From 9ee2ce3c0d78f4ec0c8dee6bba00cac6e8b2f339 Mon Sep 17 00:00:00 2001
From: Hal Finkel <hfinkel@anl.gov>
Date: Fri, 28 Sep 2018 16:32:55 -0500
Subject: [PATCH] importing new SZ files
---
thirdparty/SZ/COPYRIGHT.txt | 3 +-
.../SZ/sz/include/TightDataPointStorageD.h | 6 +
.../SZ/sz/include/TightDataPointStorageF.h | 5 +
thirdparty/SZ/sz/include/TypeManager.h | 2 +
thirdparty/SZ/sz/include/callZlib.h | 2 +
thirdparty/SZ/sz/include/dataCompression.h | 12 +
thirdparty/SZ/sz/include/pastriD.h | 274 +-
thirdparty/SZ/sz/include/pastriF.h | 274 +-
thirdparty/SZ/sz/include/sz.h | 27 +-
thirdparty/SZ/sz/include/sz_double.h | 6 +
thirdparty/SZ/sz/include/sz_double_pwr.h | 4 +
thirdparty/SZ/sz/include/sz_float.h | 8 +
thirdparty/SZ/sz/include/sz_float_pwr.h | 4 +
thirdparty/SZ/sz/include/szd_double.h | 2 +
thirdparty/SZ/sz/include/szd_double_pwr.h | 4 +
thirdparty/SZ/sz/include/szd_float.h | 5 +
thirdparty/SZ/sz/include/szd_float_pwr.h | 3 +
thirdparty/SZ/sz/include/utility.h | 43 +
thirdparty/SZ/sz/src/ByteToolkit.c | 20 +-
thirdparty/SZ/sz/src/DynamicDoubleArray.c | 2 +-
thirdparty/SZ/sz/src/DynamicFloatArray.c | 2 +-
thirdparty/SZ/sz/src/DynamicIntArray.c | 2 +-
thirdparty/SZ/sz/src/Huffman.c | 17 +-
thirdparty/SZ/sz/src/TightDataPointStorageD.c | 52 +-
thirdparty/SZ/sz/src/TightDataPointStorageF.c | 59 +-
thirdparty/SZ/sz/src/TypeManager.c | 70 +-
thirdparty/SZ/sz/src/callZlib.c | 22 +
thirdparty/SZ/sz/src/conf.c | 55 +-
thirdparty/SZ/sz/src/dataCompression.c | 278 +-
thirdparty/SZ/sz/src/sz.c | 93 +-
thirdparty/SZ/sz/src/sz_double.c | 2121 ++++++++++-
thirdparty/SZ/sz/src/sz_double_pwr.c | 186 +
thirdparty/SZ/sz/src/sz_double_ts.c | 2 +-
thirdparty/SZ/sz/src/sz_float.c | 2875 ++++++++++++++-
thirdparty/SZ/sz/src/sz_float_pwr.c | 188 +
thirdparty/SZ/sz/src/sz_int16.c | 5 +-
thirdparty/SZ/sz/src/sz_int32.c | 5 +-
thirdparty/SZ/sz/src/sz_int64.c | 5 +-
thirdparty/SZ/sz/src/sz_int8.c | 5 +-
thirdparty/SZ/sz/src/sz_uint16.c | 5 +-
thirdparty/SZ/sz/src/sz_uint32.c | 5 +-
thirdparty/SZ/sz/src/sz_uint64.c | 5 +-
thirdparty/SZ/sz/src/sz_uint8.c | 5 +-
thirdparty/SZ/sz/src/szd_double.c | 2801 ++++++++++++++-
thirdparty/SZ/sz/src/szd_double_pwr.c | 74 +
thirdparty/SZ/sz/src/szd_float.c | 3141 ++++++++++++++++-
thirdparty/SZ/sz/src/szd_float_pwr.c | 71 +
thirdparty/SZ/sz/src/szd_int16.c | 9 +-
thirdparty/SZ/sz/src/szd_int32.c | 7 +-
thirdparty/SZ/sz/src/szd_int64.c | 7 +-
thirdparty/SZ/sz/src/szd_int8.c | 7 +-
thirdparty/SZ/sz/src/szd_uint16.c | 7 +-
thirdparty/SZ/sz/src/szd_uint32.c | 7 +-
thirdparty/SZ/sz/src/szd_uint64.c | 7 +-
thirdparty/SZ/sz/src/szd_uint8.c | 7 +-
thirdparty/SZ/sz/src/szf.c | 51 +-
56 files changed, 12399 insertions(+), 565 deletions(-)
create mode 100644 thirdparty/SZ/sz/include/utility.h
diff --git a/thirdparty/SZ/COPYRIGHT.txt b/thirdparty/SZ/COPYRIGHT.txt
index de90efe..8baddd9 100644
--- a/thirdparty/SZ/COPYRIGHT.txt
+++ b/thirdparty/SZ/COPYRIGHT.txt
@@ -1,8 +1,9 @@
Copyright © 2016 , UChicago Argonne, LLC
All Rights Reserved
-[SZ, Version 1.3]
+[SZ, Version 1.4]
Sheng Di
Dingwen Tao
+Xin Liang
Franck Cappello
Argonne National Laboratory
diff --git a/thirdparty/SZ/sz/include/TightDataPointStorageD.h b/thirdparty/SZ/sz/include/TightDataPointStorageD.h
index 4fc5be8..0863fb0 100644
--- a/thirdparty/SZ/sz/include/TightDataPointStorageD.h
+++ b/thirdparty/SZ/sz/include/TightDataPointStorageD.h
@@ -23,6 +23,8 @@ typedef struct TightDataPointStorageD
char reqLength;
char radExpo; //used to compute reqLength based on segmented precisions in "pw_rel_compression"
+ double minLogValue;
+
int stateNum;
int allNodes;
@@ -52,6 +54,10 @@ typedef struct TightDataPointStorageD
unsigned char* pwrErrBoundBytes;
int pwrErrBoundBytes_size;
+
+ unsigned char* raBytes;
+ size_t raBytes_size;
+
} TightDataPointStorageD;
void new_TightDataPointStorageD_Empty(TightDataPointStorageD **self);
diff --git a/thirdparty/SZ/sz/include/TightDataPointStorageF.h b/thirdparty/SZ/sz/include/TightDataPointStorageF.h
index eca1717..7e5df7d 100644
--- a/thirdparty/SZ/sz/include/TightDataPointStorageF.h
+++ b/thirdparty/SZ/sz/include/TightDataPointStorageF.h
@@ -34,6 +34,8 @@ typedef struct TightDataPointStorageF
unsigned char* rtypeArray;
size_t rtypeArray_size;
+ float minLogValue;
+
unsigned char* typeArray; //its size is dataSeriesLength/4 (or xxx/4+1)
size_t typeArray_size;
@@ -55,6 +57,9 @@ typedef struct TightDataPointStorageF
unsigned char* pwrErrBoundBytes;
int pwrErrBoundBytes_size;
+ unsigned char* raBytes;
+ size_t raBytes_size;
+
} TightDataPointStorageF;
void new_TightDataPointStorageF_Empty(TightDataPointStorageF **self);
diff --git a/thirdparty/SZ/sz/include/TypeManager.h b/thirdparty/SZ/sz/include/TypeManager.h
index 6be71f8..7c543d3 100644
--- a/thirdparty/SZ/sz/include/TypeManager.h
+++ b/thirdparty/SZ/sz/include/TypeManager.h
@@ -19,8 +19,10 @@ extern "C" {
//TypeManager.c
size_t convertIntArray2ByteArray_fast_1b(unsigned char* intArray, size_t intArrayLength, unsigned char **result);
+size_t convertIntArray2ByteArray_fast_1b_to_result(unsigned char* intArray, size_t intArrayLength, unsigned char *result);
void convertByteArray2IntArray_fast_1b(size_t intArrayLength, unsigned char* byteArray, size_t byteArrayLength, unsigned char **intArray);
size_t convertIntArray2ByteArray_fast_2b(unsigned char* timeStepType, size_t timeStepTypeLength, unsigned char **result);
+size_t convertIntArray2ByteArray_fast_2b_inplace(unsigned char* timeStepType, size_t timeStepTypeLength, unsigned char *result);
void convertByteArray2IntArray_fast_2b(size_t stepLength, unsigned char* byteArray, size_t byteArrayLength, unsigned char **intArray);
size_t convertIntArray2ByteArray_fast_3b(unsigned char* timeStepType, size_t timeStepTypeLength, unsigned char **result);
void convertByteArray2IntArray_fast_3b(size_t stepLength, unsigned char* byteArray, size_t byteArrayLength, unsigned char **intArray);
diff --git a/thirdparty/SZ/sz/include/callZlib.h b/thirdparty/SZ/sz/include/callZlib.h
index 0622d98..1aede54 100644
--- a/thirdparty/SZ/sz/include/callZlib.h
+++ b/thirdparty/SZ/sz/include/callZlib.h
@@ -19,6 +19,8 @@ extern "C" {
#include <stdio.h>
+int isZlibFormat(unsigned char magic1, unsigned char magic2);
+
//callZlib.c
unsigned long zlib_compress(unsigned char* data, unsigned long dataLength, unsigned char** compressBytes, int level);
unsigned long zlib_compress2(unsigned char* data, unsigned long dataLength, unsigned char** compressBytes, int level);
diff --git a/thirdparty/SZ/sz/include/dataCompression.h b/thirdparty/SZ/sz/include/dataCompression.h
index 1eb0f30..6a2da0b 100644
--- a/thirdparty/SZ/sz/include/dataCompression.h
+++ b/thirdparty/SZ/sz/include/dataCompression.h
@@ -77,6 +77,18 @@ int computeBlockEdgeSize_3D(int segmentSize);
int computeBlockEdgeSize_2D(int segmentSize);
int initRandomAccessBytes(unsigned char* raBytes);
+int generateLossyCoefficients_float(float* oriData, double precision, size_t nbEle, int* reqBytesLength, int* resiBitsLength, float* medianValue, float* decData);
+int compressExactDataArray_float(float* oriData, double precision, size_t nbEle, unsigned char** leadArray, unsigned char** midArray, unsigned char** resiArray,
+int reqLength, int reqBytesLength, int resiBitsLength, float medianValue);
+
+void decompressExactDataArray_float(unsigned char* leadNum, unsigned char* exactMidBytes, unsigned char* residualMidBits, size_t nbEle, int reqLength, float medianValue, float** decData);
+
+int generateLossyCoefficients_double(double* oriData, double precision, size_t nbEle, int* reqBytesLength, int* resiBitsLength, double* medianValue, double* decData);
+int compressExactDataArray_double(double* oriData, double precision, size_t nbEle, unsigned char** leadArray, unsigned char** midArray, unsigned char** resiArray,
+int reqLength, int reqBytesLength, int resiBitsLength, double medianValue);
+
+void decompressExactDataArray_double(unsigned char* leadNum, unsigned char* exactMidBytes, unsigned char* residualMidBits, size_t nbEle, int reqLength, double medianValue, double** decData);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/pastriD.h b/thirdparty/SZ/sz/include/pastriD.h
index 1a881e6..3ee2813 100644
--- a/thirdparty/SZ/sz/include/pastriD.h
+++ b/thirdparty/SZ/sz/include/pastriD.h
@@ -10,11 +10,11 @@ static inline int64_t pastri_double_quantize(double x, double binSize){
half.d=0.5;
- //printf("pastri_double_quantize:\nx=%lf x=0x%lx\n",x,(*((uint64_t *)(&x))));
- //printf("sign(x):0x%lx\n", x);
- //printf("0.5:0x%lx\n", (*((uint64_t *)(&half))));
+// //printf("pastri_double_quantize:\nx=%lf x=0x%lx\n",x,(*((uint64_t *)(&x))));
+// //printf("sign(x):0x%lx\n", x);
+// //printf("0.5:0x%lx\n", (*((uint64_t *)(&half))));
half.ui64 |= (u1.ui64 & (uint64_t)0x8000000000000000);
- //printf("sign(x)*0.5:0x%lx\n", (*((uint64_t *)(&half))));
+// //printf("sign(x)*0.5:0x%lx\n", (*((uint64_t *)(&half))));
return (int64_t)(x + half.d);
}
@@ -26,10 +26,10 @@ static inline void pastri_double_PatternMatch(double*data,pastri_params* p,pastr
bp->nonZeros=0;
int i,sb;
for(i=0;i<p->bSize;i++){
- //printf("data[%d] = %.16lf\n",i,data[i]);//DEBUG
+// //printf("data[%d] = %.16lf\n",i,data[i]);//DEBUG
if(abs_FastD(data[i])>p->usedEb){
bp->nonZeros++;
- //if(DEBUG)printf("data[%d]:%.6e\n",i,data[i]); //DEBUG
+ ////if(DEBUG)printf("data[%d]:%.6e\n",i,data[i]); //DEBUG
}
if(abs_FastD(data[i])>absExt){
absExt=abs_FastD(data[i]);
@@ -42,40 +42,40 @@ static inline void pastri_double_PatternMatch(double*data,pastri_params* p,pastr
double patternExt=data[extIdx];
bp->binSize=2*p->usedEb;
- //if(DEBUG){printf("Extremum : data[%d] = %.6e\n",extIdx,patternExt);} //DEBUG
- //if(DEBUG){printf("patternIdx: %d\n",patternIdx);} //DEBUG
+ ////if(DEBUG){printf("Extremum : data[%d] = %.6e\n",extIdx,patternExt);} //DEBUG
+ ////if(DEBUG){printf("patternIdx: %d\n",patternIdx);} //DEBUG
- //if(DEBUG){for(i=0;i<p->sbSize;i++){printf("pattern[%d]=data[%d]=%.6e Quantized:%d\n",i,patternIdx+i,data[patternIdx+i],pastri_double_quantize(data[patternIdx+i]/binSize) );} }//DEBUG
+ ////if(DEBUG){for(i=0;i<p->sbSize;i++){printf("pattern[%d]=data[%d]=%.6e Quantized:%d\n",i,patternIdx+i,data[patternIdx+i],pastri_double_quantize(data[patternIdx+i]/binSize) );} }//DEBUG
//int64_t *patternQ=(int64_t*)(outBuf+15); //Possible Improvement!
for(i=0;i<p->sbSize;i++){
patternQ[i]=pastri_double_quantize(data[patternIdx+i],bp->binSize);
- if(D_W){printf("patternQ[%d]=%ld\n",i,patternQ[i]);}
+ //if(D_W){printf("patternQ[%d]=%ld\n",i,patternQ[i]);}
}
bp->patternBits=bitsNeeded_double((abs_FastD(patternExt)/bp->binSize)+1)+1;
bp->scaleBits=bp->patternBits;
bp->scalesBinSize=1/(double)(((uint64_t)1<<(bp->scaleBits-1))-1);
- //if(DEBUG){printf("(patternExt/binSize)+1: %.6e\n",(patternExt/binSize)+1);} //DEBUG
- //if(DEBUG){printf("scaleBits=patternBits: %d\n",scaleBits);} //DEBUG
- if(D_W){printf("scalesBinSize: %.6e\n",bp->scalesBinSize);} //DEBUG
+ ////if(DEBUG){printf("(patternExt/binSize)+1: %.6e\n",(patternExt/binSize)+1);} //DEBUG
+ ////if(DEBUG){printf("scaleBits=patternBits: %d\n",scaleBits);} //DEBUG
+ //if(D_W){printf("scalesBinSize: %.6e\n",bp->scalesBinSize);} //DEBUG
//Calculate Scales.
//The index part of the input buffer will be reused to hold Scale, Pattern, etc. values.
int localExtIdx=extIdx%p->sbSize; //Local extremum index. This is not the actual extremum of the current sb, but rather the index that correspond to the global (block) extremum.
//int64_t *scalesQ=(int64_t*)(outBuf+15+p->sbSize*8); //Possible Improvement!
int patternExtZero=(patternExt==0);
- //if(DEBUG){printf("patternExtZero: %d\n",patternExtZero);} //DEBUG
+ ////if(DEBUG){printf("patternExtZero: %d\n",patternExtZero);} //DEBUG
for(sb=0;sb<p->sbNum;sb++){
//scales[sb]=data[sb*p->sbSize+localExtIdx]/patternExt;
//scales[sb]=patternExtZero ? 0 : data[sb*p->sbSize+localExtIdx]/patternExt;
//assert(scales[sb]<=1);
scalesQ[sb]=pastri_double_quantize((patternExtZero ? 0 : data[sb*p->sbSize+localExtIdx]/patternExt),bp->scalesBinSize);
- if(D_W){printf("scalesQ[%d]=%ld\n",sb,scalesQ[sb]);}
+ //if(D_W){printf("scalesQ[%d]=%ld\n",sb,scalesQ[sb]);}
}
- //if(DEBUG){for(i=0;i<p->sbSize;i++){printf("scalesQ[%d]=%ld \n",i,scalesQ[i]);}} //DEBUG
+ ////if(DEBUG){for(i=0;i<p->sbSize;i++){printf("scalesQ[%d]=%ld \n",i,scalesQ[i]);}} //DEBUG
//int64_t *ECQ=(int64_t*)(outBuf+p->bSize*8); //ECQ is written into outBuf, just be careful when handling it.
