////////////////// |
---|
1096 | /* Process Row-0 data 0*/ |
---|
1097 | type[0] = 0; |
---|
1098 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1099 | compressSingleFloatValue(vce, spaceFillingValue[0], 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 | P1[0] = vce->data; |
---|
1104 | |
---|
1105 | /* Process Row-0 data 1*/ |
---|
1106 | pred1D = P1[0]; |
---|
1107 | diff = spaceFillingValue[1] - pred1D; |
---|
1108 | |
---|
1109 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1110 | |
---|
1111 | if (itvNum < exe_params->intvCapacity) |
---|
1112 | { |
---|
1113 | if (diff < 0) itvNum = -itvNum; |
---|
1114 | type[1] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1115 | P1[1] = pred1D + 2 * (type[1] - exe_params->intvRadius) * realPrecision; |
---|
1116 | } |
---|
1117 | else |
---|
1118 | { |
---|
1119 | if(updateReqLength==0) |
---|
1120 | { |
---|
1121 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1122 | reqBytesLength = reqLength/8; |
---|
1123 | resiBitsLength = reqLength%8; |
---|
1124 | updateReqLength = 1; |
---|
1125 | } |
---|
1126 | |
---|
1127 | type[1] = 0; |
---|
1128 | |
---|
1129 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1130 | compressSingleFloatValue(vce, spaceFillingValue[1], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1131 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1132 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1133 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1134 | P1[1] = vce->data; |
---|
1135 | } |
---|
1136 | |
---|
1137 | /* Process Row-0 data 2 --> data r3-1 */ |
---|
1138 | for (j = 2; j < r3; j++) |
---|
1139 | { |
---|
1140 | if(j%blockEdgeSize==0) |
---|
1141 | { |
---|
1142 | J++; |
---|
1143 | realPrecision = pwrErrBound[J]; |
---|
1144 | updateReqLength = 0; |
---|
1145 | } |
---|
1146 | pred1D = 2*P1[j-1] - P1[j-2]; |
---|
1147 | diff = spaceFillingValue[j] - pred1D; |
---|
1148 | |
---|
1149 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1150 | |
---|
1151 | if (itvNum < exe_params->intvCapacity) |
---|
1152 | { |
---|
1153 | if (diff < 0) itvNum = -itvNum; |
---|
1154 | type[j] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1155 | P1[j] = pred1D + 2 * (type[j] - exe_params->intvRadius) * realPrecision; |
---|
1156 | } |
---|
1157 | else |
---|
1158 | { |
---|
1159 | if(updateReqLength==0) |
---|
1160 | { |
---|
1161 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1162 | reqBytesLength = reqLength/8; |
---|
1163 | resiBitsLength = reqLength%8; |
---|
1164 | updateReqLength = 1; |
---|
1165 | } |
---|
1166 | |
---|
1167 | type[j] = 0; |
---|
1168 | |
---|
1169 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1170 | compressSingleFloatValue(vce, spaceFillingValue[j], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1171 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1172 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1173 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1174 | P1[j] = vce->data; |
---|
1175 | } |
---|
1176 | } |
---|
1177 | |
---|
1178 | /* Process Row-1 --> Row-r2-1 */ |
---|
1179 | size_t index; |
---|
1180 | K = 0; |
---|
1181 | for (i = 1; i < r2; i++) |
---|
1182 | { |
---|
1183 | /* Process row-i data 0 */ |
---|
1184 | index = i*r3; |
---|
1185 | |
---|
1186 | J = 0; |
---|
1187 | if(i%blockEdgeSize==0) |
---|
1188 | I++; |
---|
1189 | realPrecision = pwrErrBound[I*R3+J]; //J==0 |
---|
1190 | updateReqLength = 0; |
---|
1191 | |
---|
1192 | pred1D = P1[index-r3]; |
---|
1193 | diff = spaceFillingValue[index] - pred1D; |
---|
1194 | |
---|
1195 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1196 | |
---|
1197 | if (itvNum < exe_params->intvCapacity) |
---|
1198 | { |
---|
1199 | if (diff < 0) itvNum = -itvNum; |
---|
1200 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1201 | P1[index] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1202 | } |
---|
1203 | else |
---|
1204 | { |
---|
1205 | if(updateReqLength==0) |
---|
1206 | { |
---|
1207 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1208 | reqBytesLength = reqLength/8; |
---|
1209 | resiBitsLength = reqLength%8; |
---|
1210 | updateReqLength = 1; |
---|
1211 | } |
---|
1212 | |
---|
1213 | type[index] = 0; |
---|
1214 | |
---|
1215 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1216 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1217 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1218 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1219 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1220 | P1[index] = vce->data; |