@@ -92,7 +92,7 @@ static inline void pastri_double_PatternMatch(double*data,pastri_params* p,pastr
double absECQ=abs_FastD(ECQ[_1DIdx]);
if(absECQ > bp->ECQExt)
bp->ECQExt=absECQ;
- //if(DEBUG){printf("EC[%d]: %.6e Quantized:%ld \n",_1DIdx,(scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-data[_1DIdx]),ECQ[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("EC[%d]: %.6e Quantized:%ld \n",_1DIdx,(scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-data[_1DIdx]),ECQ[_1DIdx]);} //DEBUG
switch (ECQ[_1DIdx]){
case 0:
//ECQ0s++; //Currently not needed
@@ -117,8 +117,8 @@ static inline void pastri_double_PatternMatch(double*data,pastri_params* p,pastr
_1DIdx=sb*p->sbSize+i;
double decompressed=scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-ECQ[_1DIdx]*binSize;
if(abs_FastD(decompressed-data[_1DIdx])>(p->usedEb)){
- printf("p->usedEb=%.6e\n",p->usedEb);
- printf("data[%d]=%.6e decompressed[%d]=%.6e diff=%.6e\n",_1DIdx,data[_1DIdx],_1DIdx,decompressed,abs_FastD(data[_1DIdx]-decompressed));
+ //printf("p->usedEb=%.6e\n",p->usedEb);
+ //printf("data[%d]=%.6e decompressed[%d]=%.6e diff=%.6e\n",_1DIdx,data[_1DIdx],_1DIdx,decompressed,abs_FastD(data[_1DIdx]-decompressed));
assert(0);
}
}
@@ -174,8 +174,8 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
//*(uint16_t*)(&outBuf[5])=p->idxOffset[2];
//*(uint16_t*)(&outBuf[7])=p->idxOffset[3];
- if(D_W){printf("ECQ0s:%d ECQ1s:%d ECQOthers:%d Total:%d\n",p->bSize-bp->ECQ1s-bp->ECQOthers,bp->ECQ1s,bp->ECQOthers,p->bSize);} //DEBUG
- if(D_W){printf("numOutliers:%d\n",bp->numOutliers);} //DEBUG
+ //if(D_W){printf("ECQ0s:%d ECQ1s:%d ECQOthers:%d Total:%d\n",p->bSize-bp->ECQ1s-bp->ECQOthers,bp->ECQ1s,bp->ECQOthers,p->bSize);} //DEBUG
+ //if(D_W){printf("numOutliers:%d\n",bp->numOutliers);} //DEBUG
//****************************************************************************************
//if(0){ //DEBUG
@@ -183,8 +183,8 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
if((UCSparseBytes<UCNonSparseBytes) && (UCSparseBytes<CSparseBytes) && (UCSparseBytes<CNonSparseBytes) ){
//Uncompressed, Sparse bits. Just like the original GAMESS data. Includes: mode, indexOffsets, nonZeros, indexes, data
*numOutBytes=UCSparseBytes;
- if(D_G){printf("UCSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ //if(D_G){printf("UCSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
outBuf[0]=0; //mode
//*(uint16_t*)(&outBuf[9])=nonZeros;
@@ -216,7 +216,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
}
}
- if(D_G)printf("UCSparseBytes:%d \n",UCSparseBytes); //DEBUG
+ //if(D_G)printf("UCSparseBytes:%d \n",UCSparseBytes); //DEBUG
//****************************************************************************************
//}else if(0){ //DEBUG
@@ -224,23 +224,23 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
}else if((UCNonSparseBytes<UCSparseBytes) && (UCNonSparseBytes<CSparseBytes) && (UCNonSparseBytes<CNonSparseBytes) ){
//Uncompressed, NonSparse bits. Includes: mode, indexOffsets, data
*numOutBytes=UCNonSparseBytes;
- if(D_G){printf("UCNonSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ //if(D_G){printf("UCNonSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
outBuf[0]=1; //mode
//memcpy(&outBuf[9], &inBuf[p->bSize*8], UCNonSparseBytes-9);
memcpy(&outBuf[1], data, p->bSize*p->dataSize);
- if(D_G)printf("UCNonSparseBytes:%d \n",UCNonSparseBytes); //DEBUG
+ //if(D_G)printf("UCNonSparseBytes:%d \n",UCNonSparseBytes); //DEBUG
/*
for(i=0;i<UCNonSparseBytes-17;i++){
- printf("%d ",inBuf[p->bSize*8+i]);
+ //printf("%d ",inBuf[p->bSize*8+i]);
}
- printf("\n");
+ //printf("\n");
for(i=0;i<UCNonSparseBytes-17;i++){
- printf("%d ",outBuf[17+i]);
+ //printf("%d ",outBuf[17+i]);
}
- printf("\n");
+ //printf("\n");
*/
//****************************************************************************************
//}else if(1){ //DEBUG
@@ -248,9 +248,9 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
}else if((CSparseBytes<UCNonSparseBytes) && (CSparseBytes<UCSparseBytes) && (CSparseBytes<CNonSparseBytes) ){
//Includes: mode, indexOffsets, compressedBytes, patternBits, ECQBits,numOutliers,P, S, {Indexes(Sparse), ECQ}
*numOutBytes=CSparseBytes;
- if(D_G){printf("CSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
- //if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
+ //if(D_G){printf("CSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ ////if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
outBuf[0]=2; //mode
////outBuf bytes [1:8] are indexOffsets, which are already written. outBuf bytes [9:12] are reserved for compressedBytes.
@@ -269,17 +269,17 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
//Now, we are at the end of 9th byte.
bitPos=9*8;
- //if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbSize;i++){
writeBits_Fast(outBuf,&bitPos,bp->patternBits,patternQ[i]);//Pattern point
}
- //if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbNum;i++){
writeBits_Fast(outBuf,&bitPos,bp->scaleBits,scalesQ[i]);//Scale
}
- //if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
- //if(DEBUG)printf("ECQBits:%d\n",ECQBits);
+ ////if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG)printf("ECQBits:%d\n",ECQBits);
switch(bp->ECQBits){
case 2:
for(i=0;i<p->bSize;i++){
@@ -287,7 +287,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
case 0:
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x0\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x0\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2,0x10);
//writeBits_Fast(outBuf,&bitPos,2,0);//0x00
@@ -295,7 +295,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
writeBits_Fast(outBuf,&bitPos,1,0);//0x00
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2,0x11);
//writeBits_Fast(outBuf,&bitPos,2,1);//0x01
@@ -314,7 +314,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
case 0:
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x00\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x00\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,3,0);//0x000
//writeBits_Fast(outBuf,&bitPos,1,0);
@@ -322,7 +322,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
writeBits_Fast(outBuf,&bitPos,1,0);
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,3,1);//0x001
//writeBits_Fast(outBuf,&bitPos,1,0);
@@ -330,7 +330,7 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
writeBits_Fast(outBuf,&bitPos,1,1);
break;
default:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1 0x%lx\n",i,ECQ[i],ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1 0x%lx\n",i,ECQ[i],ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2+ECQBits,((uint64_t)0x11<<ECQBits)|ECQ[i]);
//writeBits_Fast(outBuf,&bitPos,2+ECQBits,(ECQ[i]&((uint64_t)0x00<<ECQBits))|((uint64_t)0x01<<ECQBits));
@@ -343,15 +343,15 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
break;
}
- //if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ ////if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
uint32_t bytePos=(bitPos+7)/8;
//*(uint32_t*)(&outBuf[9])=bytePos;
*(uint32_t*)(&outBuf[1])=bytePos;
- if(D_G)printf("bitPos:%ld CSparseBits:%d bytePos:%d CSparseBytes:%d\n",bitPos,CSparseBits,bytePos,CSparseBytes); //DEBUG
+ //if(D_G)printf("bitPos:%ld CSparseBits:%d bytePos:%d CSparseBytes:%d\n",bitPos,CSparseBits,bytePos,CSparseBytes); //DEBUG
if(D_G){assert(bitPos==CSparseBits);}
//****************************************************************************************
@@ -359,9 +359,9 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
}else {
//Includes: mode, indexOffsets, compressedBytes, patternBits, ECQBits,P, S, {ECQ}
*numOutBytes=CNonSparseBytes;
- if(D_G){printf("CNonSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
- //if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
+ //if(D_G){printf("CNonSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ ////if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
outBuf[0]=3; //mode
////outBuf bytes [1:8] are indexOffsets, which are already written. outBuf bytes [9:12] are reserved for compressedBytes.
@@ -374,33 +374,33 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
outBuf[6]=bp->ECQBits;
bitPos=7*8; //Currently, we are at the end of 7th byte.
- //if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbSize;i++){
writeBits_Fast(outBuf,&bitPos,bp->patternBits,patternQ[i]);//Pattern point
}
- //if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbNum;i++){
writeBits_Fast(outBuf,&bitPos,bp->scaleBits,scalesQ[i]);//Scale
}
- //if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
- //if(DEBUG)printf("ECQBits:%d\n",ECQBits);
+ ////if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG)printf("ECQBits:%d\n",ECQBits);
switch(bp->ECQBits){
case 2:
for(i=0;i<p->bSize;i++){
switch(ECQ[i]){
case 0:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x1\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,1,1);//0x1
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x00\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x00\n",i,ECQ[i]); //DEBUG
//writeBits_Fast(outBuf,&bitPos,2,0);//0x00
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x01\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x01\n",i,ECQ[i]); //DEBUG
//writeBits_Fast(outBuf,&bitPos,2,2); //0x01
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
@@ -412,60 +412,60 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
}
break;
default: //ECQBits>2
- //if(DEBUG) printf("AMG_W1:bitPos:%ld\n",bitPos); //DEBUG
+ ////if(DEBUG) printf("AMG_W1:bitPos:%ld\n",bitPos); //DEBUG
for(i=0;i<p->bSize;i++){
- //if(DEBUG){printf("AMG_W3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG) printf("AMG_W2:bitPos:%ld\n",bitPos); //DEBUG
- //if(DEBUG) printf("ECQ[%d]:%ld\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_W3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG) printf("AMG_W2:bitPos:%ld\n",bitPos); //DEBUG
+ ////if(DEBUG) printf("ECQ[%d]:%ld\n",i,ECQ[i]); //DEBUG
switch(ECQ[i]){
case 0:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
writeBits_Fast(outBuf,&bitPos,1,1); //0x1
//wVal=1; writeBits_Fast(outBuf,&bitPos,1,wVal); //0x1
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x000\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x000\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,3,0); //0x000
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
//wVal=0; writeBits_Fast(outBuf,&bitPos,3,wVal); //0x000
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x001\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x001\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,3,8); //0x001
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
//wVal=8; writeBits_Fast(outBuf,&bitPos,3,wVal); //0x001
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
default:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01 0x%lx\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01 0x%lx\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,2,2); //0x01
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
//wVal=2; writeBits_Fast(outBuf,&bitPos,2,wVal); //0x01
writeBits_Fast(outBuf,&bitPos,bp->ECQBits,ECQ[i]);
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
}
}
break;
}
- //if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ ////if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
@@ -473,20 +473,20 @@ static inline void pastri_double_Encode(double *data,int64_t* patternQ,int64_t*
//*(uint32_t*)(&outBuf[9])=bytePos;
*(uint32_t*)(&outBuf[1])=bytePos;
- if(D_G)printf("bitPos:%ld CNonSparseBits:%d bytePos:%d CNonSparseBytes:%d\n",bitPos,CNonSparseBits,bytePos,CNonSparseBytes); //DEBUG
+ //if(D_G)printf("bitPos:%ld CNonSparseBits:%d bytePos:%d CNonSparseBytes:%d\n",bitPos,CNonSparseBits,bytePos,CNonSparseBytes); //DEBUG
if(D_G){assert(bitPos==CNonSparseBits);}
}
- //for(i=213;i<233;i++)if(DEBUG)printf("AMG_WE:bitPos:%d buffer[%d]=0x%lx\n",i*8,i,*(uint64_t*)(&outBuf[i])); //DEBUG
+ ////for(i=213;i<233;i++)if(DEBUG)printf("AMG_WE:bitPos:%d buffer[%d]=0x%lx\n",i*8,i,*(uint64_t*)(&outBuf[i])); //DEBUG
}
static inline int pastri_double_Compress(unsigned char*inBuf,pastri_params *p,unsigned char*outBuf,int *numOutBytes){
pastri_blockParams bp;
- if(D_G2){printf("Parameters: dataSize:%d\n",p->dataSize);} //DEBUG
- if(D_G2){printf("Parameters: bfs:%d %d %d %d originalEb:%.3e\n",p->bf[0],p->bf[1],p->bf[2],p->bf[3],p->usedEb);} //DEBUG
- if(D_G2){printf("Parameters: idxRanges:%d %d %d %d\n",p->idxRange[0],p->idxRange[1],p->idxRange[2],p->idxRange[3]);} //DEBUG
- if(D_G2){printf("Parameters: sbSize:%d sbNum:%d bSize:%d\n",p->sbSize,p->sbNum,p->bSize); }//DEBUG
+ //if(D_G2){printf("Parameters: dataSize:%d\n",p->dataSize);} //DEBUG
+ //if(D_G2){printf("Parameters: bfs:%d %d %d %d originalEb:%.3e\n",p->bf[0],p->bf[1],p->bf[2],p->bf[3],p->usedEb);} //DEBUG
+ //if(D_G2){printf("Parameters: idxRanges:%d %d %d %d\n",p->idxRange[0],p->idxRange[1],p->idxRange[2],p->idxRange[3]);} //DEBUG
+ //if(D_G2){printf("Parameters: sbSize:%d sbNum:%d bSize:%d\n",p->sbSize,p->sbNum,p->bSize); }//DEBUG
int64_t patternQ[MAX_PS_SIZE];
int64_t scalesQ[MAX_PS_SIZE];
@@ -566,7 +566,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
switch(inBuf[0]){
//R:UCSparse
case 0:
- if(D_G){printf("\nDC:UCSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:UCSparse\n");} //DEBUG
//bp->nonZeros=*(uint16_t*)(&inBuf[9]);
//bytePos=11;
bp->nonZeros=*(uint16_t*)(&inBuf[1]);
@@ -591,19 +591,19 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
data[_1DIdx]=*(double*)(&inBuf[bytePos]);
bytePos+=8;
}
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
break;
//R:UCNonSparse
case 1:
- if(D_G){printf("\nDC:UCNonSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:UCNonSparse\n");} //DEBUG
//memcpy(&outBuf[p->bSize*8], &inBuf[9], p->bSize*8);
memcpy(data, &inBuf[1], p->bSize*8);
bytePos=p->bSize*8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
break;
//R:CSparse
case 2:
- if(D_G){printf("\nDC:CSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:CSparse\n");} //DEBUG
//for(j=0;j<p->bSize;j++){
// data[j]=0;
//}
@@ -614,27 +614,27 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
bp->patternBits=inBuf[5];
bp->ECQBits=inBuf[6];
- if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
+ //if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
//bp->numOutliers=*(uint16_t*)(&inBuf[15]);
//bitPos=17*8;
bp->numOutliers=*(uint16_t*)(&inBuf[7]);
bitPos=9*8;
- if(D_R){printf("bp->numOutliers:%d\n",bp->numOutliers);} //DEBUG
+ //if(D_R){printf("bp->numOutliers:%d\n",bp->numOutliers);} //DEBUG
bp->scalesBinSize=1/(double)(((uint64_t)1<<(bp->patternBits-1))-1);
bp->binSize=p->usedEb*2;
- if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
+ //if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
for(j=0;j<p->sbSize;j++){
patternQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Pattern point
- if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
+ //if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
}
for(j=0;j<p->sbNum;j++){
scalesQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Scale
- if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
+ //if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
}
/* //Splitting
@@ -648,13 +648,13 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
switch(bp->ECQBits){
case 2:
for(j=0;j<bp->numOutliers;j++){
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
ECQTemp=readBits_I64(inBuf,&bitPos,1);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -662,32 +662,32 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
////data[_1DIdx]-=ECQTemp*bp->binSize;//Splitting
ECQ[_1DIdx]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
default: //bp->ECQBits>2
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: bp->ECQBits:%d bp->numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: bp->ECQBits:%d bp->numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
for(j=0;j<bp->numOutliers;j++){
_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
temp=readBits_UI64(inBuf,&bitPos,1);
- //if(DEBUG){printf("temp:%ld\n",temp);} //DEBUG
+ ////if(DEBUG){printf("temp:%ld\n",temp);} //DEBUG
switch(temp){
case 0: //+-1
ECQTemp=readBits_I64(inBuf,&bitPos,1);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
break;
case 1: //Others
ECQTemp=readBits_I64(inBuf,&bitPos,bp->ECQBits);
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
break;
//default:
- // printf("ERROR: Bad 2-bit value: 0x%lx",temp);
+ //// printf("ERROR: Bad 2-bit value: 0x%lx",temp);
// assert(0); //AMG
// break;
}
@@ -695,7 +695,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
//data[_1DIdx]-=ECQTemp*bp->binSize;//Splitting
ECQ[_1DIdx]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
}
@@ -704,7 +704,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
//scalesQ=(int64_t*)(inBuf+15+p->sbSize*8);
bytePos=(bitPos+7)/8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
//STEP 2: PREDICT DATA(Includes INVERSE QUANTIZATION)
pastri_double_PredictData(p,bp,data,patternQ,scalesQ,ECQ);
@@ -712,7 +712,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
break;
//R:CNonSparse
case 3:
- if(D_G){printf("\nDC:CNonSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:CNonSparse\n");} //DEBUG
//for(j=0;j<p->bSize;j++){
// data[j]=0;
@@ -724,7 +724,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
bp->patternBits=inBuf[5];
bp->ECQBits=inBuf[6];
- if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
+ //if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
//bitPos=15*8;
bitPos=7*8;
@@ -732,27 +732,27 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
bp->scalesBinSize=1/(double)(((uint64_t)1<<(bp->patternBits-1))-1);
bp->binSize=p->usedEb*2;
- if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
+ //if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
for(j=0;j<p->sbSize;j++){
patternQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Pattern point
- if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
+ //if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
}
for(j=0;j<p->sbNum;j++){
scalesQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Scale
- if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
+ //if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
}
/* //Splitting
for(j=0;j<p->bSize;j++){
data[j]=scalesQ[j/p->sbSize]*patternQ[j%p->sbSize]*bp->scalesBinSize*bp->binSize;
- //if(DEBUG){printf("DC:PS[%d]=%.6e\n",j,data[j]);}
+ ////if(DEBUG){printf("DC:PS[%d]=%.6e\n",j,data[j]);}
}
*/
switch(bp->ECQBits){
case 2:
for(j=0;j<p->bSize;j++){
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
//_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
temp=readBits_UI64(inBuf,&bitPos,1);
switch(temp){
@@ -768,7 +768,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
break;
}
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -776,39 +776,39 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
//data[j]-=ECQTemp*bp->binSize; //Splitting
ECQ[j]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
default: //bp->ECQBits>2
- //if(DEBUG)printf("AMG_R1:bitPos: %ld\n",bitPos);
+ ////if(DEBUG)printf("AMG_R1:bitPos: %ld\n",bitPos);
for(j=0;j<p->bSize;j++){
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("AMG_R2:bitPos: %ld\n",bitPos);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("AMG_R2:bitPos: %ld\n",bitPos);
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
//_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
temp=readBits_UI64(inBuf,&bitPos,1);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
switch(temp){
case 0:
- //if(DEBUG)printf("Read:0");
+ ////if(DEBUG)printf("Read:0");
temp2=readBits_UI64(inBuf,&bitPos,1);
switch(temp2){
case 0:
- //if(DEBUG)printf("0");
+ ////if(DEBUG)printf("0");
ECQTemp=readBits_I64(inBuf,&bitPos,1);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("R:ECQTemp:%ld\n",ECQTemp);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("R:ECQTemp:%ld\n",ECQTemp);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
case 1:
- //if(DEBUG)printf("1\n");
+ ////if(DEBUG)printf("1\n");
ECQTemp=readBits_I64(inBuf,&bitPos,bp->ECQBits);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
default:
assert(0);
@@ -816,16 +816,16 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
}
break;
case 1:
- //if(DEBUG)printf("Read:1\n");
+ ////if(DEBUG)printf("Read:1\n");
ECQTemp=0;
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
default:
assert(0);
break;
}
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -833,7 +833,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
//data[j]-=ECQTemp*bp->binSize; //Splitting
ECQ[j]=ECQTemp;
- //if(DEBUG){printf("DC:data[%d]:%.6e\n",j,data[j]);} //DEBUG
+ ////if(DEBUG){printf("DC:data[%d]:%.6e\n",j,data[j]);} //DEBUG
}
break;
}
@@ -841,7 +841,7 @@ static inline void pastri_double_Decode(unsigned char*inBuf,pastri_params *p,pas
//patternQ=(int64_t*)(inBuf+15);
//scalesQ=(int64_t*)(inBuf+15+p->sbSize*8);
bytePos=(bitPos+7)/8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
//STEP 2: PREDICT DATA(Includes INVERSE QUANTIZATION)
pastri_double_PredictData(p,bp,data,patternQ,scalesQ,ECQ);
@@ -879,19 +879,19 @@ static inline int pastri_double_Check(unsigned char*inBuf,int dataSize,unsigned
/*
for(i=0;i<p->bSize;i++){
if(idx0[i]!=idx0_dc[i]){
- printf("idx0[%d]=%d != %d=idx0_dc[%d]",i,idx0[i],idx0_dc[i],i);
+ //printf("idx0[%d]=%d != %d=idx0_dc[%d]",i,idx0[i],idx0_dc[i],i);
assert(0);
}
if(idx1[i]!=idx1_dc[i]){
- printf("idx1[%d]=%d != %d=idx1_dc[%d]",i,idx1[i],idx1_dc[i],i);
+ //printf("idx1[%d]=%d != %d=idx1_dc[%d]",i,idx1[i],idx1_dc[i],i);
assert(0);
}
if(idx2[i]!=idx2_dc[i]){
- printf("idx2[%d]=%d != %d=idx2_dc[%d]",i,idx2[i],idx2_dc[i],i);
+ //printf("idx2[%d]=%d != %d=idx2_dc[%d]",i,idx2[i],idx2_dc[i],i);
assert(0);
}
if(idx3[i]!=idx3_dc[i]){
- printf("idx3[%d]=%d != %d=idx3_dc[%d]",i,idx3[i],idx3_dc[i],i);
+ //printf("idx3[%d]=%d != %d=idx3_dc[%d]",i,idx3[i],idx3_dc[i],i);
assert(0);
}
}
@@ -900,7 +900,7 @@ static inline int pastri_double_Check(unsigned char*inBuf,int dataSize,unsigned
//Comparing Data:
for(i=0;i<p->bSize;i++){
if(abs_FastD(data[i]-data_dc[i])>p->usedEb){
- printf("|data[%d]-data_dc[%d]|>originalEb : %.3e - %.3e = %.3e > %.3e\n",i,i,data[i],data_dc[i],abs_FastD(data[i]-data_dc[i]),p->usedEb);
+ //printf("|data[%d]-data_dc[%d]|>originalEb : %.3e - %.3e = %.3e > %.3e\n",i,i,data[i],data_dc[i],abs_FastD(data[i]-data_dc[i]),p->usedEb);
assert(0);
}
}
diff --git a/thirdparty/SZ/sz/include/pastriF.h b/thirdparty/SZ/sz/include/pastriF.h
index 08c9c14..5c1d587 100644
--- a/thirdparty/SZ/sz/include/pastriF.h
+++ b/thirdparty/SZ/sz/include/pastriF.h
@@ -10,11 +10,11 @@ static inline int64_t pastri_float_quantize(float x, float binSize){
half.d=0.5;
- //printf("pastri_float_quantize:\nx=%lf x=0x%lx\n",x,(*((uint64_t *)(&x))));
- //printf("sign(x):0x%lx\n", x);
- //printf("0.5:0x%lx\n", (*((uint64_t *)(&half))));
+ ////printf("pastri_float_quantize:\nx=%lf x=0x%lx\n",x,(*((uint64_t *)(&x))));
+ ////printf("sign(x):0x%lx\n", x);
+ ////printf("0.5:0x%lx\n", (*((uint64_t *)(&half))));
half.ui64 |= (u1.ui64 & (uint64_t)0x8000000000000000);
- //printf("sign(x)*0.5:0x%lx\n", (*((uint64_t *)(&half))));
+ ////printf("sign(x)*0.5:0x%lx\n", (*((uint64_t *)(&half))));
return (int64_t)(x + half.d);
}
@@ -26,10 +26,10 @@ static inline void pastri_float_PatternMatch(float*data,pastri_params* p,pastri_
bp->nonZeros=0;
int i,sb;
for(i=0;i<p->bSize;i++){
- //printf("data[%d] = %.16lf\n",i,data[i]);//DEBUG
+ ////printf("data[%d] = %.16lf\n",i,data[i]);//DEBUG
if(abs_FastD(data[i])>p->usedEb){
bp->nonZeros++;
- //if(DEBUG)printf("data[%d]:%.6e\n",i,data[i]); //DEBUG
+ ////if(DEBUG)printf("data[%d]:%.6e\n",i,data[i]); //DEBUG
}
if(abs_FastD(data[i])>absExt){
absExt=abs_FastD(data[i]);
@@ -42,40 +42,40 @@ static inline void pastri_float_PatternMatch(float*data,pastri_params* p,pastri_
float patternExt=data[extIdx];
bp->binSize=2*p->usedEb;
- //if(DEBUG){printf("Extremum : data[%d] = %.6e\n",extIdx,patternExt);} //DEBUG
- //if(DEBUG){printf("patternIdx: %d\n",patternIdx);} //DEBUG
+ ////if(DEBUG){printf("Extremum : data[%d] = %.6e\n",extIdx,patternExt);} //DEBUG
+ ////if(DEBUG){printf("patternIdx: %d\n",patternIdx);} //DEBUG
- //if(DEBUG){for(i=0;i<p->sbSize;i++){printf("pattern[%d]=data[%d]=%.6e Quantized:%d\n",i,patternIdx+i,data[patternIdx+i],pastri_float_quantize(data[patternIdx+i]/binSize) );} }//DEBUG
+ ////if(DEBUG){for(i=0;i<p->sbSize;i++){printf("pattern[%d]=data[%d]=%.6e Quantized:%d\n",i,patternIdx+i,data[patternIdx+i],pastri_float_quantize(data[patternIdx+i]/binSize) );} }//DEBUG
//int64_t *patternQ=(int64_t*)(outBuf+15); //Possible Improvement!
for(i=0;i<p->sbSize;i++){
patternQ[i]=pastri_float_quantize(data[patternIdx+i],bp->binSize);
- if(D_W){printf("patternQ[%d]=%ld\n",i,patternQ[i]);}
+ //if(D_W){printf("patternQ[%d]=%ld\n",i,patternQ[i]);}
}
bp->patternBits=bitsNeeded_float((abs_FastD(patternExt)/bp->binSize)+1)+1;
bp->scaleBits=bp->patternBits;
bp->scalesBinSize=1/(float)(((uint64_t)1<<(bp->scaleBits-1))-1);
- //if(DEBUG){printf("(patternExt/binSize)+1: %.6e\n",(patternExt/binSize)+1);} //DEBUG
- //if(DEBUG){printf("scaleBits=patternBits: %d\n",scaleBits);} //DEBUG
- if(D_W){printf("scalesBinSize: %.6e\n",bp->scalesBinSize);} //DEBUG
+ ////if(DEBUG){printf("(patternExt/binSize)+1: %.6e\n",(patternExt/binSize)+1);} //DEBUG
+ ////if(DEBUG){printf("scaleBits=patternBits: %d\n",scaleBits);} //DEBUG
+ //if(D_W){printf("scalesBinSize: %.6e\n",bp->scalesBinSize);} //DEBUG
//Calculate Scales.
//The index part of the input buffer will be reused to hold Scale, Pattern, etc. values.
int localExtIdx=extIdx%p->sbSize; //Local extremum index. This is not the actual extremum of the current sb, but rather the index that correspond to the global (block) extremum.
//int64_t *scalesQ=(int64_t*)(outBuf+15+p->sbSize*8); //Possible Improvement!
int patternExtZero=(patternExt==0);
- //if(DEBUG){printf("patternExtZero: %d\n",patternExtZero);} //DEBUG
+ ////if(DEBUG){printf("patternExtZero: %d\n",patternExtZero);} //DEBUG
for(sb=0;sb<p->sbNum;sb++){
//scales[sb]=data[sb*p->sbSize+localExtIdx]/patternExt;
//scales[sb]=patternExtZero ? 0 : data[sb*p->sbSize+localExtIdx]/patternExt;
//assert(scales[sb]<=1);
scalesQ[sb]=pastri_float_quantize((patternExtZero ? 0 : data[sb*p->sbSize+localExtIdx]/patternExt),bp->scalesBinSize);
- if(D_W){printf("scalesQ[%d]=%ld\n",sb,scalesQ[sb]);}
+ //if(D_W){printf("scalesQ[%d]=%ld\n",sb,scalesQ[sb]);}
}
- //if(DEBUG){for(i=0;i<p->sbSize;i++){printf("scalesQ[%d]=%ld \n",i,scalesQ[i]);}} //DEBUG
+ ////if(DEBUG){for(i=0;i<p->sbSize;i++){printf("scalesQ[%d]=%ld \n",i,scalesQ[i]);}} //DEBUG
//int64_t *ECQ=(int64_t*)(outBuf+p->bSize*8); //ECQ is written into outBuf, just be careful when handling it.
@@ -92,7 +92,7 @@ static inline void pastri_float_PatternMatch(float*data,pastri_params* p,pastri_
float absECQ=abs_FastD(ECQ[_1DIdx]);
if(absECQ > bp->ECQExt)
bp->ECQExt=absECQ;
- //if(DEBUG){printf("EC[%d]: %.6e Quantized:%ld \n",_1DIdx,(scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-data[_1DIdx]),ECQ[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("EC[%d]: %.6e Quantized:%ld \n",_1DIdx,(scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-data[_1DIdx]),ECQ[_1DIdx]);} //DEBUG
switch (ECQ[_1DIdx]){
case 0:
//ECQ0s++; //Currently not needed
@@ -117,8 +117,8 @@ static inline void pastri_float_PatternMatch(float*data,pastri_params* p,pastri_
_1DIdx=sb*p->sbSize+i;
float decompressed=scalesQ[sb]*patternQ[i]*scalesBinSize*binSize-ECQ[_1DIdx]*binSize;
if(abs_FastD(decompressed-data[_1DIdx])>(p->usedEb)){
- printf("p->usedEb=%.6e\n",p->usedEb);
- printf("data[%d]=%.6e decompressed[%d]=%.6e diff=%.6e\n",_1DIdx,data[_1DIdx],_1DIdx,decompressed,abs_FastD(data[_1DIdx]-decompressed));
+ //printf("p->usedEb=%.6e\n",p->usedEb);
+ //printf("data[%d]=%.6e decompressed[%d]=%.6e diff=%.6e\n",_1DIdx,data[_1DIdx],_1DIdx,decompressed,abs_FastD(data[_1DIdx]-decompressed));
assert(0);
}
}
@@ -174,8 +174,8 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
//*(uint16_t*)(&outBuf[5])=p->idxOffset[2];
//*(uint16_t*)(&outBuf[7])=p->idxOffset[3];
- if(D_W){printf("ECQ0s:%d ECQ1s:%d ECQOthers:%d Total:%d\n",p->bSize-bp->ECQ1s-bp->ECQOthers,bp->ECQ1s,bp->ECQOthers,p->bSize);} //DEBUG
- if(D_W){printf("numOutliers:%d\n",bp->numOutliers);} //DEBUG
+ //if(D_W){printf("ECQ0s:%d ECQ1s:%d ECQOthers:%d Total:%d\n",p->bSize-bp->ECQ1s-bp->ECQOthers,bp->ECQ1s,bp->ECQOthers,p->bSize);} //DEBUG
+ //if(D_W){printf("numOutliers:%d\n",bp->numOutliers);} //DEBUG
//****************************************************************************************
//if(0){ //DEBUG
@@ -183,8 +183,8 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
if((UCSparseBytes<UCNonSparseBytes) && (UCSparseBytes<CSparseBytes) && (UCSparseBytes<CNonSparseBytes) ){
//Uncompressed, Sparse bits. Just like the original GAMESS data. Includes: mode, indexOffsets, nonZeros, indexes, data
*numOutBytes=UCSparseBytes;
- if(D_G){printf("UCSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ //if(D_G){printf("UCSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
outBuf[0]=0; //mode
//*(uint16_t*)(&outBuf[9])=nonZeros;
@@ -216,7 +216,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
}
}
- if(D_G)printf("UCSparseBytes:%d \n",UCSparseBytes); //DEBUG
+ //if(D_G)printf("UCSparseBytes:%d \n",UCSparseBytes); //DEBUG
//****************************************************************************************
//}else if(0){ //DEBUG
@@ -224,23 +224,23 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
}else if((UCNonSparseBytes<UCSparseBytes) && (UCNonSparseBytes<CSparseBytes) && (UCNonSparseBytes<CNonSparseBytes) ){
//Uncompressed, NonSparse bits. Includes: mode, indexOffsets, data
*numOutBytes=UCNonSparseBytes;
- if(D_G){printf("UCNonSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ //if(D_G){printf("UCNonSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
outBuf[0]=1; //mode
//memcpy(&outBuf[9], &inBuf[p->bSize*8], UCNonSparseBytes-9);
memcpy(&outBuf[1], data, p->bSize*p->dataSize);
- if(D_G)printf("UCNonSparseBytes:%d \n",UCNonSparseBytes); //DEBUG
+ //if(D_G)printf("UCNonSparseBytes:%d \n",UCNonSparseBytes); //DEBUG
/*
for(i=0;i<UCNonSparseBytes-17;i++){
- printf("%d ",inBuf[p->bSize*8+i]);
+ //printf("%d ",inBuf[p->bSize*8+i]);
}
- printf("\n");
+ //printf("\n");
for(i=0;i<UCNonSparseBytes-17;i++){
- printf("%d ",outBuf[17+i]);
+ //printf("%d ",outBuf[17+i]);
}
- printf("\n");
+ //printf("\n");
*/
//****************************************************************************************
//}else if(1){ //DEBUG
@@ -248,9 +248,9 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
}else if((CSparseBytes<UCNonSparseBytes) && (CSparseBytes<UCSparseBytes) && (CSparseBytes<CNonSparseBytes) ){
//Includes: mode, indexOffsets, compressedBytes, patternBits, ECQBits,numOutliers,P, S, {Indexes(Sparse), ECQ}
*numOutBytes=CSparseBytes;
- if(D_G){printf("CSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
- //if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
+ //if(D_G){printf("CSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ ////if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
outBuf[0]=2; //mode
////outBuf bytes [1:8] are indexOffsets, which are already written. outBuf bytes [9:12] are reserved for compressedBytes.
@@ -269,17 +269,17 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
//Now, we are at the end of 9th byte.
bitPos=9*8;
- //if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbSize;i++){
writeBits_Fast(outBuf,&bitPos,bp->patternBits,patternQ[i]);//Pattern point
}
- //if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbNum;i++){
writeBits_Fast(outBuf,&bitPos,bp->scaleBits,scalesQ[i]);//Scale
}
- //if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
- //if(DEBUG)printf("ECQBits:%d\n",ECQBits);
+ ////if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG)printf("ECQBits:%d\n",ECQBits);
switch(bp->ECQBits){
case 2:
for(i=0;i<p->bSize;i++){
@@ -287,7 +287,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
case 0:
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x0\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x0\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2,0x10);
//writeBits_Fast(outBuf,&bitPos,2,0);//0x00
@@ -295,7 +295,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
writeBits_Fast(outBuf,&bitPos,1,0);//0x00
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2,0x11);
//writeBits_Fast(outBuf,&bitPos,2,1);//0x01
@@ -314,7 +314,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
case 0:
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x00\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x00\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,3,0);//0x000
//writeBits_Fast(outBuf,&bitPos,1,0);
@@ -322,7 +322,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
writeBits_Fast(outBuf,&bitPos,1,0);
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,3,1);//0x001
//writeBits_Fast(outBuf,&bitPos,1,0);
@@ -330,7 +330,7 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
writeBits_Fast(outBuf,&bitPos,1,1);
break;
default:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1 0x%lx\n",i,ECQ[i],ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1 0x%lx\n",i,ECQ[i],ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,bp->_1DIdxBits,i);
//writeBits_Fast(outBuf,&bitPos,2+ECQBits,((uint64_t)0x11<<ECQBits)|ECQ[i]);
//writeBits_Fast(outBuf,&bitPos,2+ECQBits,(ECQ[i]&((uint64_t)0x00<<ECQBits))|((uint64_t)0x01<<ECQBits));
@@ -343,15 +343,15 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
break;
}
- //if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ ////if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
uint32_t bytePos=(bitPos+7)/8;
//*(uint32_t*)(&outBuf[9])=bytePos;
*(uint32_t*)(&outBuf[1])=bytePos;
- if(D_G)printf("bitPos:%ld CSparseBits:%d bytePos:%d CSparseBytes:%d\n",bitPos,CSparseBits,bytePos,CSparseBytes); //DEBUG
+ //if(D_G)printf("bitPos:%ld CSparseBits:%d bytePos:%d CSparseBytes:%d\n",bitPos,CSparseBits,bytePos,CSparseBytes); //DEBUG
if(D_G){assert(bitPos==CSparseBits);}
//****************************************************************************************
@@ -359,9 +359,9 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
}else {
//Includes: mode, indexOffsets, compressedBytes, patternBits, ECQBits,P, S, {ECQ}
*numOutBytes=CNonSparseBytes;
- if(D_G){printf("CNonSparse\n");} //DEBUG
- if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
- //if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
+ //if(D_G){printf("CNonSparse\n");} //DEBUG
+ //if(D_G)printf("ECQBits:%d\n",bp->ECQBits); //DEBUG
+ ////if(DEBUG){printf("patternBits:%d _1DIdxBits:%d\n",patternBits,_1DIdxBits);} //DEBUG
outBuf[0]=3; //mode
////outBuf bytes [1:8] are indexOffsets, which are already written. outBuf bytes [9:12] are reserved for compressedBytes.
@@ -374,33 +374,33 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
outBuf[6]=bp->ECQBits;
bitPos=7*8; //Currently, we are at the end of 7th byte.