---|
1221 | } |
---|
1222 | |
---|
1223 | /* Process row-i data 1 --> data r3-1*/ |
---|
1224 | for (j = 1; j < r3; j++) //note that this j refers to fastest dimension (lowest order) |
---|
1225 | { |
---|
1226 | index = i*r3+j; |
---|
1227 | if(j%blockEdgeSize==0) |
---|
1228 | { |
---|
1229 | J++; |
---|
1230 | realPrecision = pwrErrBound[I*R3+J]; |
---|
1231 | updateReqLength = 0; |
---|
1232 | } |
---|
1233 | |
---|
1234 | pred2D = P1[index-1] + P1[index-r3] - P1[index-r3-1]; |
---|
1235 | |
---|
1236 | diff = spaceFillingValue[index] - pred2D; |
---|
1237 | |
---|
1238 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1239 | |
---|
1240 | if (itvNum < exe_params->intvCapacity) |
---|
1241 | { |
---|
1242 | if (diff < 0) itvNum = -itvNum; |
---|
1243 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1244 | P1[index] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1245 | } |
---|
1246 | else |
---|
1247 | { |
---|
1248 | if(updateReqLength==0) |
---|
1249 | { |
---|
1250 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1251 | reqBytesLength = reqLength/8; |
---|
1252 | resiBitsLength = reqLength%8; |
---|
1253 | updateReqLength = 1; |
---|
1254 | } |
---|
1255 | |
---|
1256 | type[index] = 0; |
---|
1257 | |
---|
1258 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1259 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1260 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1261 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1262 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1263 | P1[index] = vce->data; |
---|
1264 | } |
---|
1265 | } |
---|
1266 | } |
---|
1267 | |
---|
1268 | /////////////////////////// Process layer-1 --> layer-r1-1 /////////////////////////// |
---|
1269 | |
---|
1270 | for (k = 1; k < r1; k++) |
---|
1271 | { |
---|
1272 | /* Process Row-0 data 0*/ |
---|
1273 | index = k*r23; |
---|
1274 | I = 0; |
---|
1275 | J = 0; |
---|
1276 | if(k%blockEdgeSize==0) |
---|
1277 | K++; |
---|
1278 | realPrecision = pwrErrBound[K*R23]; //J==0 |
---|
1279 | updateReqLength = 0; |
---|
1280 | |
---|
1281 | pred1D = P1[0]; |
---|
1282 | diff = spaceFillingValue[index] - pred1D; |
---|
1283 | |
---|
1284 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1285 | |
---|
1286 | if (itvNum < exe_params->intvCapacity) |
---|
1287 | { |
---|
1288 | if (diff < 0) itvNum = -itvNum; |
---|
1289 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1290 | P0[0] = pred1D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1291 | } |
---|
1292 | else |
---|
1293 | { |
---|
1294 | if(updateReqLength==0) |
---|
1295 | { |
---|
1296 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1297 | reqBytesLength = reqLength/8; |
---|
1298 | resiBitsLength = reqLength%8; |
---|
1299 | updateReqLength = 1; |
---|
1300 | } |
---|
1301 | |
---|
1302 | type[index] = 0; |
---|
1303 | |
---|
1304 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1305 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1306 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1307 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1308 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1309 | P0[0] = vce->data; |
---|
1310 | } |
---|
1311 | |
---|
1312 | /* Process Row-0 data 1 --> data r3-1 */ |
---|
1313 | for (j = 1; j < r3; j++) |
---|
1314 | { |
---|
1315 | index = k*r23+j; |
---|
1316 | |
---|
1317 | if(j%blockEdgeSize==0) |
---|
1318 | { |
---|
1319 | J++; |
---|
1320 | realPrecision = pwrErrBound[K*R23+J]; |
---|
1321 | updateReqLength = 0; |
---|
1322 | } |
---|
1323 | pred2D = P0[j-1] + P1[j] - P1[j-1]; |
---|
1324 | diff = spaceFillingValue[index] - pred2D; |
---|
1325 | |
---|
1326 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1327 | |
---|
1328 | if (itvNum < exe_params->intvCapacity) |
---|
1329 | { |
---|
1330 | if (diff < 0) itvNum = -itvNum; |
---|
1331 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1332 | P0[j] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1333 | /* if(type[index]==0) |
---|
1334 | printf("err:type[%d]=0, index4\n", index); */ |
---|
1335 | } |
---|
1336 | else |
---|
1337 | { |
---|
1338 | if(updateReqLength==0) |
---|
1339 | { |
---|
1340 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1341 | reqBytesLength = reqLength/8; |