- //if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_B:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbSize;i++){
writeBits_Fast(outBuf,&bitPos,bp->patternBits,patternQ[i]);//Pattern point
}
- //if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG){printf("bitPos_P:%ld\n",bitPos);} //DEBUG
for(i=0;i<p->sbNum;i++){
writeBits_Fast(outBuf,&bitPos,bp->scaleBits,scalesQ[i]);//Scale
}
- //if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
- //if(DEBUG)printf("ECQBits:%d\n",ECQBits);
+ ////if(DEBUG){printf("bitPos_S:%ld\n",bitPos);} //DEBUG
+ ////if(DEBUG)printf("ECQBits:%d\n",ECQBits);
switch(bp->ECQBits){
case 2:
for(i=0;i<p->bSize;i++){
switch(ECQ[i]){
case 0:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x1\n",i,ECQ[i]); //DEBUG
writeBits_Fast(outBuf,&bitPos,1,1);//0x1
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x00\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x00\n",i,ECQ[i]); //DEBUG
//writeBits_Fast(outBuf,&bitPos,2,0);//0x00
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%d Written:0x01\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%d Written:0x01\n",i,ECQ[i]); //DEBUG
//writeBits_Fast(outBuf,&bitPos,2,2); //0x01
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
@@ -412,60 +412,60 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
}
break;
default: //ECQBits>2
- //if(DEBUG) printf("AMG_W1:bitPos:%ld\n",bitPos); //DEBUG
+ ////if(DEBUG) printf("AMG_W1:bitPos:%ld\n",bitPos); //DEBUG
for(i=0;i<p->bSize;i++){
- //if(DEBUG){printf("AMG_W3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG) printf("AMG_W2:bitPos:%ld\n",bitPos); //DEBUG
- //if(DEBUG) printf("ECQ[%d]:%ld\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_W3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG) printf("AMG_W2:bitPos:%ld\n",bitPos); //DEBUG
+ ////if(DEBUG) printf("ECQ[%d]:%ld\n",i,ECQ[i]); //DEBUG
switch(ECQ[i]){
case 0:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x1\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
writeBits_Fast(outBuf,&bitPos,1,1); //0x1
//wVal=1; writeBits_Fast(outBuf,&bitPos,1,wVal); //0x1
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
case 1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x000\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x000\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,3,0); //0x000
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
//wVal=0; writeBits_Fast(outBuf,&bitPos,3,wVal); //0x000
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
case -1:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x001\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x001\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,3,8); //0x001
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
//wVal=8; writeBits_Fast(outBuf,&bitPos,3,wVal); //0x001
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
default:
- //if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01 0x%lx\n",i,ECQ[i]); //DEBUG
- //if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("Index:%d ECQ:%ld Written:0x01 0x%lx\n",i,ECQ[i]); //DEBUG
+ ////if(DEBUG){printf("AMG_WB3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&outBuf[bitPos/8]));}; //DEBUG
//temp1=bitPos;
//writeBits_Fast(outBuf,&bitPos,2,2); //0x01
writeBits_Fast(outBuf,&bitPos,1,0);
writeBits_Fast(outBuf,&bitPos,1,1);
//wVal=2; writeBits_Fast(outBuf,&bitPos,2,wVal); //0x01
writeBits_Fast(outBuf,&bitPos,bp->ECQBits,ECQ[i]);
- //if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_WA3:bitPos:%ld buffer[%ld]=0x%lx\n",temp1,temp1/8,*(uint64_t*)(&outBuf[temp1/8]));}; //DEBUG
break;
}
}
break;
}
- //if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ ////if(DEBUG){printf("bitPos_E:%ld\n",bitPos);} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: ECQBits:%d numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
@@ -473,20 +473,20 @@ static inline void pastri_float_Encode(float *data,int64_t* patternQ,int64_t* sc
//*(uint32_t*)(&outBuf[9])=bytePos;
*(uint32_t*)(&outBuf[1])=bytePos;
- if(D_G)printf("bitPos:%ld CNonSparseBits:%d bytePos:%d CNonSparseBytes:%d\n",bitPos,CNonSparseBits,bytePos,CNonSparseBytes); //DEBUG
+ //if(D_G)printf("bitPos:%ld CNonSparseBits:%d bytePos:%d CNonSparseBytes:%d\n",bitPos,CNonSparseBits,bytePos,CNonSparseBytes); //DEBUG
if(D_G){assert(bitPos==CNonSparseBits);}
}
- //for(i=213;i<233;i++)if(DEBUG)printf("AMG_WE:bitPos:%d buffer[%d]=0x%lx\n",i*8,i,*(uint64_t*)(&outBuf[i])); //DEBUG
+ ////for(i=213;i<233;i++)if(DEBUG)printf("AMG_WE:bitPos:%d buffer[%d]=0x%lx\n",i*8,i,*(uint64_t*)(&outBuf[i])); //DEBUG
}
static inline int pastri_float_Compress(unsigned char*inBuf,pastri_params *p,unsigned char*outBuf,int *numOutBytes){
pastri_blockParams bp;
- if(D_G2){printf("Parameters: dataSize:%d\n",p->dataSize);} //DEBUG
- if(D_G2){printf("Parameters: bfs:%d %d %d %d originalEb:%.3e\n",p->bf[0],p->bf[1],p->bf[2],p->bf[3],p->usedEb);} //DEBUG
- if(D_G2){printf("Parameters: idxRanges:%d %d %d %d\n",p->idxRange[0],p->idxRange[1],p->idxRange[2],p->idxRange[3]);} //DEBUG
- if(D_G2){printf("Parameters: sbSize:%d sbNum:%d bSize:%d\n",p->sbSize,p->sbNum,p->bSize); }//DEBUG
+ //if(D_G2){printf("Parameters: dataSize:%d\n",p->dataSize);} //DEBUG
+ //if(D_G2){printf("Parameters: bfs:%d %d %d %d originalEb:%.3e\n",p->bf[0],p->bf[1],p->bf[2],p->bf[3],p->usedEb);} //DEBUG
+ //if(D_G2){printf("Parameters: idxRanges:%d %d %d %d\n",p->idxRange[0],p->idxRange[1],p->idxRange[2],p->idxRange[3]);} //DEBUG
+ //if(D_G2){printf("Parameters: sbSize:%d sbNum:%d bSize:%d\n",p->sbSize,p->sbNum,p->bSize); }//DEBUG
int64_t patternQ[MAX_PS_SIZE];
int64_t scalesQ[MAX_PS_SIZE];
@@ -566,7 +566,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
switch(inBuf[0]){
//R:UCSparse
case 0:
- if(D_G){printf("\nDC:UCSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:UCSparse\n");} //DEBUG
//bp->nonZeros=*(uint16_t*)(&inBuf[9]);
//bytePos=11;
bp->nonZeros=*(uint16_t*)(&inBuf[1]);
@@ -591,19 +591,19 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
data[_1DIdx]=*(float*)(&inBuf[bytePos]);
bytePos+=8;
}
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
break;
//R:UCNonSparse
case 1:
- if(D_G){printf("\nDC:UCNonSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:UCNonSparse\n");} //DEBUG
//memcpy(&outBuf[p->bSize*8], &inBuf[9], p->bSize*8);
memcpy(data, &inBuf[1], p->bSize*8);
bytePos=p->bSize*8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
break;
//R:CSparse
case 2:
- if(D_G){printf("\nDC:CSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:CSparse\n");} //DEBUG
//for(j=0;j<p->bSize;j++){
// data[j]=0;
//}
@@ -614,27 +614,27 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
bp->patternBits=inBuf[5];
bp->ECQBits=inBuf[6];
- if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
+ //if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
//bp->numOutliers=*(uint16_t*)(&inBuf[15]);
//bitPos=17*8;
bp->numOutliers=*(uint16_t*)(&inBuf[7]);
bitPos=9*8;
- if(D_R){printf("bp->numOutliers:%d\n",bp->numOutliers);} //DEBUG
+ //if(D_R){printf("bp->numOutliers:%d\n",bp->numOutliers);} //DEBUG
bp->scalesBinSize=1/(float)(((uint64_t)1<<(bp->patternBits-1))-1);
bp->binSize=p->usedEb*2;
- if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
+ //if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
for(j=0;j<p->sbSize;j++){
patternQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Pattern point
- if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
+ //if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
}
for(j=0;j<p->sbNum;j++){
scalesQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Scale
- if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
+ //if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
}
/* //Splitting
@@ -648,13 +648,13 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
switch(bp->ECQBits){
case 2:
for(j=0;j<bp->numOutliers;j++){
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
ECQTemp=readBits_I64(inBuf,&bitPos,1);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -662,32 +662,32 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
////data[_1DIdx]-=ECQTemp*bp->binSize;//Splitting
ECQ[_1DIdx]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
default: //bp->ECQBits>2
- if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: bp->ECQBits:%d bp->numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
+ //if(D_C){if(!((bp->ECQBits>=2)||((bp->ECQBits==1) && (bp->numOutliers==0)))){printf("ERROR: bp->ECQBits:%d bp->numOutliers:%d This should not have happened!\n",bp->ECQBits,bp->numOutliers);assert(0);}} //DEBUG
for(j=0;j<bp->numOutliers;j++){
_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
temp=readBits_UI64(inBuf,&bitPos,1);
- //if(DEBUG){printf("temp:%ld\n",temp);} //DEBUG
+ ////if(DEBUG){printf("temp:%ld\n",temp);} //DEBUG
switch(temp){
case 0: //+-1
ECQTemp=readBits_I64(inBuf,&bitPos,1);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
break;
case 1: //Others
ECQTemp=readBits_I64(inBuf,&bitPos,bp->ECQBits);
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
- //if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(D_R)printf("R:ECQ[%d]: %ld \n",_1DIdx,ECQTemp);
break;
//default:
- // printf("ERROR: Bad 2-bit value: 0x%lx",temp);
+ //// printf("ERROR: Bad 2-bit value: 0x%lx",temp);
// assert(0); //AMG
// break;
}
@@ -695,7 +695,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
//data[_1DIdx]-=ECQTemp*bp->binSize;//Splitting
ECQ[_1DIdx]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
}
@@ -704,7 +704,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
//scalesQ=(int64_t*)(inBuf+15+p->sbSize*8);
bytePos=(bitPos+7)/8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
//STEP 2: PREDICT DATA(Includes INVERSE QUANTIZATION)
pastri_float_PredictData(p,bp,data,patternQ,scalesQ,ECQ);
@@ -712,7 +712,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
break;
//R:CNonSparse
case 3:
- if(D_G){printf("\nDC:CNonSparse\n");} //DEBUG
+ //if(D_G){printf("\nDC:CNonSparse\n");} //DEBUG
//for(j=0;j<p->bSize;j++){
// data[j]=0;
@@ -724,7 +724,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
bp->patternBits=inBuf[5];
bp->ECQBits=inBuf[6];
- if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
+ //if(D_R){printf("bp->patternBits:%d bp->ECQBits:%d bp->_1DIdxBits:%d\n",bp->patternBits,bp->ECQBits,bp->_1DIdxBits);} //DEBUG
//bitPos=15*8;
bitPos=7*8;
@@ -732,27 +732,27 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
bp->scalesBinSize=1/(float)(((uint64_t)1<<(bp->patternBits-1))-1);
bp->binSize=p->usedEb*2;
- if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
+ //if(D_R){printf("bp->scalesBinSize:%.6e bp->binSize:%.6e bp->scalesBinSize*bp->binSize:%.6e\n",bp->scalesBinSize,bp->binSize,bp->scalesBinSize*bp->binSize);} //DEBUG
for(j=0;j<p->sbSize;j++){
patternQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Pattern point
- if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
+ //if(D_R){printf("R:patternQ[%d]=%ld\n",j,patternQ[j]);}
}
for(j=0;j<p->sbNum;j++){
scalesQ[j]=readBits_I64(inBuf,&bitPos,bp->patternBits);//Scale
- if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
+ //if(D_R){printf("R:scalesQ[%d]=%ld\n",j,scalesQ[j]);}
}
/* //Splitting
for(j=0;j<p->bSize;j++){
data[j]=scalesQ[j/p->sbSize]*patternQ[j%p->sbSize]*bp->scalesBinSize*bp->binSize;
- //if(DEBUG){printf("DC:PS[%d]=%.6e\n",j,data[j]);}
+ ////if(DEBUG){printf("DC:PS[%d]=%.6e\n",j,data[j]);}
}
*/
switch(bp->ECQBits){
case 2:
for(j=0;j<p->bSize;j++){
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
//_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
temp=readBits_UI64(inBuf,&bitPos,1);
switch(temp){
@@ -768,7 +768,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
break;
}
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -776,39 +776,39 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
//data[j]-=ECQTemp*bp->binSize; //Splitting
ECQ[j]=ECQTemp;
- //if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
+ ////if(DEBUG){printf("decompressed[%d]:%.6e\n",_1DIdx,data[_1DIdx]);} //DEBUG
}
break;
default: //bp->ECQBits>2
- //if(DEBUG)printf("AMG_R1:bitPos: %ld\n",bitPos);
+ ////if(DEBUG)printf("AMG_R1:bitPos: %ld\n",bitPos);
for(j=0;j<p->bSize;j++){
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("AMG_R2:bitPos: %ld\n",bitPos);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("AMG_R2:bitPos: %ld\n",bitPos);
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
- //if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits));} //DEBUG
+ ////if(DEBUG){printf("readBits_UI64:%ld\n",readBits_I64(inBuf,&bitPos,2));} //DEBUG
//_1DIdx=readBits_UI64(inBuf,&bitPos,bp->_1DIdxBits);
temp=readBits_UI64(inBuf,&bitPos,1);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
switch(temp){
case 0:
- //if(DEBUG)printf("Read:0");
+ ////if(DEBUG)printf("Read:0");
temp2=readBits_UI64(inBuf,&bitPos,1);
switch(temp2){
case 0:
- //if(DEBUG)printf("0");
+ ////if(DEBUG)printf("0");
ECQTemp=readBits_I64(inBuf,&bitPos,1);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("R:ECQTemp:%ld\n",ECQTemp);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("R:ECQTemp:%ld\n",ECQTemp);
ECQTemp= ((ECQTemp<<63)>>63)|(uint64_t)0x1;
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
case 1:
- //if(DEBUG)printf("1\n");
+ ////if(DEBUG)printf("1\n");
ECQTemp=readBits_I64(inBuf,&bitPos,bp->ECQBits);
- //if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG){printf("AMG_R3:bitPos:%ld buffer[%ld]=0x%lx\n",bitPos,bitPos/8,*(uint64_t*)(&inBuf[bitPos/8]));}; //DEBUG
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
default:
assert(0);
@@ -816,16 +816,16 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
}
break;
case 1:
- //if(DEBUG)printf("Read:1\n");
+ ////if(DEBUG)printf("Read:1\n");
ECQTemp=0;
- //if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
+ ////if(DEBUG)printf("R:ECQ[%d]: %ld\n",j,ECQTemp);
break;
default:
assert(0);
break;
}
- //if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
+ ////if(DEBUG){printf("_1DIdx:%ld ECQTemp:0x%ld\n",_1DIdx,ECQTemp);} //DEBUG
//continue;
//sb=_1DIdx/p->sbSize;
//localIdx=_1DIdx%p->sbSize;
@@ -833,7 +833,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
//data[j]-=ECQTemp*bp->binSize; //Splitting
ECQ[j]=ECQTemp;
- //if(DEBUG){printf("DC:data[%d]:%.6e\n",j,data[j]);} //DEBUG
+ ////if(DEBUG){printf("DC:data[%d]:%.6e\n",j,data[j]);} //DEBUG
}
break;
}
@@ -841,7 +841,7 @@ static inline void pastri_float_Decode(unsigned char*inBuf,pastri_params *p,past
//patternQ=(int64_t*)(inBuf+15);
//scalesQ=(int64_t*)(inBuf+15+p->sbSize*8);
bytePos=(bitPos+7)/8;
- if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
+ //if(D_G){printf("\nDC:bytePos:%ld\n",bytePos);} //DEBUG
//STEP 2: PREDICT DATA(Includes INVERSE QUANTIZATION)
pastri_float_PredictData(p,bp,data,patternQ,scalesQ,ECQ);
@@ -879,19 +879,19 @@ static inline int pastri_float_Check(unsigned char*inBuf,int dataSize,unsigned c
/*
for(i=0;i<p->bSize;i++){
if(idx0[i]!=idx0_dc[i]){
- printf("idx0[%d]=%d != %d=idx0_dc[%d]",i,idx0[i],idx0_dc[i],i);
+ //printf("idx0[%d]=%d != %d=idx0_dc[%d]",i,idx0[i],idx0_dc[i],i);
assert(0);
}
if(idx1[i]!=idx1_dc[i]){
- printf("idx1[%d]=%d != %d=idx1_dc[%d]",i,idx1[i],idx1_dc[i],i);
+ //printf("idx1[%d]=%d != %d=idx1_dc[%d]",i,idx1[i],idx1_dc[i],i);
assert(0);
}
if(idx2[i]!=idx2_dc[i]){
- printf("idx2[%d]=%d != %d=idx2_dc[%d]",i,idx2[i],idx2_dc[i],i);
+ //printf("idx2[%d]=%d != %d=idx2_dc[%d]",i,idx2[i],idx2_dc[i],i);
assert(0);
}
if(idx3[i]!=idx3_dc[i]){
- printf("idx3[%d]=%d != %d=idx3_dc[%d]",i,idx3[i],idx3_dc[i],i);
+ //printf("idx3[%d]=%d != %d=idx3_dc[%d]",i,idx3[i],idx3_dc[i],i);
assert(0);
}
}
@@ -900,7 +900,7 @@ static inline int pastri_float_Check(unsigned char*inBuf,int dataSize,unsigned c
//Comparing Data:
for(i=0;i<p->bSize;i++){
if(abs_FastD(data[i]-data_dc[i])>p->usedEb){
- printf("|data[%d]-data_dc[%d]|>originalEb : %.3e - %.3e = %.3e > %.3e\n",i,i,data[i],data_dc[i],abs_FastD(data[i]-data_dc[i]),p->usedEb);
+ //printf("|data[%d]-data_dc[%d]|>originalEb : %.3e - %.3e = %.3e > %.3e\n",i,i,data[i],data_dc[i],abs_FastD(data[i]-data_dc[i]),p->usedEb);
assert(0);
}
}
diff --git a/thirdparty/SZ/sz/include/sz.h b/thirdparty/SZ/sz/include/sz.h
index 31c3260..42f1fc8 100644
--- a/thirdparty/SZ/sz/include/sz.h
+++ b/thirdparty/SZ/sz/include/sz.h
@@ -54,6 +54,7 @@
#include "pastri.h"
#include "sz_float_ts.h"
#include "szd_float_ts.h"
+#include "utility.h"
#ifdef _WIN32
#define PATH_SEPARATOR ';'
@@ -74,11 +75,11 @@ extern "C" {
//typedef long int64_t;
//typedef unsigned long uint64_t;
-#define SZ_VERNUM 0x0140
-#define SZ_VER_MAJOR 1
-#define SZ_VER_MINOR 4
-#define SZ_VER_BUILD 13
-#define SZ_VER_REVISION 5
+#define SZ_VERNUM 0x0200
+#define SZ_VER_MAJOR 2
+#define SZ_VER_MINOR 0
+#define SZ_VER_BUILD 2
+#define SZ_VER_REVISION 0
#define PASTRI 103
#define HZ 102
@@ -130,6 +131,9 @@ extern "C" {
#define SZ_DEFAULT_COMPRESSION 2
#define SZ_TEMPORAL_COMPRESSION 3
+#define SZ_NO_REGRESSION 0
+#define SZ_WITH_LINEAR_REGRESSION 1
+
#define SZ_PWR_MIN_TYPE 0
#define SZ_PWR_AVG_TYPE 1
#define SZ_PWR_MAX_TYPE 2
@@ -151,6 +155,10 @@ extern "C" {
#define MetaDataByteLength 20
#define numOfBufferedSteps 1 //the number of time steps in the buffer
+
+
+#define GZIP_COMPRESSOR 0 //i.e., ZLIB_COMPRSSOR
+#define ZSTD_COMPRESSOR 1
//Note: the following setting should be consistent with stateNum in Huffman.h
//#define intvCapacity 65536
@@ -236,6 +244,7 @@ typedef struct sz_params
unsigned int quantization_intervals;
unsigned int maxRangeRadius;
int sol_ID;// it's always SZ, unless the setting is PASTRI compression mode (./configure --enable-pastri)
+ int losslessCompressor;
int sampleDistance; //2 bytes
float predThreshold; // 2 bytes
int szMode; //* 0 (best speed) or 1 (better compression with Gzip) or 3 temporal-dimension based compression
@@ -278,6 +287,10 @@ typedef struct sz_tsc_metainfo
int currentStep;
char metadata_filename[256];
FILE *metadata_file;
+ unsigned char* bit_array; //sihuan added
+ size_t intersect_size; //sihuan added
+ int64_t* hist_index; //sihuan added: prestep index
+
} sz_tsc_metadata;
extern int versionNumber[4];
@@ -289,6 +302,8 @@ extern int sysEndianType; //*sysEndianType is actually set automatically.