---|
1342 | resiBitsLength = reqLength%8; |
---|
1343 | updateReqLength = 1; |
---|
1344 | } |
---|
1345 | |
---|
1346 | type[index] = 0; |
---|
1347 | |
---|
1348 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1349 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1350 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1351 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1352 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1353 | P0[j] = vce->data; |
---|
1354 | } |
---|
1355 | } |
---|
1356 | |
---|
1357 | /* Process Row-1 --> Row-r2-1 */ |
---|
1358 | size_t index2D; |
---|
1359 | for (i = 1; i < r2; i++) |
---|
1360 | { |
---|
1361 | /* Process Row-i data 0 */ |
---|
1362 | index = k*r23 + i*r3; |
---|
1363 | J = 0; |
---|
1364 | if(i%blockEdgeSize==0) |
---|
1365 | I++; |
---|
1366 | realPrecision = pwrErrBound[K*R23+I*R3+J]; //J==0 |
---|
1367 | updateReqLength = 0; |
---|
1368 | |
---|
1369 | index2D = i*r3; |
---|
1370 | pred2D = P0[index2D-r3] + P1[index2D] - P1[index2D-r3]; |
---|
1371 | diff = spaceFillingValue[index] - pred2D; |
---|
1372 | |
---|
1373 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1374 | |
---|
1375 | if (itvNum < exe_params->intvCapacity) |
---|
1376 | { |
---|
1377 | if (diff < 0) itvNum = -itvNum; |
---|
1378 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1379 | P0[index2D] = pred2D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1380 | } |
---|
1381 | else |
---|
1382 | { |
---|
1383 | if(updateReqLength==0) |
---|
1384 | { |
---|
1385 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1386 | reqBytesLength = reqLength/8; |
---|
1387 | resiBitsLength = reqLength%8; |
---|
1388 | updateReqLength = 1; |
---|
1389 | } |
---|
1390 | |
---|
1391 | type[index] = 0; |
---|
1392 | |
---|
1393 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1394 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1395 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1396 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1397 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1398 | P0[index2D] = vce->data; |
---|
1399 | } |
---|
1400 | |
---|
1401 | /* Process Row-i data 1 --> data r3-1 */ |
---|
1402 | for (j = 1; j < r3; j++) |
---|
1403 | { |
---|
1404 | index = k*r23 + i*r3 + j; |
---|
1405 | if(j%blockEdgeSize==0) |
---|
1406 | { |
---|
1407 | J++; |
---|
1408 | realPrecision = pwrErrBound[K*R23+I*R3+J]; |
---|
1409 | updateReqLength = 0; |
---|
1410 | } |
---|
1411 | index2D = i*r3 + j; |
---|
1412 | pred3D = P0[index2D-1] + P0[index2D-r3]+ P1[index2D] - P0[index2D-r3-1] - P1[index2D-r3] - P1[index2D-1] + P1[index2D-r3-1]; |
---|
1413 | diff = spaceFillingValue[index] - pred3D; |
---|
1414 | |
---|
1415 | itvNum = fabs(diff)/realPrecision + 1; |
---|
1416 | |
---|
1417 | if (itvNum < exe_params->intvCapacity) |
---|
1418 | { |
---|
1419 | if (diff < 0) itvNum = -itvNum; |
---|
1420 | type[index] = (int) (itvNum/2) + exe_params->intvRadius; |
---|
1421 | P0[index2D] = pred3D + 2 * (type[index] - exe_params->intvRadius) * realPrecision; |
---|
1422 | } |
---|
1423 | else |
---|
1424 | { |
---|
1425 | if(updateReqLength==0) |
---|
1426 | { |
---|
1427 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1428 | reqBytesLength = reqLength/8; |
---|
1429 | resiBitsLength = reqLength%8; |
---|
1430 | updateReqLength = 1; |
---|
1431 | } |
---|
1432 | |
---|
1433 | type[index] = 0; |
---|
1434 | |
---|
1435 | addDBA_Data(resiBitLengthArray, (unsigned char)resiBitsLength); |
---|
1436 | compressSingleFloatValue(vce, spaceFillingValue[index], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1437 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1438 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1439 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1440 | P0[index2D] = vce->data; |
---|
1441 | } |
---|
1442 | } |
---|
1443 | } |
---|
1444 | |
---|
1445 | float *Pt; |
---|
1446 | Pt = P1; |
---|
1447 | P1 = P0; |
---|
1448 | P0 = Pt; |
---|
1449 | } |
---|
1450 | if(r23!=1) |
---|
1451 | free(P0); |
---|
1452 | free(P1); |
---|
1453 | int exactDataNum = exactLeadNumArray->size; |
---|
1454 | |
---|
1455 | TightDataPointStorageF* tdps; |
---|
1456 | |
---|
1457 | new_TightDataPointStorageF2(&tdps, dataLength, exactDataNum, |
---|
1458 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
1459 | exactLeadNumArray->array, |
---|
1460 | resiBitArray->array, resiBitArray->size, |
---|
1461 | resiBitLengthArray->array, resiBitLengthArray->size, |
---|
1462 | realPrecision, medianValue, (char)reqLength, quantization_intervals, pwrErrBoundBytes, pwrErrBoundBytes_size, radExpo); |
---|
1463 | |
---|
1464 | //sdi:Debug |
---|
1465 | /* int sum =0; |
---|
1466 | for(i=0;i<dataLength;i++) |
---|
1467 | if(type[i]==0) sum++; |
---|
1468 | printf("opt_quantizations=%d, exactDataNum=%d, sum=%d\n",quantization_intervals, exactDataNum, sum); |
---|
1469 | */ |
---|
1470 | |
---|
1471 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1472 | |
---|
1473 | //free memory |
---|
1474 | free_DBA(resiBitLengthArray); |
---|
1475 | free_DIA(exactLeadNumArray); |
---|
1476 | free_DIA(resiBitArray); |
---|
1477 | free(type); |
---|
1478 | |
---|
1479 | |
---|
1480 | free(pwrErrBound); |
---|
1481 | |
---|
1482 | free(vce); |
---|
1483 | free(lce); |
---|
1484 | free_TightDataPointStorageF(tdps); |
---|
1485 | free(exactMidByteArray); |
---|
1486 | } |
---|
1487 | |
---|
1488 | void createRangeGroups_float(float** posGroups, float** negGroups, int** posFlags, int** negFlags) |
---|
1489 | { |
---|
1490 | size_t size = GROUP_COUNT*sizeof(float); |
---|
1491 | size_t size2 = GROUP_COUNT*sizeof(int); |
---|
1492 | *posGroups = (float*)malloc(size); |
---|
1493 | *negGroups = (float*)malloc(size); |
---|
1494 | *posFlags = (int*)malloc(size2); |
---|
1495 | *negFlags = (int*)malloc(size2); |
---|
1496 | memset(*posGroups, 0, size); |
---|
1497 | memset(*negGroups, 0, size); |
---|
1498 | memset(*posFlags, 0, size2); |
---|
1499 | memset(*negFlags, 0, size2); |
---|
1500 | } |
---|
1501 | |
---|
1502 | void compressGroupIDArray_float(char* groupID, TightDataPointStorageF* tdps) |
---|
1503 | { |
---|
1504 | size_t dataLength = tdps->dataSeriesLength; |
---|
1505 | int* standGroupID = (int*)malloc(dataLength*sizeof(int)); |
---|
1506 | |
---|
1507 | size_t i; |
---|
1508 | standGroupID[0] = groupID[0]+GROUP_COUNT; //plus an offset such that it would not be a negative number. |
---|
1509 | char lastGroupIDValue = groupID[0], curGroupIDValue; |
---|
1510 | int offset = 2*(GROUP_COUNT + 2); |
---|
1511 | for(i=1; i<dataLength;i++) |
---|
1512 | { |
---|
1513 | curGroupIDValue = groupID[i]; |
---|
1514 | standGroupID[i] = (curGroupIDValue - lastGroupIDValue) + offset; |
---|
1515 | lastGroupIDValue = curGroupIDValue; |
---|
1516 | } |
---|
1517 | |
---|
1518 | unsigned char* out = NULL; |
---|
1519 | size_t outSize; |
---|
1520 | |
---|
1521 | HuffmanTree* huffmanTree = SZ_Reset(); |
---|
1522 | encode_withTree(huffmanTree, standGroupID, dataLength, &out, &outSize); |
---|
1523 | SZ_ReleaseHuffman(huffmanTree); |
---|
1524 | |
---|
1525 | tdps->pwrErrBoundBytes = out; //groupIDArray |
---|
1526 | tdps->pwrErrBoundBytes_size = outSize; |
---|
1527 | |
---|
1528 | free(standGroupID); |
---|
1529 | } |
---|
1530 | |
---|
1531 | TightDataPointStorageF* SZ_compress_float_1D_MDQ_pwrGroup(float* oriData, size_t dataLength, int errBoundMode, |
---|
1532 | double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSize, float medianValue_f) |
---|
1533 | { |
---|
1534 | size_t i; |
---|
1535 | float *posGroups, *negGroups, *groups; |
---|
1536 | float pos_01_group = 0, neg_01_group = 0; //[0,1] and [-1,0] |
---|
1537 | int *posFlags, *negFlags, *flags; |
---|
1538 | int pos_01_flag = 0, neg_01_flag = 0; |
---|
1539 | createRangeGroups_float(&posGroups, &negGroups, &posFlags, &negFlags); |
---|
1540 | size_t nbBins = (size_t)(1/pwrErrRatio); |
---|
1541 | if(nbBins%2==1) |
---|
1542 | nbBins++; |
---|
1543 | exe_params->intvRadius = nbBins; |
---|
1544 | |
---|
1545 | int reqLength, status; |
---|
1546 | float medianValue = medianValue_f; |
---|
1547 | float realPrecision = (float)getRealPrecision_float(valueRangeSize, errBoundMode, absErrBound, relBoundRatio, &status); |
---|
1548 | if(realPrecision<0) |
---|
1549 | realPrecision = pwrErrRatio; |
---|
1550 | float realGroupPrecision; //precision (error) based on group ID |
---|
1551 | getPrecisionReqLength_float(realPrecision); |
---|
1552 | short radExpo = getExponent_float(valueRangeSize/2); |
---|
1553 | short lastGroupNum = 0, groupNum, grpNum = 0; |
---|
1554 | |
---|
1555 | double* groupErrorBounds = generateGroupErrBounds(errBoundMode, realPrecision, pwrErrRatio); |
---|
1556 | exe_params->intvRadius = generateGroupMaxIntervalCount(groupErrorBounds); |
---|
1557 | |
---|