extern sz_params *confparams_cpr;
extern sz_params *confparams_dec;
extern sz_exedata *exe_params;
+extern int sz_with_regression;
+
//------------------------------------------------
extern SZ_VarSet* sz_varset;
extern sz_multisteps *multisteps; //compression based on multiple time steps (time-dimension based compression)
@@ -356,8 +371,6 @@ void filloutDimArray(size_t* dim, size_t r5, size_t r4, size_t r3, size_t r2, si
size_t compute_total_batch_size();
-int isZlibFormat(unsigned char magic1, unsigned char magic2);
-
void SZ_registerVar(char* varName, int dataType, void* data,
int errBoundMode, double absErrBound, double relBoundRatio, double pwRelBoundRatio,
size_t r5, size_t r4, size_t r3, size_t r2, size_t r1);
diff --git a/thirdparty/SZ/sz/include/sz_double.h b/thirdparty/SZ/sz/include/sz_double.h
index b186d12..9962366 100644
--- a/thirdparty/SZ/sz/include/sz_double.h
+++ b/thirdparty/SZ/sz/include/sz_double.h
@@ -75,6 +75,12 @@ size_t r1, size_t r2, size_t r3, size_t s1, size_t s2, size_t s3, size_t e1, siz
TightDataPointStorageD* SZ_compress_double_4D_MDQ_subblock(double *oriData, double realPrecision, double valueRangeSize, double medianValue_d,
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);
+unsigned int optimize_intervals_double_2D_with_freq_and_dense_pos(double *oriData, size_t r1, size_t r2, double realPrecision, double * dense_pos, double * max_freq, double * mean_freq);
+unsigned int optimize_intervals_double_3D_with_freq_and_dense_pos(double *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, double * dense_pos, double * max_freq, double * mean_freq);
+unsigned char * SZ_compress_double_2D_MDQ_nonblocked_with_blocked_regression(double *oriData, size_t r1, size_t r2, double realPrecision, size_t * comp_size);
+unsigned char * SZ_compress_double_3D_MDQ_nonblocked_with_blocked_regression(double *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size);
+
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/sz_double_pwr.h b/thirdparty/SZ/sz/include/sz_double_pwr.h
index ce81629..9cb9978 100644
--- a/thirdparty/SZ/sz/include/sz_double_pwr.h
+++ b/thirdparty/SZ/sz/include/sz_double_pwr.h
@@ -37,6 +37,10 @@ double absErrBound, double relBoundRatio, double pwrErrRatio, double valueRangeS
void SZ_compress_args_double_NoCkRngeNoGzip_1D_pwrgroup(unsigned char** newByteData, double *oriData,
size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio, double valueRangeSize, double medianValue_f, size_t *outSize);
+void SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr_pre_log(unsigned char** newByteData, double *oriData, double globalPrecision, size_t dataLength, size_t *outSize, double min, double max);
+void SZ_compress_args_double_NoCkRngeNoGzip_2D_pwr_pre_log(unsigned char** newByteData, double *oriData, double globalPrecision, size_t r1, size_t r2, size_t *outSize, double min, double max);
+void SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr_pre_log(unsigned char** newByteData, double *oriData, double globalPrecision, size_t r1, size_t r2, size_t r3, size_t *outSize, double min, double max);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/sz_float.h b/thirdparty/SZ/sz/include/sz_float.h
index 6ab9231..d08827c 100644
--- a/thirdparty/SZ/sz/include/sz_float.h
+++ b/thirdparty/SZ/sz/include/sz_float.h
@@ -128,6 +128,14 @@ size_t r1, size_t r2, size_t r3, size_t s1, size_t s2, size_t s3, size_t e1, siz
TightDataPointStorageF* SZ_compress_float_4D_MDQ_subblock(float *oriData, double realPrecision, float valueRangeSize, float medianValue_f,
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);
+
+unsigned int optimize_intervals_float_2D_with_freq_and_dense_pos(float *oriData, size_t r1, size_t r2, double realPrecision, float * dense_pos, float * max_freq, float * mean_freq);
+unsigned int optimize_intervals_float_3D_with_freq_and_dense_pos(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, float * dense_pos, float * max_freq, float * mean_freq);
+
+unsigned char * SZ_compress_float_2D_MDQ_nonblocked_with_blocked_regression(float *oriData, size_t r1, size_t r2, double realPrecision, size_t * comp_size);
+unsigned char * SZ_compress_float_3D_MDQ_nonblocked_with_blocked_regression(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size);
+unsigned char * SZ_compress_float_3D_MDQ_random_access_with_blocked_regression(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/sz_float_pwr.h b/thirdparty/SZ/sz/include/sz_float_pwr.h
index 3ef0e7c..a2432b3 100644
--- a/thirdparty/SZ/sz/include/sz_float_pwr.h
+++ b/thirdparty/SZ/sz/include/sz_float_pwr.h
@@ -44,6 +44,10 @@ double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSi
void SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(unsigned char** newByteData, float *oriData,
size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio, float valueRangeSize, float medianValue_f, size_t *outSize);
+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);
+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);
+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);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/szd_double.h b/thirdparty/SZ/sz/include/szd_double.h
index daf3622..15bb81c 100644
--- a/thirdparty/SZ/sz/include/szd_double.h
+++ b/thirdparty/SZ/sz/include/szd_double.h
@@ -24,6 +24,8 @@ void getSnapshotData_double_1D(double** data, size_t dataSeriesLength, TightData
void getSnapshotData_double_2D(double** data, size_t r1, size_t r2, TightDataPointStorageD* tdps, int errBoundMode);
void getSnapshotData_double_3D(double** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageD* tdps, int errBoundMode);
void getSnapshotData_double_4D(double** data, size_t r1, size_t r2, size_t r3, size_t r4, TightDataPointStorageD* tdps, int errBoundMode);
+void decompressDataSeries_double_2D_nonblocked_with_blocked_regression(double** data, size_t r1, size_t r2, unsigned char* comp_data);
+void decompressDataSeries_double_3D_nonblocked_with_blocked_regression(double** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data);
int SZ_decompress_args_double(double** newData, size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, unsigned char* cmpBytes, size_t cmpSize);
diff --git a/thirdparty/SZ/sz/include/szd_double_pwr.h b/thirdparty/SZ/sz/include/szd_double_pwr.h
index 5d3b257..2756685 100644
--- a/thirdparty/SZ/sz/include/szd_double_pwr.h
+++ b/thirdparty/SZ/sz/include/szd_double_pwr.h
@@ -21,6 +21,10 @@ double* extractRealPrecision_3D_double(size_t R1, size_t R2, size_t R3, int bloc
void decompressDataSeries_double_3D_pwr(double** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageD* tdps);
void decompressDataSeries_double_1D_pwrgroup(double** data, size_t dataSeriesLength, TightDataPointStorageD* tdps);
+void decompressDataSeries_double_1D_pwr_pre_log(double** data, size_t dataSeriesLength, TightDataPointStorageD* tdps);
+void decompressDataSeries_double_2D_pwr_pre_log(double** data, size_t r1, size_t r2, TightDataPointStorageD* tdps);
+void decompressDataSeries_double_3D_pwr_pre_log(double** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageD* tdps);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/szd_float.h b/thirdparty/SZ/sz/include/szd_float.h
index 8aaf42c..b2168fe 100644
--- a/thirdparty/SZ/sz/include/szd_float.h
+++ b/thirdparty/SZ/sz/include/szd_float.h
@@ -31,6 +31,11 @@ size_t decompressDataSeries_float_2D_RA_block(float * data, float mean, size_t d
int SZ_decompress_args_float(float** newData, size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, unsigned char* cmpBytes, size_t cmpSize);
size_t decompressDataSeries_float_3D_RA_block(float * 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, int * type, float * unpredictable_data);
+
+void decompressDataSeries_float_2D_nonblocked_with_blocked_regression(float** data, size_t r1, size_t r2, unsigned char* comp_data);
+void decompressDataSeries_float_3D_nonblocked_with_blocked_regression(float** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data);
+void decompressDataSeries_float_3D_random_access_with_blocked_regression(float** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data);
+
#ifdef __cplusplus
}
#endif
diff --git a/thirdparty/SZ/sz/include/szd_float_pwr.h b/thirdparty/SZ/sz/include/szd_float_pwr.h
index 0907517..7f57b45 100644
--- a/thirdparty/SZ/sz/include/szd_float_pwr.h
+++ b/thirdparty/SZ/sz/include/szd_float_pwr.h
@@ -22,6 +22,9 @@ void decompressDataSeries_float_3D_pwr(float** data, size_t r1, size_t r2, size_
char* decompressGroupIDArray(unsigned char* bytes, size_t dataLength);
void decompressDataSeries_float_1D_pwrgroup(float** data, size_t dataSeriesLength, TightDataPointStorageF* tdps);
+void decompressDataSeries_float_1D_pwr_pre_log(float** data, size_t dataSeriesLength, TightDataPointStorageF* tdps);
+void decompressDataSeries_float_2D_pwr_pre_log(float** data, size_t r1, size_t r2, TightDataPointStorageF* tdps);
+void decompressDataSeries_float_3D_pwr_pre_log(float** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageF* tdps);
#ifdef __cplusplus
}
diff --git a/thirdparty/SZ/sz/include/utility.h b/thirdparty/SZ/sz/include/utility.h
new file mode 100644
index 0000000..b66c141
--- /dev/null
+++ b/thirdparty/SZ/sz/include/utility.h
@@ -0,0 +1,43 @@
+/**
+ * @file utility.h
+ * @author Sheng Di, Sihuan Li
+ * @date July, 2018
+ * @brief Header file for the utility.c.
+ * (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory.
+ * See COPYRIGHT in top-level directory.
+ */
+
+#ifndef _UTILITY_H
+#define _UTILITY_H
+
+#include "sz.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+//sihuan added: use a assistant struct to do sorting and swap that are easy to implement: should
+//consider optimizing the performance later.
+typedef struct sort_ast_particle{
+ int64_t id;
+ float var[6];
+} sort_ast_particle;
+
+int compare_struct(const void* obj1, const void* obj2);//sihuan added: the compare function in the qsort parameter for 2 structures
+void reorder_vars(SZ_VarSet* vset);//sihuan added: reorder the variables increasingly by their index
+size_t intersectAndsort(int64_t* preIndex, size_t preLen, SZ_VarSet* curVar, size_t dataLen, unsigned char* bitarray);
+//sihuan added: find intersection and keep new var sorted by id
+void write_reordered_tofile(SZ_VarSet* curVar, size_t dataLen);
+//sihuan added: write the reordered input to files for further decompression validation
+float calculate_delta_t(size_t size);//sihuan added
+
+int is_lossless_compressed_data(unsigned char* compressedBytes, size_t cmpSize);
+unsigned long sz_lossless_compress(int losslessCompressor, int level, unsigned char* data, unsigned long dataLength, unsigned char** compressBytes);
+unsigned long sz_lossless_decompress(int losslessCompressor, unsigned char* compressBytes, unsigned long cmpSize, unsigned char** oriData, unsigned long targetOriSize);
+unsigned long sz_lossless_decompress65536bytes(int losslessCompressor, unsigned char* compressBytes, unsigned long cmpSize, unsigned char** oriData);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* ----- #ifndef _UTILITY_H ----- */
diff --git a/thirdparty/SZ/sz/src/ByteToolkit.c b/thirdparty/SZ/sz/src/ByteToolkit.c
index 31dbf66..43fd811 100644
--- a/thirdparty/SZ/sz/src/ByteToolkit.c
+++ b/thirdparty/SZ/sz/src/ByteToolkit.c
@@ -430,7 +430,7 @@ long bytesToLong(unsigned char* bytes)
}
//the byte to input is in the big-endian format
-float bytesToFloat(unsigned char* bytes)
+inline float bytesToFloat(unsigned char* bytes)
{
lfloat buf;
memcpy(buf.byte, bytes, 4);
@@ -439,7 +439,7 @@ float bytesToFloat(unsigned char* bytes)
return buf.value;
}
-void floatToBytes(unsigned char *b, float num)
+inline void floatToBytes(unsigned char *b, float num)
{
lfloat buf;
buf.value = num;
@@ -449,7 +449,7 @@ void floatToBytes(unsigned char *b, float num)
}
//the byte to input is in the big-endian format
-double bytesToDouble(unsigned char* bytes)
+inline double bytesToDouble(unsigned char* bytes)
{
ldouble buf;
memcpy(buf.byte, bytes, 8);
@@ -458,7 +458,7 @@ double bytesToDouble(unsigned char* bytes)
return buf.value;
}
-void doubleToBytes(unsigned char *b, double num)
+inline void doubleToBytes(unsigned char *b, double num)
{
ldouble buf;
buf.value = num;
@@ -507,7 +507,7 @@ int extractBytes(unsigned char* byteArray, size_t k, int validLength)
return result;
}
-int getMaskRightCode(int m) {
+inline int getMaskRightCode(int m) {
switch (m) {
case 1:
return 0x01;
@@ -530,16 +530,16 @@ int getMaskRightCode(int m) {
}
}
-int getLeftMovingCode(int kMod8)
+inline int getLeftMovingCode(int kMod8)
{
return getMaskRightCode(8 - kMod8);
}
-int getRightMovingSteps(int kMod8, int resiBitLength) {
+inline int getRightMovingSteps(int kMod8, int resiBitLength) {
return 8 - kMod8 - resiBitLength;
}
-int getRightMovingCode(int kMod8, int resiBitLength)
+inline int getRightMovingCode(int kMod8, int resiBitLength)
{
int rightMovingSteps = 8 - kMod8 - resiBitLength;
if(rightMovingSteps < 0)
@@ -814,7 +814,7 @@ void convertULongArrayToBytes(uint64_t* states, size_t stateLength, unsigned cha
}
-size_t bytesToSize(unsigned char* bytes)
+inline size_t bytesToSize(unsigned char* bytes)
{
size_t result = 0;
if(exe_params->SZ_SIZE_TYPE==4)
@@ -824,7 +824,7 @@ size_t bytesToSize(unsigned char* bytes)
return result;
}
-void sizeToBytes(unsigned char* outBytes, size_t size)
+inline void sizeToBytes(unsigned char* outBytes, size_t size)
{
if(exe_params->SZ_SIZE_TYPE==4)
intToBytes_bigEndian(outBytes, size);//4
diff --git a/thirdparty/SZ/sz/src/DynamicDoubleArray.c b/thirdparty/SZ/sz/src/DynamicDoubleArray.c
index 20eb579..54bbb10 100644
--- a/thirdparty/SZ/sz/src/DynamicDoubleArray.c
+++ b/thirdparty/SZ/sz/src/DynamicDoubleArray.c
@@ -21,7 +21,7 @@ void new_DDA(DynamicDoubleArray **dda, size_t cap) {
void convertDDAtoDoubles(DynamicDoubleArray *dba, double **data)
{
- int size = dba->size;
+ size_t size = dba->size;
if(size>0)
*data = (double*)malloc(size * sizeof(double));
else
diff --git a/thirdparty/SZ/sz/src/DynamicFloatArray.c b/thirdparty/SZ/sz/src/DynamicFloatArray.c
index f775827..1a80a48 100644
--- a/thirdparty/SZ/sz/src/DynamicFloatArray.c
+++ b/thirdparty/SZ/sz/src/DynamicFloatArray.c
@@ -21,7 +21,7 @@ void new_DFA(DynamicFloatArray **dfa, size_t cap) {
void convertDFAtoFloats(DynamicFloatArray *dfa, float **data)
{
- int size = dfa->size;
+ size_t size = dfa->size;
if(size>0)
*data = (float*)malloc(size * sizeof(float));
else
diff --git a/thirdparty/SZ/sz/src/DynamicIntArray.c b/thirdparty/SZ/sz/src/DynamicIntArray.c
index 3196ab9..9b713aa 100644
--- a/thirdparty/SZ/sz/src/DynamicIntArray.c
+++ b/thirdparty/SZ/sz/src/DynamicIntArray.c
@@ -21,7 +21,7 @@ void new_DIA(DynamicIntArray **dia, size_t cap) {
void convertDIAtoInts(DynamicIntArray *dia, unsigned char **data)
{
- int size = dia->size;
+ size_t size = dia->size;
if(size>0)
*data = (unsigned char*)malloc(size * sizeof(char));
else
diff --git a/thirdparty/SZ/sz/src/Huffman.c b/thirdparty/SZ/sz/src/Huffman.c
index d067609..6004090 100644
--- a/thirdparty/SZ/sz/src/Huffman.c
+++ b/thirdparty/SZ/sz/src/Huffman.c
@@ -651,23 +651,26 @@ node reconstruct_HuffTree_from_bytes_anyStates(HuffmanTree *huffmanTree, unsigne
void encode_withTree(HuffmanTree* huffmanTree, int *s, size_t length, unsigned char **out, size_t *outSize)
{
- size_t i, nodeCount = 0;
+ size_t i;
+ int nodeCount = 0;
unsigned char *treeBytes, buffer[4];
init(huffmanTree, s, length);
for (i = 0; i < huffmanTree->stateNum; i++)
- if (huffmanTree->code[i]) nodeCount++;
+ if (huffmanTree->code[i]) nodeCount++;
nodeCount = nodeCount*2-1;
unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree,nodeCount, &treeBytes);
//printf("treeByteSize=%d\n", treeByteSize);
*out = (unsigned char*)malloc(length*sizeof(int)+treeByteSize);
intToBytes_bigEndian(buffer, nodeCount);
memcpy(*out, buffer, 4);
- memcpy(*out+4, treeBytes, treeByteSize);
+ intToBytes_bigEndian(buffer, huffmanTree->stateNum/2); //real number of intervals
+ memcpy(*out+4, buffer, 4);
+ memcpy(*out+8, treeBytes, treeByteSize);
free(treeBytes);
size_t enCodeSize = 0;
- encode(huffmanTree, s, length, *out+4+treeByteSize, &enCodeSize);
- *outSize = 4+treeByteSize+enCodeSize;
+ encode(huffmanTree, s, length, *out+8+treeByteSize, &enCodeSize);
+ *outSize = 8+treeByteSize+enCodeSize;
//unsigned short state[length];
//decode(*out+4+treeByteSize, enCodeSize, qqq[0], state);