1558 | computeReqLength_float(realPrecision, radExpo, &reqLength, &medianValue); |
---|
1559 | |
---|
1560 | int* type = (int*) malloc(dataLength*sizeof(int)); |
---|
1561 | char *groupID = (char*) malloc(dataLength*sizeof(char)); |
---|
1562 | char *gp = groupID; |
---|
1563 | |
---|
1564 | float* spaceFillingValue = oriData; |
---|
1565 | |
---|
1566 | DynamicIntArray *exactLeadNumArray; |
---|
1567 | new_DIA(&exactLeadNumArray, DynArrayInitLen); |
---|
1568 | |
---|
1569 | DynamicByteArray *exactMidByteArray; |
---|
1570 | new_DBA(&exactMidByteArray, DynArrayInitLen); |
---|
1571 | |
---|
1572 | DynamicIntArray *resiBitArray; |
---|
1573 | new_DIA(&resiBitArray, DynArrayInitLen); |
---|
1574 | |
---|
1575 | unsigned char preDataBytes[4]; |
---|
1576 | intToBytes_bigEndian(preDataBytes, 0); |
---|
1577 | |
---|
1578 | int reqBytesLength = reqLength/8; |
---|
1579 | int resiBitsLength = reqLength%8; |
---|
1580 | |
---|
1581 | FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement)); |
---|
1582 | LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement)); |
---|
1583 | |
---|
1584 | int state; |
---|
1585 | float curData, decValue; |
---|
1586 | float pred; |
---|
1587 | float predAbsErr; |
---|
1588 | double interval = 0; |
---|
1589 | |
---|
1590 | //add the first data |
---|
1591 | type[0] = 0; |
---|
1592 | compressSingleFloatValue(vce, spaceFillingValue[0], realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1593 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1594 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1595 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1596 | |
---|
1597 | curData = spaceFillingValue[0]; |
---|
1598 | groupNum = computeGroupNum_float(vce->data); |
---|
1599 | |
---|
1600 | if(curData > 0 && groupNum >= 0) |
---|
1601 | { |
---|
1602 | groups = posGroups; |
---|
1603 | flags = posFlags; |
---|
1604 | grpNum = groupNum; |
---|
1605 | } |
---|
1606 | else if(curData < 0 && groupNum >= 0) |
---|
1607 | { |
---|
1608 | groups = negGroups; |
---|
1609 | flags = negFlags; |
---|
1610 | grpNum = groupNum; |
---|
1611 | } |
---|
1612 | else if(curData >= 0 && groupNum == -1) |
---|
1613 | { |
---|
1614 | groups = &pos_01_group; |
---|
1615 | flags = &pos_01_flag; |
---|
1616 | grpNum = 0; |
---|
1617 | } |
---|
1618 | else //curData < 0 && groupNum == -1 |
---|
1619 | { |
---|
1620 | groups = &neg_01_group; |
---|
1621 | flags = &neg_01_flag; |
---|
1622 | grpNum = 0; |
---|
1623 | } |
---|
1624 | |
---|
1625 | listAdd_float_group(groups, flags, groupNum, spaceFillingValue[0], vce->data, gp); |
---|
1626 | gp++; |
---|
1627 | |
---|
1628 | for(i=1;i<dataLength;i++) |
---|
1629 | { |
---|
1630 | curData = oriData[i]; |
---|
1631 | //printf("i=%d, posGroups[3]=%f, negGroups[3]=%f\n", i, posGroups[3], negGroups[3]); |
---|
1632 | |
---|
1633 | groupNum = computeGroupNum_float(curData); |
---|
1634 | |
---|
1635 | if(curData > 0 && groupNum >= 0) |
---|
1636 | { |
---|
1637 | groups = posGroups; |
---|
1638 | flags = posFlags; |
---|
1639 | grpNum = groupNum; |
---|
1640 | } |
---|
1641 | else if(curData < 0 && groupNum >= 0) |
---|
1642 | { |
---|
1643 | groups = negGroups; |
---|
1644 | flags = negFlags; |
---|
1645 | grpNum = groupNum; |
---|
1646 | } |
---|
1647 | else if(curData >= 0 && groupNum == -1) |
---|
1648 | { |
---|
1649 | groups = &pos_01_group; |
---|
1650 | flags = &pos_01_flag; |
---|
1651 | grpNum = 0; |
---|
1652 | } |
---|
1653 | else //curData < 0 && groupNum == -1 |
---|
1654 | { |
---|
1655 | groups = &neg_01_group; |
---|
1656 | flags = &neg_01_flag; |
---|
1657 | grpNum = 0; |
---|
1658 | } |
---|
1659 | |
---|
1660 | if(groupNum>=GROUP_COUNT) |
---|
1661 | { |
---|
1662 | type[i] = 0; |
---|
1663 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1664 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1665 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1666 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1667 | listAdd_float_group(groups, flags, lastGroupNum, curData, vce->data, gp); //set the group number to be last one in order to get the groupID array as smooth as possible. |
---|
1668 | } |
---|
1669 | else if(flags[grpNum]==0) //the dec value may not be in the same group |
---|
1670 | { |
---|
1671 | type[i] = 0; |
---|
1672 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1673 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1674 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1675 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1676 | //decGroupNum = computeGroupNum_float(vce->data); |
---|
1677 | |
---|
1678 | //if(decGroupNum < groupNum) |
---|
1679 | // decValue = curData>0?pow(2, groupNum):-pow(2, groupNum); |
---|
1680 | //else if(decGroupNum > groupNum) |
---|
1681 | // decValue = curData>0?pow(2, groupNum+1):-pow(2, groupNum+1); |
---|
1682 | //else |
---|
1683 | // decValue = vce->data; |
---|
1684 | |
---|
1685 | decValue = vce->data; |
---|
1686 | listAdd_float_group(groups, flags, groupNum, curData, decValue, gp); |
---|
1687 | lastGroupNum = curData>0?groupNum + 2: -(groupNum+2); |
---|
1688 | } |
---|
1689 | else //if flags[groupNum]==1, the dec value must be in the same group |
---|
1690 | { |
---|
1691 | pred = groups[grpNum]; |
---|
1692 | predAbsErr = fabs(curData - pred); |
---|
1693 | realGroupPrecision = groupErrorBounds[grpNum]; //compute real error bound |
---|
1694 | interval = realGroupPrecision*2; |
---|
1695 | state = (predAbsErr/realGroupPrecision+1)/2; |
---|
1696 | if(curData>=pred) |
---|
1697 | { |
---|
1698 | type[i] = exe_params->intvRadius+state; |
---|
1699 | decValue = pred + state*interval; |
---|
1700 | } |
---|
1701 | else //curData<pred |
---|
1702 | { |
---|
1703 | type[i] = exe_params->intvRadius-state; |
---|
1704 | decValue = pred - state*interval; |
---|
1705 | } |
---|
1706 | //decGroupNum = computeGroupNum_float(pred); |
---|
1707 | |
---|
1708 | if((decValue>0&&curData<0)||(decValue<0&&curData>=0)) |
---|
1709 | decValue = 0; |
---|
1710 | //else |
---|
1711 | //{ |
---|
1712 | // if(decGroupNum < groupNum) |
---|
1713 | // decValue = curData>0?pow(2, groupNum):-pow(2, groupNum); |
---|
1714 | // else if(decGroupNum > groupNum) |
---|
1715 | // decValue = curData>0?pow(2, groupNum+1):-pow(2, groupNum+1); |
---|
1716 | // else |
---|
1717 | // decValue = pred; |
---|
1718 | //} |
---|
1719 | |
---|
1720 | if(fabs(curData-decValue)>realGroupPrecision) |
---|
1721 | { |
---|
1722 | type[i] = 0; |
---|
1723 | compressSingleFloatValue(vce, curData, realPrecision, medianValue, reqLength, reqBytesLength, resiBitsLength); |
---|
1724 | updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce); |
---|
1725 | memcpy(preDataBytes,vce->curBytes,4); |
---|
1726 | addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce); |
---|
1727 | |
---|
1728 | decValue = vce->data; |
---|
1729 | } |
---|
1730 | |
---|
1731 | listAdd_float_group(groups, flags, groupNum, curData, decValue, gp); |
---|
1732 | lastGroupNum = curData>=0?groupNum + 2: -(groupNum+2); |
---|
1733 | } |
---|
1734 | gp++; |
---|
1735 | |
---|
1736 | } |
---|
1737 | |
---|
1738 | int exactDataNum = exactLeadNumArray->size; |
---|
1739 | |
---|
1740 | TightDataPointStorageF* tdps; |
---|
1741 | |
---|
1742 | //combineTypeAndGroupIDArray(nbBins, dataLength, &type, groupID); |
---|
1743 | |
---|
1744 | new_TightDataPointStorageF(&tdps, dataLength, exactDataNum, |
---|
1745 | type, exactMidByteArray->array, exactMidByteArray->size, |
---|
1746 | exactLeadNumArray->array, |
---|
1747 | resiBitArray->array, resiBitArray->size, |
---|
1748 | resiBitsLength, |
---|
1749 | realPrecision, medianValue, (char)reqLength, nbBins, NULL, 0, radExpo); |
---|
1750 | |
---|
1751 | compressGroupIDArray_float(groupID, tdps); |
---|
1752 | |
---|
1753 | free(posGroups); |
---|
1754 | free(negGroups); |
---|
1755 | free(posFlags); |
---|
1756 | free(negFlags); |
---|
1757 | free(groupID); |
---|
1758 | free(groupErrorBounds); |
---|
1759 | |
---|
1760 | free_DIA(exactLeadNumArray); |
---|
1761 | free_DIA(resiBitArray); |
---|
1762 | free(type); |
---|
1763 | free(vce); |
---|
1764 | free(lce); |
---|
1765 | free(exactMidByteArray); //exactMidByteArray->array has been released in free_TightDataPointStorageF(tdps); |
---|
1766 | |
---|
1767 | return tdps; |
---|
1768 | } |
---|
1769 | |
---|
1770 | void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(unsigned char** newByteData, float *oriData, |
---|
1771 | size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSize, float medianValue_f, size_t *outSize) |
---|
1772 | { |
---|
1773 | TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ_pwrGroup(oriData, dataLength, confparams_cpr->errorBoundMode, |
---|
1774 | absErrBound, relBoundRatio, pwrErrRatio, |
---|
1775 | valueRangeSize, medianValue_f); |
---|
1776 | |
---|