@@ -682,7 +685,7 @@ void decode_withTree(HuffmanTree* huffmanTree, unsigned char *s, size_t targetLe
{
size_t encodeStartIndex;
size_t nodeCount = bytesToInt_bigEndian(s);
- node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,s+4, nodeCount);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,s+8, nodeCount);
//sdi: Debug
/* build_code(root, 0, 0, 0);
@@ -701,7 +704,7 @@ void decode_withTree(HuffmanTree* huffmanTree, unsigned char *s, size_t targetLe
encodeStartIndex = 1+2*nodeCount*sizeof(unsigned short)+nodeCount*sizeof(unsigned char)+nodeCount*sizeof(unsigned int);
else
encodeStartIndex = 1+3*nodeCount*sizeof(unsigned int)+nodeCount*sizeof(unsigned char);
- decode(s+4+encodeStartIndex, targetLength, root, out);
+ decode(s+8+encodeStartIndex, targetLength, root, out);
}
void SZ_ReleaseHuffman(HuffmanTree* huffmanTree)
diff --git a/thirdparty/SZ/sz/src/TightDataPointStorageD.c b/thirdparty/SZ/sz/src/TightDataPointStorageD.c
index 6ece9db..f31bf85 100644
--- a/thirdparty/SZ/sz/src/TightDataPointStorageD.c
+++ b/thirdparty/SZ/sz/src/TightDataPointStorageD.c
@@ -46,6 +46,10 @@ void new_TightDataPointStorageD_Empty(TightDataPointStorageD **this)
(*this)->segment_size = 0;
(*this)->pwrErrBoundBytes = NULL;
(*this)->pwrErrBoundBytes_size = 0;
+
+ (*this)->raBytes = NULL;
+ (*this)->raBytes_size = 0;
+
}
int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsigned char* flatBytes, size_t flatBytesLength)
@@ -83,10 +87,13 @@ int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsi
sz_params* params = convertBytesToSZParams(&(flatBytes[index]));
int mode = confparams_dec->szMode;
int predictionMode = confparams_dec->predictionMode;
+ int losslessCompressor = confparams_dec->losslessCompressor;
if(confparams_dec!=NULL)
free(confparams_dec);
confparams_dec = params;
confparams_dec->szMode = mode;
+ confparams_dec->losslessCompressor = losslessCompressor;
+
if(mode==SZ_TEMPORAL_COMPRESSION)
{
confparams_dec->szMode = SZ_TEMPORAL_COMPRESSION;
@@ -94,6 +101,8 @@ int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsi
}
index += MetaDataByteLength;
+ int isRandomAccess = (sameRByte >> 7) & 0x01;
+
unsigned char dsLengthBytes[8];
for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++)
dsLengthBytes[i] = flatBytes[index++];
@@ -121,6 +130,13 @@ int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsi
else
(*this)->allSameData = 0;
+ if(isRandomAccess == 1)
+ {
+ (*this)->raBytes_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE;
+ (*this)->raBytes = &(flatBytes[index]);
+ return errorBoundMode;
+ }
+
int rtype_ = sameRByte & 0x08; //1000
unsigned char byteBuf[8];
@@ -204,12 +220,16 @@ int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsi
{
(*this)->leadNumArray_size = (logicLeadNumBitsNum >> 3) + 1;
}
+
+ int minLogValueSize = 0;
+ if(errorBoundMode>=PW_REL)
+ minLogValueSize = 8;
if ((*this)->rtypeArray != NULL)
{
(*this)->residualMidBits_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE - 4 - radExpoL - segmentL - pwrErrBoundBytesL - 4 - 8 - 1 - 8
- exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - 8 - (*this)->rtypeArray_size
- - (*this)->typeArray_size - (*this)->leadNumArray_size
+ - minLogValueSize - (*this)->typeArray_size - (*this)->leadNumArray_size
- (*this)->exactMidBytes_size - pwrErrBoundBytes_size;
for (i = 0; i < (*this)->rtypeArray_size; i++)
(*this)->rtypeArray[i] = flatBytes[index++];
@@ -217,10 +237,15 @@ int new_TightDataPointStorageD_fromFlatBytes(TightDataPointStorageD **this, unsi
else
{
(*this)->residualMidBits_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE - 4 - radExpoL - segmentL - pwrErrBoundBytesL - 4 - 8 - 1 - 8
- - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - (*this)->typeArray_size
+ - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - minLogValueSize - (*this)->typeArray_size
- (*this)->leadNumArray_size - (*this)->exactMidBytes_size - pwrErrBoundBytes_size;
}
+ if(errorBoundMode >= PW_REL){
+ (*this)->minLogValue = bytesToDouble(&flatBytes[index]);
+ index+=8;
+ }
+
(*this)->typeArray = &flatBytes[index];
//retrieve the number of states (i.e., stateNum)
(*this)->allNodes = bytesToInt_bigEndian((*this)->typeArray); //the first 4 bytes store the stateNum
@@ -422,6 +447,13 @@ void convertTDPStoBytes_double(TightDataPointStorageD* tdps, unsigned char* byte
for(i = 0;i<exe_params->SZ_SIZE_TYPE;i++)//ST
bytes[k++] = exactMidBytesLength[i];
+ if(confparams_cpr->errorBoundMode>=PW_REL)
+ {
+ doubleToBytes(exactMidBytesLength, tdps->minLogValue);
+ for(i = 0;i < 8; i++)
+ bytes[k++] = exactMidBytesLength[i];
+ }
+
memcpy(&(bytes[k]), tdps->typeArray, tdps->typeArray_size);
k += tdps->typeArray_size;
if(confparams_cpr->errorBoundMode>=PW_REL)
@@ -521,6 +553,14 @@ void convertTDPStoBytes_double_reserve(TightDataPointStorageD* tdps, unsigned ch
memcpy(&(bytes[k]), tdps->rtypeArray, tdps->rtypeArray_size);
k += tdps->rtypeArray_size;
+
+ if(confparams_cpr->errorBoundMode>=PW_REL)
+ {
+ doubleToBytes(exactMidBytesLength, tdps->minLogValue);
+ for(i = 0;i < 8; i++)
+ bytes[k++] = exactMidBytesLength[i];
+ }
+
memcpy(&(bytes[k]), tdps->typeArray, tdps->typeArray_size);
k += tdps->typeArray_size;
if(confparams_cpr->errorBoundMode>=PW_REL)
@@ -583,15 +623,19 @@ void convertTDPStoFlatBytes_double(TightDataPointStorageD *tdps, unsigned char**
{
size_t residualMidBitsLength = tdps->residualMidBits == NULL ? 0 : tdps->residualMidBits_size;
size_t segmentL = 0, radExpoL = 0, pwrBoundArrayL = 0;
+
+ int minLogValueSize = 0;
if(confparams_cpr->errorBoundMode>=PW_REL)
{
segmentL = exe_params->SZ_SIZE_TYPE;
radExpoL = 1;
pwrBoundArrayL = 4;
+ minLogValueSize = 8;
}
size_t totalByteLength = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrBoundArrayL + 4 + 8 + 1 + 8
+ exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE
+ + minLogValueSize /*max absolute log value*/
+ tdps->typeArray_size + tdps->leadNumArray_size
+ tdps->exactMidBytes_size + residualMidBitsLength + tdps->pwrErrBoundBytes_size;
@@ -605,16 +649,18 @@ void convertTDPStoFlatBytes_double(TightDataPointStorageD *tdps, unsigned char**
{
size_t residualMidBitsLength = tdps->residualMidBits == NULL ? 0 : tdps->residualMidBits_size;
size_t segmentL = 0, radExpoL = 0, pwrBoundArrayL = 0;
+ int minLogValueSize = 0;
if(confparams_cpr->errorBoundMode>=PW_REL)
{
segmentL = exe_params->SZ_SIZE_TYPE;
radExpoL = 1;
pwrBoundArrayL = 4;
+ minLogValueSize = 8;
}
size_t totalByteLength = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrBoundArrayL + 4 + 8 + 1 + 8
+ exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + 8 + tdps->rtypeArray_size
- + tdps->typeArray_size + tdps->leadNumArray_size
+ + minLogValueSize + tdps->typeArray_size + tdps->leadNumArray_size
+ tdps->exactMidBytes_size + residualMidBitsLength + tdps->pwrErrBoundBytes_size;
sameByte = (unsigned char) (sameByte | 0x08); // 00001000, the 4th bit
diff --git a/thirdparty/SZ/sz/src/TightDataPointStorageF.c b/thirdparty/SZ/sz/src/TightDataPointStorageF.c
index 23a69b8..a30f8d9 100644
--- a/thirdparty/SZ/sz/src/TightDataPointStorageF.c
+++ b/thirdparty/SZ/sz/src/TightDataPointStorageF.c
@@ -46,6 +46,9 @@ void new_TightDataPointStorageF_Empty(TightDataPointStorageF **this)
(*this)->segment_size = 0;
(*this)->pwrErrBoundBytes = NULL;
(*this)->pwrErrBoundBytes_size = 0;
+
+ (*this)->raBytes = NULL;
+ (*this)->raBytes_size = 0;
}
int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsigned char* flatBytes, size_t flatBytesLength)
@@ -81,10 +84,13 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
sz_params* params = convertBytesToSZParams(&(flatBytes[index]));
int mode = confparams_dec->szMode;
int predictionMode = confparams_dec->predictionMode;
+ int losslessCompressor = confparams_dec->losslessCompressor;
if(confparams_dec!=NULL)
free(confparams_dec);
confparams_dec = params;
confparams_dec->szMode = mode;
+ confparams_dec->losslessCompressor = losslessCompressor;
+
if(mode==SZ_TEMPORAL_COMPRESSION)
{
confparams_dec->szMode = SZ_TEMPORAL_COMPRESSION;
@@ -92,6 +98,8 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
}
index += MetaDataByteLength;
+
+ int isRandomAccess = (sameRByte >> 7) & 0x01;
unsigned char dsLengthBytes[8];
for (i = 0; i < exe_params->SZ_SIZE_TYPE; i++)
@@ -117,6 +125,12 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
}
else
(*this)->allSameData = 0;
+ if(isRandomAccess == 1)
+ {
+ (*this)->raBytes_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE;
+ (*this)->raBytes = &(flatBytes[index]);
+ return errorBoundMode;
+ }
int rtype_ = sameRByte & 0x08; //=00001000
unsigned char byteBuf[8];
@@ -165,7 +179,7 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
{
for(i = 0;i<exe_params->SZ_SIZE_TYPE;i++)
byteBuf[i] = flatBytes[index++];
- (*this)->rtypeArray_size = bytesToSize(byteBuf);//(ST)
+ (*this)->rtypeArray_size = bytesToSize(byteBuf);//(ST)
}
else
(*this)->rtypeArray_size = 0;
@@ -199,11 +213,15 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
(*this)->leadNumArray_size = (logicLeadNumBitsNum >> 3) + 1;
}
+ int minLogValueSize = 0;
+ if(errorBoundMode>=PW_REL)
+ minLogValueSize = 4;
+
if ((*this)->rtypeArray != NULL)
{
(*this)->residualMidBits_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE - 4 - radExpoL - segmentL - pwrErrBoundBytesL - 4 - 4 - 1 - 8
- - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - 4 - (*this)->rtypeArray_size
- - (*this)->typeArray_size - (*this)->leadNumArray_size
+ - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - minLogValueSize - exe_params->SZ_SIZE_TYPE - 4 - (*this)->rtypeArray_size
+ - minLogValueSize - (*this)->typeArray_size - (*this)->leadNumArray_size
- (*this)->exactMidBytes_size - pwrErrBoundBytes_size;
for (i = 0; i < (*this)->rtypeArray_size; i++)
(*this)->rtypeArray[i] = flatBytes[index++];
@@ -211,9 +229,15 @@ int new_TightDataPointStorageF_fromFlatBytes(TightDataPointStorageF **this, unsi
else
{
(*this)->residualMidBits_size = flatBytesLength - 3 - 1 - MetaDataByteLength - exe_params->SZ_SIZE_TYPE - 4 - radExpoL - segmentL - pwrErrBoundBytesL - 4 - 4 - 1 - 8
- - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - (*this)->typeArray_size
+ - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - exe_params->SZ_SIZE_TYPE - minLogValueSize - (*this)->typeArray_size
- (*this)->leadNumArray_size - (*this)->exactMidBytes_size - pwrErrBoundBytes_size;
- }
+ }
+
+ if(errorBoundMode>=PW_REL)
+ {
+ (*this)->minLogValue = bytesToFloat(&flatBytes[index]);
+ index+=4;
+ }
(*this)->typeArray = &flatBytes[index];
//retrieve the number of states (i.e., stateNum)
@@ -418,6 +442,13 @@ void convertTDPStoBytes_float(TightDataPointStorageF* tdps, unsigned char* bytes
for(i = 0;i<exe_params->SZ_SIZE_TYPE;i++)//ST
bytes[k++] = exactMidBytesLength[i];
+ if(confparams_cpr->errorBoundMode>=PW_REL)
+ {
+ floatToBytes(exactMidBytesLength, tdps->minLogValue);
+ for(i=0;i<4;i++)
+ bytes[k++] = exactMidBytesLength[i];
+ }
+
memcpy(&(bytes[k]), tdps->typeArray, tdps->typeArray_size);
k += tdps->typeArray_size;
if(confparams_cpr->errorBoundMode>=PW_REL)
@@ -519,6 +550,14 @@ void convertTDPStoBytes_float_reserve(TightDataPointStorageF* tdps, unsigned cha
memcpy(&(bytes[k]), tdps->rtypeArray, tdps->rtypeArray_size);
k += tdps->rtypeArray_size;
+
+ if(confparams_cpr->errorBoundMode>=PW_REL)
+ {
+ floatToBytes(exactMidBytesLength, tdps->minLogValue);
+ for(i=0;i<4;i++)
+ bytes[k++] = exactMidBytesLength[i];
+ }
+
memcpy(&(bytes[k]), tdps->typeArray, tdps->typeArray_size);
k += tdps->typeArray_size;
if(confparams_cpr->errorBoundMode>=PW_REL)
@@ -581,15 +620,17 @@ void convertTDPStoFlatBytes_float(TightDataPointStorageF *tdps, unsigned char**
{
size_t residualMidBitsLength = tdps->residualMidBits == NULL ? 0 : tdps->residualMidBits_size;
size_t segmentL = 0, radExpoL = 0, pwrBoundArrayL = 0;
+ int minLogValueSize = 0;
if(confparams_cpr->errorBoundMode>=PW_REL)
{
segmentL = exe_params->SZ_SIZE_TYPE;
radExpoL = 1;
pwrBoundArrayL = 4;
+ minLogValueSize = 4;
}
size_t totalByteLength = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrBoundArrayL + 4 + 4 + 1 + 8
- + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE
+ + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + minLogValueSize
+ tdps->typeArray_size + tdps->leadNumArray_size
+ tdps->exactMidBytes_size + residualMidBitsLength + tdps->pwrErrBoundBytes_size;
@@ -603,16 +644,18 @@ void convertTDPStoFlatBytes_float(TightDataPointStorageF *tdps, unsigned char**
{
size_t residualMidBitsLength = tdps->residualMidBits == NULL ? 0 : tdps->residualMidBits_size;
size_t segmentL = 0, radExpoL = 0, pwrBoundArrayL = 0;
+ int minLogValueSize = 0;
if(confparams_cpr->errorBoundMode>=PW_REL)
{
segmentL = exe_params->SZ_SIZE_TYPE;
radExpoL = 1;
pwrBoundArrayL = 4;
+ minLogValueSize = 4;
}
size_t totalByteLength = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrBoundArrayL + 4 + 4 + 1 + 8
+ exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + 4 + tdps->rtypeArray_size
- + tdps->typeArray_size + tdps->leadNumArray_size
+ + minLogValueSize + tdps->typeArray_size + tdps->leadNumArray_size
+ tdps->exactMidBytes_size + residualMidBitsLength + tdps->pwrErrBoundBytes_size;
sameByte = (unsigned char) (sameByte | 0x08); // 00001000, the 4th bit
@@ -721,7 +764,7 @@ void convertTDPStoFlatBytes_float_args(TightDataPointStorageF *tdps, unsigned ch
* to free the memory used in the compression
* */
void free_TightDataPointStorageF(TightDataPointStorageF *tdps)
-{
+{
if(tdps->rtypeArray!=NULL)
free(tdps->rtypeArray);
if(tdps->typeArray!=NULL)
diff --git a/thirdparty/SZ/sz/src/TypeManager.c b/thirdparty/SZ/sz/src/TypeManager.c
index 42474fb..638f3cb 100644
--- a/thirdparty/SZ/sz/src/TypeManager.c
+++ b/thirdparty/SZ/sz/src/TypeManager.c
@@ -43,7 +43,33 @@ size_t convertIntArray2ByteArray_fast_1b(unsigned char* intArray, size_t intArra
}
return byteLength;
}
-
+
+size_t convertIntArray2ByteArray_fast_1b_to_result(unsigned char* intArray, size_t intArrayLength, unsigned char *result)
+{
+ size_t byteLength = 0;
+ size_t i, j;
+ if(intArrayLength%8==0)
+ byteLength = intArrayLength/8;
+ else
+ byteLength = intArrayLength/8+1;
+
+ size_t n = 0;
+ int tmp, type;
+ for(i = 0;i<byteLength;i++)
+ {
+ tmp = 0;
+ for(j = 0;j<8&&n<intArrayLength;j++)
+ {
+ type = intArray[n];
+ if(type == 1)
+ tmp = (tmp | (1 << (7-j)));
+ n++;
+ }
+ result[i] = (unsigned char)tmp;
+ }
+ return byteLength;
+}
+
void convertByteArray2IntArray_fast_1b(size_t intArrayLength, unsigned char* byteArray, size_t byteArrayLength, unsigned char **intArray)
{
if(intArrayLength > byteArrayLength*8)
@@ -148,6 +174,46 @@ size_t convertIntArray2ByteArray_fast_2b(unsigned char* timeStepType, size_t tim
return byteLength;
}
+size_t convertIntArray2ByteArray_fast_2b_inplace(unsigned char* timeStepType, size_t timeStepTypeLength, unsigned char *result)
+{
+ size_t i, j, byteLength = 0;
+ if(timeStepTypeLength%4==0)
+ byteLength = timeStepTypeLength*2/8;
+ else
+ byteLength = timeStepTypeLength*2/8+1;
+
+ size_t n = 0;
+ for(i = 0;i<byteLength;i++)
+ {
+ int tmp = 0;
+ for(j = 0;j<4&&n<timeStepTypeLength;j++)
+ {
+ int type = timeStepType[n];
+ switch(type)
+ {
+ case 0:
+
+ break;
+ case 1:
+ tmp = (tmp | (1 << (6-j*2)));
+ break;
+ case 2:
+ tmp = (tmp | (2 << (6-j*2)));
+ break;
+ case 3:
+ tmp = (tmp | (3 << (6-j*2)));
+ break;
+ default:
+ printf("Error: wrong timestep type...: type[%zu]=%d\n", n, type);
+ exit(0);
+ }
+ n++;
+ }
+ result[i] = (unsigned char)tmp;
+ }
+ return byteLength;
+}
+
void convertByteArray2IntArray_fast_2b(size_t stepLength, unsigned char* byteArray, size_t byteArrayLength, unsigned char **intArray)
{
if(stepLength > byteArrayLength*4)
@@ -291,7 +357,7 @@ void convertByteArray2IntArray_fast_3b(size_t stepLength, unsigned char* byteArr
}
}
-int getLeftMovingSteps(size_t k, unsigned char resiBitLength)
+inline int getLeftMovingSteps(size_t k, unsigned char resiBitLength)
{
return 8 - k%8 - resiBitLength;
}
diff --git a/thirdparty/SZ/sz/src/callZlib.c b/thirdparty/SZ/sz/src/callZlib.c
index 0e392b7..4e4bb6f 100644
--- a/thirdparty/SZ/sz/src/callZlib.c
+++ b/thirdparty/SZ/sz/src/callZlib.c
@@ -27,6 +27,25 @@
} \
}
+int isZlibFormat(unsigned char magic1, unsigned char magic2)
+{
+ if(magic1==104&&magic2==5) //DC+BS
+ return 1;
+ if(magic1==104&&magic2==129) //DC+DC
+ return 1;
+ if(magic1==104&&magic2==222) //DC+BC
+ return 1;
+ if(magic1==120&&magic2==1) //BC+BS
+ return 1;
+ if(magic1==120&&magic2==94) //BC+?
+ return 1;
+ if(magic1==120&&magic2==156) //BC+DC
+ return 1;
+ if(magic1==120&&magic2==218) //BC+BS
+ return 1;
+ return 0;
+}
+
/*zlib_compress() is only valid for median-size data compression. */
unsigned long zlib_compress(unsigned char* data, unsigned long dataLength, unsigned char** compressBytes, int level)
{
@@ -195,6 +214,9 @@ unsigned long zlib_compress5(unsigned char* data, unsigned long dataLength, unsi
strm.zfree = Z_NULL;
strm.opaque = Z_NULL;
ret = deflateInit(&strm, level);
+ //int windowBits = 15;
+ //ret = deflateInit2(&strm, level, Z_DEFLATED, windowBits, DEF_MEM_LEVEL, Z_DEFAULT_STRATEGY);//Z_FIXED); //Z_DEFAULT_STRATEGY
+
if (ret != Z_OK)
return ret;
diff --git a/thirdparty/SZ/sz/src/conf.c b/thirdparty/SZ/sz/src/conf.c
index 8e6959d..cc5ccf3 100644
--- a/thirdparty/SZ/sz/src/conf.c
+++ b/thirdparty/SZ/sz/src/conf.c
@@ -102,11 +102,16 @@ int SZ_ReadConf(const char* sz_cfgFile) {
confparams_cpr->sampleDistance = 100;
confparams_cpr->szMode = SZ_BEST_COMPRESSION;
-
- confparams_cpr->gzipMode = 1; //fast mode
+ confparams_cpr->losslessCompressor = ZSTD_COMPRESSOR; //other option: GZIP_COMPRESSOR;
+ if(confparams_cpr->losslessCompressor==ZSTD_COMPRESSOR)
+ confparams_cpr->gzipMode = 3; //fast mode
+ else
+ confparams_cpr->gzipMode = 1; //high speed mode
confparams_cpr->errorBoundMode = PSNR;
confparams_cpr->psnr = 90;
+ confparams_cpr->absErrBound = 1E-4;
+ confparams_cpr->relBoundRatio = 1E-4;
confparams_cpr->pw_relBoundRatio = 1E-3;
confparams_cpr->segment_size = 36;
@@ -114,6 +119,8 @@ int SZ_ReadConf(const char* sz_cfgFile) {
confparams_cpr->pwr_type = SZ_PWR_MIN_TYPE;
confparams_cpr->snapshotCmprStep = 5;
+
+ sz_with_regression = SZ_WITH_LINEAR_REGRESSION;
return SZ_SCES;
}
@@ -212,7 +219,26 @@ int SZ_ReadConf(const char* sz_cfgFile) {
return SZ_NSCS;
}
- modeBuf = iniparser_getstring(ini, "PARAMETER:gzipMode", NULL);
+ modeBuf = iniparser_getstring(ini, "PARAMETER:losslessCompressor", "ZSTD_COMPRESSOR");
+ if(strcmp(modeBuf, "GZIP_COMPRESSOR")==0)
+ confparams_cpr->losslessCompressor = GZIP_COMPRESSOR;
+ else if(strcmp(modeBuf, "ZSTD_COMPRESSOR")==0)
+ confparams_cpr->losslessCompressor = ZSTD_COMPRESSOR;
+ else
+ {
+ printf("[SZ] Error: Wrong losslessCompressor setting (please check sz.config file)\n");\
+ printf("No Such a lossless compressor: %s\n", modeBuf);
+ iniparser_freedict(ini);
+ return SZ_NSCS;
+ }
+
+ modeBuf = iniparser_getstring(ini, "PARAMETER:withLinearRegression", "YES");
+ if(strcmp(modeBuf, "YES")==0 || strcmp(modeBuf, "yes")==0)
+ sz_with_regression = SZ_WITH_LINEAR_REGRESSION;
+ else
+ sz_with_regression = SZ_NO_REGRESSION;
+
+ modeBuf = iniparser_getstring(ini, "PARAMETER:gzipMode", "Gzip_BEST_SPEED");
if(modeBuf==NULL)
{
printf("[SZ] Error: Null Gzip mode setting (please check sz.config file)\n");
@@ -233,6 +259,29 @@ int SZ_ReadConf(const char* sz_cfgFile) {
return SZ_NSCS;
}
+ modeBuf = iniparser_getstring(ini, "PARAMETER:zstdMode", "Zstd_HIGH_SPEED");
+ if(modeBuf==NULL)
+ {
+ printf("[SZ] Error: Null Zstd mode setting (please check sz.config file)\n");
+ iniparser_freedict(ini);
+ return SZ_NSCS;
+ }
+ else if(strcmp(modeBuf, "Zstd_BEST_SPEED")==0)
+ confparams_cpr->gzipMode = 1;
+ else if(strcmp(modeBuf, "Zstd_HIGH_SPEED")==0)
+ confparams_cpr->gzipMode = 3;
+ else if(strcmp(modeBuf, "Zstd_HIGH_COMPRESSION")==0)
+ confparams_cpr->gzipMode = 19;
+ else if(strcmp(modeBuf, "Zstd_BEST_COMPRESSION")==0)
+ confparams_cpr->gzipMode = 22;
+ else if(strcmp(modeBuf, "Zstd_DEFAULT_COMPRESSION")==0)
+ confparams_cpr->gzipMode = 3;
+ else
+ {
+ printf("[SZ] Error: Wrong zstd Mode (please check sz.config file)\n");
+ return SZ_NSCS;
+ }
+
//TODO
confparams_cpr->snapshotCmprStep = (int)iniparser_getint(ini, "PARAMETER:snapshotCmprStep", 5);
diff --git a/thirdparty/SZ/sz/src/dataCompression.c b/thirdparty/SZ/sz/src/dataCompression.c
index 0bb5ce1..212a104 100644
--- a/thirdparty/SZ/sz/src/dataCompression.c
+++ b/thirdparty/SZ/sz/src/dataCompression.c
@@ -66,14 +66,14 @@ long computeRangeSize_int(void* oriData, int dataType, size_t size, int64_t* val
else if(dataType == SZ_UINT32)
{
unsigned int* data = (unsigned int*)oriData;
- int data_;
+ unsigned int data_;
min = data[0], max = min;
computeMinMax(data);
}
else if(dataType == SZ_INT32)
{
int* data = (int*)oriData;
- unsigned int data_;
+ int data_;
min = data[0], max = min;
computeMinMax(data);
}
@@ -595,3 +595,277 @@ int initRandomAccessBytes(unsigned char* raBytes)
return k;
}
+
+//The following functions are float-precision version of dealing with the unpredictable data points
+int generateLossyCoefficients_float(float* oriData, double precision, size_t nbEle, int* reqBytesLength, int* resiBitsLength, float* medianValue, float* decData)
+{
+ float valueRangeSize;
+
+ computeRangeSize_float(oriData, nbEle, &valueRangeSize, medianValue);
+ short radExpo = getExponent_float(valueRangeSize/2);
+
+ int reqLength;
+ computeReqLength_float(precision, radExpo, &reqLength, medianValue);
+
+ *reqBytesLength = reqLength/8;
+ *resiBitsLength = reqLength%8;
+
+ size_t i = 0;
+ for(i = 0;i < nbEle;i++)
+ {
+ float normValue = oriData[i] - *medianValue;
+
+ lfloat lfBuf;
+ lfBuf.value = normValue;
+
+ int ignBytesLength = 32 - reqLength;
+ if(ignBytesLength<0)
+ ignBytesLength = 0;
+
+ lfBuf.ivalue = (lfBuf.ivalue >> ignBytesLength) << ignBytesLength;
+
+ //float tmpValue = lfBuf.value;
+
+ decData[i] = lfBuf.value + *medianValue;
+ }
+ return reqLength;
+}
+
+/**
+ * @param float* oriData: inplace argument (input / output)
+ *
+ * */
+int compressExactDataArray_float(float* oriData, double precision, size_t nbEle, unsigned char** leadArray, unsigned char** midArray, unsigned char** resiArray,
+int reqLength, int reqBytesLength, int resiBitsLength, float medianValue)
+{
+ //allocate memory for coefficient compression arrays
+ DynamicIntArray *exactLeadNumArray;
+ new_DIA(&exactLeadNumArray, DynArrayInitLen);
+ DynamicByteArray *exactMidByteArray;
+ new_DBA(&exactMidByteArray, DynArrayInitLen);
+ DynamicIntArray *resiBitArray;
+ new_DIA(&resiBitArray, DynArrayInitLen);
+ unsigned char preDataBytes[4] = {0,0,0,0};
+
+ //allocate memory for vce and lce
+ FloatValueCompressElement *vce = (FloatValueCompressElement*)malloc(sizeof(FloatValueCompressElement));
+ LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
+
+ size_t i = 0;
+ for(i = 0;i < nbEle;i++)
+ {
+ compressSingleFloatValue(vce, oriData[i], precision, medianValue, reqLength, reqBytesLength, resiBitsLength);
+ updateLossyCompElement_Float(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
+ memcpy(preDataBytes,vce->curBytes,4);
+ addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
+ oriData[i] = vce->data;
+ }
+ convertDIAtoInts(exactLeadNumArray, leadArray);
+ convertDBAtoBytes(exactMidByteArray,midArray);
+ convertDIAtoInts(resiBitArray, resiArray);
+
+ size_t midArraySize = exactMidByteArray->size;
+
+ free(vce);
+ free(lce);
+
+ free_DIA(exactLeadNumArray);
+ free_DBA(exactMidByteArray);
+ free_DIA(resiBitArray);
+
+ return midArraySize;
+}
+
+void decompressExactDataArray_float(unsigned char* leadNum, unsigned char* exactMidBytes, unsigned char* residualMidBits, size_t nbEle, int reqLength, float medianValue, float** decData)
+{
+ *decData = (float*)malloc(nbEle*sizeof(float));
+ size_t i = 0, j = 0, k = 0, l = 0, p = 0, curByteIndex = 0;
+ float exactData = 0;
+ unsigned char preBytes[4] = {0,0,0,0};
+ unsigned char curBytes[4];
+ int resiBits;
+ unsigned char leadingNum;
+
+ int reqBytesLength = reqLength/8;
+ int resiBitsLength = reqLength%8;
+
+ for(i = 0; i<nbEle;i++)
+ {
+ // compute resiBits
+ resiBits = 0;
+ if (resiBitsLength != 0) {
+ int kMod8 = k % 8;
+ int rightMovSteps = getRightMovingSteps(kMod8, resiBitsLength);
+ if (rightMovSteps > 0) {
+ int code = getRightMovingCode(kMod8, resiBitsLength);
+ resiBits = (residualMidBits[p] & code) >> rightMovSteps;
+ } else if (rightMovSteps < 0) {
+ int code1 = getLeftMovingCode(kMod8);
+ int code2 = getRightMovingCode(kMod8, resiBitsLength);
+ int leftMovSteps = -rightMovSteps;
+ rightMovSteps = 8 - leftMovSteps;
+ resiBits = (residualMidBits[p] & code1) << leftMovSteps;
+ p++;
+ resiBits = resiBits
+ | ((residualMidBits[p] & code2) >> rightMovSteps);
+ } else // rightMovSteps == 0
+ {
+ int code = getRightMovingCode(kMod8, resiBitsLength);
+ resiBits = (residualMidBits[p] & code);
+ p++;
+ }
+ k += resiBitsLength;
+ }
+
+ // recover the exact data
+ memset(curBytes, 0, 4);
+ leadingNum = leadNum[l++];
+ memcpy(curBytes, preBytes, leadingNum);
+ for (j = leadingNum; j < reqBytesLength; j++)
+ curBytes[j] = exactMidBytes[curByteIndex++];
+ if (resiBitsLength != 0) {
+ unsigned char resiByte = (unsigned char) (resiBits << (8 - resiBitsLength));
+ curBytes[reqBytesLength] = resiByte;
+ }
+
+ exactData = bytesToFloat(curBytes);
+ (*decData)[i] = exactData + medianValue;
+ memcpy(preBytes,curBytes,4);
+ }
+}
+
+//double-precision version of dealing with unpredictable data points in sz 2.0
+int generateLossyCoefficients_double(double* oriData, double precision, size_t nbEle, int* reqBytesLength, int* resiBitsLength, double* medianValue, double* decData)
+{
+ double valueRangeSize;
+
+ computeRangeSize_double(oriData, nbEle, &valueRangeSize, medianValue);
+ short radExpo = getExponent_double(valueRangeSize/2);
+
+ int reqLength;
+ computeReqLength_double(precision, radExpo, &reqLength, medianValue);
+
+ *reqBytesLength = reqLength/8;
+ *resiBitsLength = reqLength%8;
+
+ size_t i = 0;
+ for(i = 0;i < nbEle;i++)
+ {
+ double normValue = oriData[i] - *medianValue;
+
+ ldouble ldBuf;
+ ldBuf.value = normValue;
+
+ int ignBytesLength = 64 - reqLength;
+ if(ignBytesLength<0)
+ ignBytesLength = 0;
+
+ ldBuf.lvalue = (ldBuf.lvalue >> ignBytesLength) << ignBytesLength;
+
+ decData[i] = ldBuf.value + *medianValue;
+ }
+ return reqLength;
+}
+
+/**
+ * @param double* oriData: inplace argument (input / output)
+ *
+ * */
+int compressExactDataArray_double(double* oriData, double precision, size_t nbEle, unsigned char** leadArray, unsigned char** midArray, unsigned char** resiArray,
+int reqLength, int reqBytesLength, int resiBitsLength, double medianValue)
+{
+ //allocate memory for coefficient compression arrays
+ DynamicIntArray *exactLeadNumArray;
+ new_DIA(&exactLeadNumArray, DynArrayInitLen);
+ DynamicByteArray *exactMidByteArray;
+ new_DBA(&exactMidByteArray, DynArrayInitLen);
+ DynamicIntArray *resiBitArray;
+ new_DIA(&resiBitArray, DynArrayInitLen);
+ unsigned char preDataBytes[8] = {0,0,0,0,0,0,0,0};
+
+ //allocate memory for vce and lce
+ DoubleValueCompressElement *vce = (DoubleValueCompressElement*)malloc(sizeof(DoubleValueCompressElement));
+ LossyCompressionElement *lce = (LossyCompressionElement*)malloc(sizeof(LossyCompressionElement));
+
+ size_t i = 0;
+ for(i = 0;i < nbEle;i++)
+ {
+ compressSingleDoubleValue(vce, oriData[i], precision, medianValue, reqLength, reqBytesLength, resiBitsLength);
+ updateLossyCompElement_Double(vce->curBytes, preDataBytes, reqBytesLength, resiBitsLength, lce);
+ memcpy(preDataBytes,vce->curBytes,8);
+ addExactData(exactMidByteArray, exactLeadNumArray, resiBitArray, lce);
+ oriData[i] = vce->data;
+ }
+ convertDIAtoInts(exactLeadNumArray, leadArray);
+ convertDBAtoBytes(exactMidByteArray,midArray);
+ convertDIAtoInts(resiBitArray, resiArray);
+
+ size_t midArraySize = exactMidByteArray->size;
+
+ free(vce);
+ free(lce);
+
+ free_DIA(exactLeadNumArray);
+ free_DBA(exactMidByteArray);
+ free_DIA(resiBitArray);
+
+ return midArraySize;
+}
+
+void decompressExactDataArray_double(unsigned char* leadNum, unsigned char* exactMidBytes, unsigned char* residualMidBits, size_t nbEle, int reqLength, double medianValue, double** decData)
+{
+ *decData = (double*)malloc(nbEle*sizeof(double));
+ size_t i = 0, j = 0, k = 0, l = 0, p = 0, curByteIndex = 0;
+ double exactData = 0;
+ unsigned char preBytes[8] = {0,0,0,0,0,0,0,0};
+ unsigned char curBytes[8];
+ int resiBits;
+ unsigned char leadingNum;
+
+ int reqBytesLength = reqLength/8;
+ int resiBitsLength = reqLength%8;
+
+ for(i = 0; i<nbEle;i++)
+ {
+ // compute resiBits
+ resiBits = 0;
+ if (resiBitsLength != 0) {
+ int kMod8 = k % 8;
+ int rightMovSteps = getRightMovingSteps(kMod8, resiBitsLength);
+ if (rightMovSteps > 0) {
+ int code = getRightMovingCode(kMod8, resiBitsLength);
+ resiBits = (residualMidBits[p] & code) >> rightMovSteps;
+ } else if (rightMovSteps < 0) {
+ int code1 = getLeftMovingCode(kMod8);
+ int code2 = getRightMovingCode(kMod8, resiBitsLength);
+ int leftMovSteps = -rightMovSteps;
+ rightMovSteps = 8 - leftMovSteps;
+ resiBits = (residualMidBits[p] & code1) << leftMovSteps;
+ p++;
+ resiBits = resiBits
+ | ((residualMidBits[p] & code2) >> rightMovSteps);
+ } else // rightMovSteps == 0
+ {
+ int code = getRightMovingCode(kMod8, resiBitsLength);
+ resiBits = (residualMidBits[p] & code);
+ p++;
+ }
+ k += resiBitsLength;
+ }
+
+ // recover the exact data
+ memset(curBytes, 0, 8);
+ leadingNum = leadNum[l++];
+ memcpy(curBytes, preBytes, leadingNum);
+ for (j = leadingNum; j < reqBytesLength; j++)
+ curBytes[j] = exactMidBytes[curByteIndex++];
+ if (resiBitsLength != 0) {
+ unsigned char resiByte = (unsigned char) (resiBits << (8 - resiBitsLength));
+ curBytes[reqBytesLength] = resiByte;
+ }
+
+ exactData = bytesToDouble(curBytes);
+ (*decData)[i] = exactData + medianValue;
+ memcpy(preBytes,curBytes,8);
+ }
+}
diff --git a/thirdparty/SZ/sz/src/sz.c b/thirdparty/SZ/sz/src/sz.c
index 97cb00d..6cdc35b 100644
--- a/thirdparty/SZ/sz/src/sz.c
+++ b/thirdparty/SZ/sz/src/sz.c
@@ -22,6 +22,7 @@
#include "rw.h"
#include "Huffman.h"
#include "conf.h"
+#include "utility.h"
//#include "CurveFillingCompressStorage.h"
int versionNumber[4] = {SZ_VER_MAJOR,SZ_VER_MINOR,SZ_VER_BUILD,SZ_VER_REVISION};
@@ -36,6 +37,8 @@ sz_params *confparams_dec = NULL; //used for decompression
sz_exedata *exe_params = NULL;
+int sz_with_regression = SZ_WITH_LINEAR_REGRESSION; //SZ_NO_REGRESSION
+
/*following global variables are desgined for time-series based compression*/
/*sz_varset is not used in the single-snapshot data compression*/
SZ_VarSet* sz_varset = NULL;
@@ -69,31 +72,15 @@ int SZ_Init(const char *configFilePath)
int SZ_Init_Params(sz_params *params)
{
- int x = 1;
- char *y = (char*)&x;
- int endianType = BIG_ENDIAN_SYSTEM;
- if(*y==1) endianType = LITTLE_ENDIAN_SYSTEM;
+ SZ_Init(NULL);
- sysEndianType = endianType;
- exe_params->SZ_SIZE_TYPE = sizeof(size_t);
+ if(params->losslessCompressor!=GZIP_COMPRESSOR && params->losslessCompressor!=ZSTD_COMPRESSOR)
+ params->losslessCompressor = ZSTD_COMPRESSOR;
- // set default values
- if(params->max_quant_intervals > 0)
+ if(params->max_quant_intervals > 0)
params->maxRangeRadius = params->max_quant_intervals/2;
- else
- params->max_quant_intervals = params->maxRangeRadius*2;
-
- exe_params->intvCapacity = params->maxRangeRadius*2;
- exe_params->intvRadius = params->maxRangeRadius;
-
- if(params->quantization_intervals>0)
- {
- updateQuantizationInfo(params->quantization_intervals);
- exe_params->optQuantMode = 0;
- }
- else
- exe_params->optQuantMode = 1;
-
+
+ memcpy(confparams_cpr, params, sizeof(sz_params));
if(params->quantization_intervals%2!=0)
{
@@ -101,9 +88,6 @@ int SZ_Init_Params(sz_params *params)
return SZ_NSCS;
}
- confparams_cpr = (sz_params*)malloc(sizeof(sz_params));
- memcpy(confparams_cpr, params, sizeof(sz_params));
-
return SZ_SCES;
}
@@ -535,6 +519,14 @@ sz_metadata* SZ_getMetadata(unsigned char* bytes)