1777 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1778 | |
---|
1779 | if(*outSize>dataLength*sizeof(float)) |
---|
1780 | SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize); |
---|
1781 | |
---|
1782 | free_TightDataPointStorageF(tdps); |
---|
1783 | } |
---|
1784 | |
---|
1785 | #include <stdbool.h> |
---|
1786 | |
---|
1787 | void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log(unsigned char** newByteData, float *oriData, double pwrErrRatio, size_t dataLength, size_t *outSize, float min, float max){ |
---|
1788 | |
---|
1789 | float * log_data = (float *) malloc(dataLength * sizeof(float)); |
---|
1790 | |
---|
1791 | unsigned char * signs = (unsigned char *) malloc(dataLength); |
---|
1792 | memset(signs, 0, dataLength); |
---|
1793 | // preprocess |
---|
1794 | float max_abs_log_data; |
---|
1795 | if(min == 0) max_abs_log_data = fabs(log2(fabs(max))); |
---|
1796 | else if(max == 0) max_abs_log_data = fabs(log2(fabs(min))); |
---|
1797 | else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max))); |
---|
1798 | float min_log_data = max_abs_log_data; |
---|
1799 | bool positive = true; |
---|
1800 | for(size_t i=0; i<dataLength; i++){ |
---|
1801 | if(oriData[i] < 0){ |
---|
1802 | signs[i] = 1; |
---|
1803 | log_data[i] = -oriData[i]; |
---|
1804 | positive = false; |
---|
1805 | } |
---|
1806 | else |
---|
1807 | log_data[i] = oriData[i]; |
---|
1808 | if(log_data[i] > 0){ |
---|
1809 | log_data[i] = log2(log_data[i]); |
---|
1810 | if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i]; |
---|
1811 | if(log_data[i] < min_log_data) min_log_data = log_data[i]; |
---|
1812 | } |
---|
1813 | } |
---|
1814 | |
---|
1815 | float valueRangeSize, medianValue_f; |
---|
1816 | computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f); |
---|
1817 | if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data); |
---|
1818 | double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7; |
---|
1819 | for(size_t i=0; i<dataLength; i++){ |
---|
1820 | if(oriData[i] == 0){ |
---|
1821 | log_data[i] = min_log_data - 2.0001*realPrecision; |
---|
1822 | } |
---|
1823 | } |
---|
1824 | |
---|
1825 | TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ(log_data, dataLength, realPrecision, valueRangeSize, medianValue_f); |
---|
1826 | tdps->minLogValue = min_log_data - 1.0001*realPrecision; |
---|
1827 | free(log_data); |
---|
1828 | if(!positive){ |
---|
1829 | unsigned char * comp_signs; |
---|
1830 | // compress signs |
---|
1831 | unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs); |
---|
1832 | tdps->pwrErrBoundBytes = comp_signs; |
---|
1833 | tdps->pwrErrBoundBytes_size = signSize; |
---|
1834 | } |
---|
1835 | else{ |
---|
1836 | tdps->pwrErrBoundBytes = NULL; |
---|
1837 | tdps->pwrErrBoundBytes_size = 0; |
---|
1838 | } |
---|
1839 | free(signs); |
---|
1840 | |
---|
1841 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1842 | if(*outSize>dataLength*sizeof(float)) |
---|
1843 | SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize); |
---|
1844 | |
---|
1845 | free_TightDataPointStorageF(tdps); |
---|
1846 | } |
---|
1847 | |
---|
1848 | void SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr_pre_log(unsigned char** newByteData, float *oriData, double pwrErrRatio, size_t r1, size_t r2, size_t *outSize, float min, float max){ |
---|
1849 | |
---|
1850 | size_t dataLength = r1 * r2; |
---|
1851 | float * log_data = (float *) malloc(dataLength * sizeof(float)); |
---|
1852 | |
---|
1853 | unsigned char * signs = (unsigned char *) malloc(dataLength); |
---|
1854 | memset(signs, 0, dataLength); |
---|
1855 | // preprocess |
---|
1856 | float max_abs_log_data; |
---|
1857 | if(min == 0) max_abs_log_data = fabs(log2(fabs(max))); |
---|
1858 | else if(max == 0) max_abs_log_data = fabs(log2(fabs(min))); |
---|
1859 | else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max))); |
---|
1860 | float min_log_data = max_abs_log_data; |
---|
1861 | bool positive = true; |
---|
1862 | for(size_t i=0; i<dataLength; i++){ |
---|
1863 | if(oriData[i] < 0){ |
---|
1864 | signs[i] = 1; |
---|
1865 | log_data[i] = -oriData[i]; |
---|
1866 | positive = false; |
---|
1867 | } |
---|
1868 | else |
---|
1869 | log_data[i] = oriData[i]; |
---|
1870 | if(log_data[i] > 0){ |
---|
1871 | log_data[i] = log2(log_data[i]); |
---|
1872 | if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i]; |
---|
1873 | if(log_data[i] < min_log_data) min_log_data = log_data[i]; |
---|
1874 | } |
---|
1875 | } |
---|
1876 | |
---|
1877 | float valueRangeSize, medianValue_f; |
---|
1878 | computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f); |
---|
1879 | if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data); |
---|
1880 | double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7; |
---|
1881 | for(size_t i=0; i<dataLength; i++){ |
---|
1882 | if(oriData[i] == 0){ |
---|
1883 | log_data[i] = min_log_data - 2.0001*realPrecision; |
---|
1884 | } |
---|
1885 | } |
---|
1886 | |
---|
1887 | TightDataPointStorageF* tdps = SZ_compress_float_2D_MDQ(log_data, r1, r2, realPrecision, valueRangeSize, medianValue_f); |
---|
1888 | tdps->minLogValue = min_log_data - 1.0001*realPrecision; |
---|
1889 | free(log_data); |
---|
1890 | if(!positive){ |
---|
1891 | unsigned char * comp_signs; |
---|
1892 | // compress signs |
---|
1893 | unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs); |
---|
1894 | tdps->pwrErrBoundBytes = comp_signs; |
---|
1895 | tdps->pwrErrBoundBytes_size = signSize; |
---|
1896 | } |
---|
1897 | else{ |
---|
1898 | tdps->pwrErrBoundBytes = NULL; |
---|
1899 | tdps->pwrErrBoundBytes_size = 0; |
---|
1900 | } |
---|
1901 | free(signs); |
---|
1902 | |
---|
1903 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1904 | if(*outSize>dataLength*sizeof(float)) |
---|
1905 | SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize); |
---|
1906 | |
---|
1907 | free_TightDataPointStorageF(tdps); |
---|
1908 | } |
---|
1909 | |
---|
1910 | void SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_pre_log(unsigned char** newByteData, float *oriData, double pwrErrRatio, size_t r1, size_t r2, size_t r3, size_t *outSize, float min, float max){ |
---|
1911 | |
---|
1912 | size_t dataLength = r1 * r2 * r3; |
---|
1913 | float * log_data = (float *) malloc(dataLength * sizeof(float)); |
---|
1914 | |
---|
1915 | unsigned char * signs = (unsigned char *) malloc(dataLength); |
---|
1916 | memset(signs, 0, dataLength); |
---|
1917 | // preprocess |
---|
1918 | float max_abs_log_data; |
---|
1919 | if(min == 0) max_abs_log_data = fabs(log2(fabs(max))); |
---|
1920 | else if(max == 0) max_abs_log_data = fabs(log2(fabs(min))); |
---|
1921 | else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max))); |
---|
1922 | float min_log_data = max_abs_log_data; |
---|
1923 | bool positive = true; |
---|
1924 | for(size_t i=0; i<dataLength; i++){ |
---|
1925 | if(oriData[i] < 0){ |
---|
1926 | signs[i] = 1; |
---|
1927 | log_data[i] = -oriData[i]; |
---|
1928 | positive = false; |
---|
1929 | } |
---|
1930 | else |
---|
1931 | log_data[i] = oriData[i]; |
---|
1932 | if(log_data[i] > 0){ |
---|
1933 | log_data[i] = log2(log_data[i]); |
---|
1934 | if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i]; |
---|
1935 | if(log_data[i] < min_log_data) min_log_data = log_data[i]; |
---|
1936 | } |
---|
1937 | } |
---|
1938 | |
---|
1939 | float valueRangeSize, medianValue_f; |
---|
1940 | computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f); |
---|
1941 | if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data); |
---|
1942 | double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7; |
---|
1943 | for(size_t i=0; i<dataLength; i++){ |
---|
1944 | if(oriData[i] == 0){ |
---|
1945 | log_data[i] = min_log_data - 2.0001*realPrecision; |
---|
1946 | } |
---|
1947 | } |
---|
1948 | |
---|
1949 | TightDataPointStorageF* tdps = SZ_compress_float_3D_MDQ(log_data, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f); |
---|
1950 | tdps->minLogValue = min_log_data - 1.0001*realPrecision; |
---|
1951 | free(log_data); |
---|
1952 | if(!positive){ |
---|
1953 | unsigned char * comp_signs; |
---|
1954 | // compress signs |
---|
1955 | unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs); |
---|
1956 | tdps->pwrErrBoundBytes = comp_signs; |
---|
1957 | tdps->pwrErrBoundBytes_size = signSize; |
---|
1958 | } |
---|
1959 | else{ |
---|
1960 | tdps->pwrErrBoundBytes = NULL; |
---|
1961 | tdps->pwrErrBoundBytes_size = 0; |
---|
1962 | } |
---|
1963 | free(signs); |
---|
1964 | |
---|
1965 | convertTDPStoFlatBytes_float(tdps, newByteData, outSize); |
---|
1966 | if(*outSize>dataLength*sizeof(float)) |
---|
1967 | SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize); |
---|
1968 | |
---|
1969 | free_TightDataPointStorageF(tdps); |
---|
1970 | } |
---|