isConstant = sameRByte & 0x01;
//confparams_dec->szMode = (sameRByte & 0x06)>>1;
isLossless = (sameRByte & 0x10)>>4;
+
+ int isRandomAccess = (sameRByte >> 7) & 0x01;
+
+ if(exe_params==NULL)
+ {
+ exe_params = (sz_exedata *)malloc(sizeof(struct sz_exedata));
+ memset(exe_params, 0, sizeof(struct sz_exedata));
+ }
exe_params->SZ_SIZE_TYPE = ((sameRByte & 0x40)>>6)==1?8:4;
sz_params* params = convertBytesToSZParams(&(bytes[index]));
@@ -547,8 +539,8 @@ sz_metadata* SZ_getMetadata(unsigned char* bytes)
index++; //jump to the dataLength info byte address
dataSeriesLength = bytesToSize(&(bytes[index]));// 4 or 8
index += exe_params->SZ_SIZE_TYPE;
- index += 4; //max_quant_intervals
-
+ //index += 4; //max_quant_intervals
+
sz_metadata* metadata = (sz_metadata*)malloc(sizeof(struct sz_metadata));
metadata->versionNumber[0] = versions[0];
@@ -564,19 +556,27 @@ sz_metadata* SZ_getMetadata(unsigned char* bytes)
int defactoNBBins = 0; //real # bins
if(isConstant==0 && isLossless==0)
{
- int radExpoL = 0, segmentL = 0, pwrErrBoundBytesL = 0;
- if(metadata->conf_params->errorBoundMode >= PW_REL)
+ if(isRandomAccess==1)
{
- radExpoL = 1;
- segmentL = exe_params->SZ_SIZE_TYPE;
- pwrErrBoundBytesL = 4;
+ unsigned char* raBytes = &(bytes[index]);
+ defactoNBBins = bytesToInt_bigEndian(raBytes + sizeof(int) + sizeof(double));
}
-
- int offset_typearray = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrErrBoundBytesL + 4 + 4 + 1 + 8
- + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE;
- size_t nodeCount = bytesToInt_bigEndian(bytes+offset_typearray);
- defactoNBBins = (nodeCount+1)/2;
- }
+ else
+ {
+ int radExpoL = 0, segmentL = 0, pwrErrBoundBytesL = 0;
+ if(metadata->conf_params->errorBoundMode >= PW_REL)
+ {
+ radExpoL = 1;
+ segmentL = exe_params->SZ_SIZE_TYPE;
+ pwrErrBoundBytesL = 4;
+ }
+
+ int offset_typearray = 3 + 1 + MetaDataByteLength + exe_params->SZ_SIZE_TYPE + 4 + radExpoL + segmentL + pwrErrBoundBytesL + 4 + (4 + params->dataType*4) + 1 + 8
+ + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + exe_params->SZ_SIZE_TYPE + 4;
+ defactoNBBins = bytesToInt_bigEndian(bytes+offset_typearray);
+ }
+
+ }
metadata->defactoNBBins = defactoNBBins;
return metadata;
@@ -779,23 +779,6 @@ size_t compute_total_batch_size()
return totalSize;
}
-int isZlibFormat(unsigned char magic1, unsigned char magic2)
-{
- if(magic1==104&&magic2==5) //DC+BS
- return 1;
- if(magic1==104&&magic2==129) //DC+DC
- return 1;
- if(magic1==104&&magic2==222) //DC+BC
- return 1;
- if(magic1==120&&magic2==1) //BC+BS
- return 1;
- if(magic1==120&&magic2==156) //BC+DC
- return 1;
- if(magic1==120&&magic2==218) //BC+BS
- return 1;
- return 0;
-}
-
void SZ_registerVar(char* varName, int dataType, void* data,
int errBoundMode, double absErrBound, double relBoundRatio, double pwRelBoundRatio,
size_t r5, size_t r4, size_t r3, size_t r2, size_t r1)
diff --git a/thirdparty/SZ/sz/src/sz_double.c b/thirdparty/SZ/sz/src/sz_double.c
index 51819bd..d8bf875 100644
--- a/thirdparty/SZ/sz/src/sz_double.c
+++ b/thirdparty/SZ/sz/src/sz_double.c
@@ -25,6 +25,7 @@
#include "zlib.h"
#include "rw.h"
#include "sz_double_ts.h"
+#include "utility.h"
unsigned char* SZ_skip_compress_double(double* data, size_t dataLength, size_t* outSize)
{
@@ -328,7 +329,7 @@ size_t dataLength, double realPrecision, double valueRangeSize, double medianVal
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = fabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1516,8 +1517,8 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwrErrRatio)
{
if(errBoundMode>=PW_REL)
{
- //SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr(newByteData, oriData, realPrecision, r1, outSize, min, max);
- SZ_compress_args_double_NoCkRngeNoGzip_1D_pwrgroup(newByteData, oriData, r1, absErr_Bound, relBoundRatio, pwrErrRatio, valueRangeSize, medianValue, outSize);
+ SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr(newByteData, oriData, pwrErrRatio, r1, outSize, min, max);
+ //SZ_compress_args_double_NoCkRngeNoGzip_1D_pwrgroup(newByteData, oriData, r1, absErr_Bound, relBoundRatio, pwrErrRatio, valueRangeSize, medianValue, outSize);
}
else
SZ_compress_args_double_NoCkRngeNoGzip_1D(newByteData, oriData, r1, realPrecision, outSize, valueRangeSize, medianValue);
@@ -1562,7 +1563,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
exit(0);
return SZ_NSCS;
}
- }
+ }
int status = SZ_SCES;
size_t dataLength = computeDataLength(r5,r4,r3,r2,r1);
@@ -1600,13 +1601,12 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
{
if(confparams_cpr->errorBoundMode>=PW_REL)
{
- //SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr(&tmpByteData, oriData, realPrecision, r1, &tmpOutSize, min, max);
- SZ_compress_args_double_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio,
- valueRangeSize, medianValue, &tmpOutSize);
+ SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r1, &tmpOutSize, min, max);
+ //SZ_compress_args_double_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio, valueRangeSize, medianValue, &tmpOutSize);
}
else
#ifdef HAVE_TIMECMPR
- if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
multisteps->compressionType = SZ_compress_args_double_NoCkRngeNoGzip_1D(&tmpByteData, oriData, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
@@ -1616,40 +1616,58 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
if (r3==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_double_NoCkRngeNoGzip_2D_pwr(&tmpByteData, oriData, realPrecision, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_double_NoCkRngeNoGzip_2D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r2, r1, &tmpOutSize, min, max);
else
#ifdef HAVE_TIMECMPR
if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
multisteps->compressionType = SZ_compress_args_double_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
- SZ_compress_args_double_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_double_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_double_2D_MDQ_nonblocked_with_blocked_regression(oriData, r2, r1, realPrecision, &tmpOutSize);
+ }
}
else
if (r4==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r3, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r3, r2, r1, &tmpOutSize, min, max);
else
#ifdef HAVE_TIMECMPR
if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
multisteps->compressionType = SZ_compress_args_double_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
- SZ_compress_args_double_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_double_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_double_3D_MDQ_nonblocked_with_blocked_regression(oriData, r3, r2, r1, realPrecision, &tmpOutSize);
+ }
+
+
}
else
if (r5==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r4*r3, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r4*r3, r2, r1, &tmpOutSize, min, max);
else
#ifdef HAVE_TIMECMPR
if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
multisteps->compressionType = SZ_compress_args_double_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
-#endif
- SZ_compress_args_double_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+#endif
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_double_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_double_3D_MDQ_nonblocked_with_blocked_regression(oriData, r4*r3, r2, r1, realPrecision, &tmpOutSize);
+ }
+
}
else
{
@@ -1665,7 +1683,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
@@ -3121,11 +3139,8 @@ unsigned int optimize_intervals_double_3D_opt(double *oriData, size_t r1, size_t
if(radiusIndex>=confparams_cpr->maxRangeRadius)
{
radiusIndex = confparams_cpr->maxRangeRadius - 1;
- //printf("radiusIndex=%d\n", radiusIndex);
}
intervals[radiusIndex]++;
- // printf("TEST: %ld, i: %ld\tj: %ld\tk: %ld\n", data_pos - oriData);
- // fflush(stdout);
offset_count += confparams_cpr->sampleDistance;
if(offset_count >= r3){
n2_count ++;
@@ -3141,9 +3156,6 @@ unsigned int optimize_intervals_double_3D_opt(double *oriData, size_t r1, size_t
}
else data_pos += confparams_cpr->sampleDistance;
}
- // printf("sample_count: %ld\n", sample_count);
- // fflush(stdout);
- // if(*max_freq < 0.15) *max_freq *= 2;
//compute the appropriate number
size_t targetCount = totalSampleSize*confparams_cpr->predThreshold;
size_t sum = 0;
@@ -3161,7 +3173,6 @@ unsigned int optimize_intervals_double_3D_opt(double *oriData, size_t r1, size_t
if(powerOf2<32)
powerOf2 = 32;
free(intervals);
- //printf("targetCount=%d, sum=%d, totalSampleSize=%d, ratio=%f, accIntervals=%d, powerOf2=%d\n", targetCount, sum, totalSampleSize, (double)sum/(double)totalSampleSize, accIntervals, powerOf2);
return powerOf2;
}
@@ -3172,7 +3183,7 @@ unsigned int optimize_intervals_double_2D_opt(double *oriData, size_t r1, size_t
double pred_value = 0, pred_err;
size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
- size_t totalSampleSize = 0;//(r1-1)*(r2-1)/confparams_cpr->sampleDistance;
+ size_t totalSampleSize = 0;
size_t offset_count = confparams_cpr->sampleDistance - 1; // count r2 offset
size_t offset_count_2;
@@ -3226,12 +3237,11 @@ unsigned int optimize_intervals_double_1D_opt(double *oriData, size_t dataLength
double pred_value = 0, pred_err;
size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
- size_t totalSampleSize = 0;//dataLength/confparams_cpr->sampleDistance;
+ size_t totalSampleSize = 0;
double * data_pos = oriData + 2;
while(data_pos - oriData < dataLength){
totalSampleSize++;
- //pred_value = 2*data_pos[-1] - data_pos[-2];
pred_value = data_pos[-1];
pred_err = fabs(pred_value - *data_pos);
radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
@@ -3260,6 +3270,2063 @@ unsigned int optimize_intervals_double_1D_opt(double *oriData, size_t dataLength
powerOf2 = 32;
free(intervals);
- //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2);
return powerOf2;
}
+
+/*The above code is for sz 1.4.13; the following code is for sz 2.0*/
+unsigned int optimize_intervals_double_2D_with_freq_and_dense_pos(double *oriData, size_t r1, size_t r2, double realPrecision, double * dense_pos, double * max_freq, double * mean_freq)
+{
+ double mean = 0.0;
+ size_t len = r1 * r2;
+ size_t mean_distance = (int) (sqrt(len));
+
+ double * data_pos = oriData;
+ size_t mean_count = 0;
+ while(data_pos - oriData < len){
+ mean += *data_pos;
+ mean_count ++;
+ data_pos += mean_distance;
+ }
+ if(mean_count > 0) mean /= mean_count;
+ size_t range = 8192;
+ size_t radius = 4096;
+ size_t * freq_intervals = (size_t *) malloc(range*sizeof(size_t));
+ memset(freq_intervals, 0, range*sizeof(size_t));
+
+ unsigned int maxRangeRadius = confparams_cpr->maxRangeRadius;
+ int sampleDistance = confparams_cpr->sampleDistance;
+ double predThreshold = confparams_cpr->predThreshold;
+
+ size_t i;
+ size_t radiusIndex;
+ double pred_value = 0, pred_err;
+ size_t *intervals = (size_t*)malloc(maxRangeRadius*sizeof(size_t));
+ memset(intervals, 0, maxRangeRadius*sizeof(size_t));
+
+ double mean_diff;
+ ptrdiff_t freq_index;
+ size_t freq_count = 0;
+ size_t n1_count = 1;
+ size_t offset_count = sampleDistance - 1;
+ size_t offset_count_2 = 0;
+ size_t sample_count = 0;
+ data_pos = oriData + r2 + offset_count;
+ while(data_pos - oriData < len){
+ pred_value = data_pos[-1] + data_pos[-r2] - data_pos[-r2-1];
+ pred_err = fabs(pred_value - *data_pos);
+ if(pred_err < realPrecision) freq_count ++;
+ radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
+ if(radiusIndex>=maxRangeRadius)
+ radiusIndex = maxRangeRadius - 1;
+ intervals[radiusIndex]++;
+
+ mean_diff = *data_pos - mean;
+ if(mean_diff > 0) freq_index = (ptrdiff_t)(mean_diff/realPrecision) + radius;
+ else freq_index = (ptrdiff_t)(mean_diff/realPrecision) - 1 + radius;
+ if(freq_index <= 0){
+ freq_intervals[0] ++;
+ }
+ else if(freq_index >= range){
+ freq_intervals[range - 1] ++;
+ }
+ else{
+ freq_intervals[freq_index] ++;
+ }
+ offset_count += sampleDistance;
+ if(offset_count >= r2){
+ n1_count ++;
+ offset_count_2 = n1_count % sampleDistance;
+ data_pos += (r2 + sampleDistance - offset_count) + (sampleDistance - offset_count_2);
+ offset_count = (sampleDistance - offset_count_2);
+ if(offset_count == 0) offset_count ++;
+ }
+ else data_pos += sampleDistance;
+ sample_count ++;
+ }
+ *max_freq = freq_count * 1.0/ sample_count;
+
+ //compute the appropriate number
+ size_t targetCount = sample_count*predThreshold;
+ size_t sum = 0;
+ for(i=0;i<maxRangeRadius;i++)
+ {
+ sum += intervals[i];
+ if(sum>targetCount)
+ break;
+ }
+ if(i>=maxRangeRadius)
+ i = maxRangeRadius-1;
+ unsigned int accIntervals = 2*(i+1);
+ unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
+
+ if(powerOf2<32)
+ powerOf2 = 32;
+
+ // collect frequency
+ size_t max_sum = 0;
+ size_t max_index = 0;
+ size_t tmp_sum;
+ size_t * freq_pos = freq_intervals + 1;
+ for(size_t i=1; i<range-2; i++){
+ tmp_sum = freq_pos[0] + freq_pos[1];
+ if(tmp_sum > max_sum){
+ max_sum = tmp_sum;
+ max_index = i;
+ }
+ freq_pos ++;
+ }
+ *dense_pos = mean + realPrecision * (ptrdiff_t)(max_index + 1 - radius);
+ *mean_freq = max_sum * 1.0 / sample_count;
+
+ free(freq_intervals);
+ free(intervals);
+ return powerOf2;
+}
+
+unsigned int optimize_intervals_double_3D_with_freq_and_dense_pos(double *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, double * dense_pos, double * max_freq, double * mean_freq)
+{
+ double mean = 0.0;
+ size_t len = r1 * r2 * r3;
+ size_t mean_distance = (int) (sqrt(len));
+ double * data_pos = oriData;
+ size_t offset_count = 0;
+ size_t offset_count_2 = 0;
+ size_t mean_count = 0;
+ while(data_pos - oriData < len){
+ mean += *data_pos;
+ mean_count ++;
+ data_pos += mean_distance;
+ offset_count += mean_distance;
+ offset_count_2 += mean_distance;
+ if(offset_count >= r3){
+ offset_count = 0;
+ data_pos -= 1;
+ }
+ if(offset_count_2 >= r2 * r3){
+ offset_count_2 = 0;
+ data_pos -= 1;
+ }
+ }
+ if(mean_count > 0) mean /= mean_count;
+ size_t range = 8192;
+ size_t radius = 4096;
+ size_t * freq_intervals = (size_t *) malloc(range*sizeof(size_t));
+ memset(freq_intervals, 0, range*sizeof(size_t));
+
+ unsigned int maxRangeRadius = confparams_cpr->maxRangeRadius;
+ int sampleDistance = confparams_cpr->sampleDistance;
+ double predThreshold = confparams_cpr->predThreshold;
+
+ size_t i;
+ size_t radiusIndex;
+ size_t r23=r2*r3;
+ double pred_value = 0, pred_err;
+ size_t *intervals = (size_t*)malloc(maxRangeRadius*sizeof(size_t));
+ memset(intervals, 0, maxRangeRadius*sizeof(size_t));
+
+ double mean_diff;
+ ptrdiff_t freq_index;
+ size_t freq_count = 0;
+ size_t sample_count = 0;
+
+ offset_count = confparams_cpr->sampleDistance - 2; // count r3 offset
+ data_pos = oriData + r23 + r3 + offset_count;
+ size_t n1_count = 1, n2_count = 1; // count i,j sum
+
+ while(data_pos - oriData < len){
+
+ 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];
+ pred_err = fabs(pred_value - *data_pos);
+ if(pred_err < realPrecision) freq_count ++;
+ radiusIndex = (pred_err/realPrecision+1)/2;
+ if(radiusIndex>=maxRangeRadius)
+ {
+ radiusIndex = maxRangeRadius - 1;
+ }
+ intervals[radiusIndex]++;
+
+ mean_diff = *data_pos - mean;
+ if(mean_diff > 0) freq_index = (ptrdiff_t)(mean_diff/realPrecision) + radius;
+ else freq_index = (ptrdiff_t)(mean_diff/realPrecision) - 1 + radius;
+ if(freq_index <= 0){
+ freq_intervals[0] ++;
+ }
+ else if(freq_index >= range){
+ freq_intervals[range - 1] ++;
+ }
+ else{
+ freq_intervals[freq_index] ++;
+ }
+ offset_count += sampleDistance;
+ if(offset_count >= r3){
+ n2_count ++;
+ if(n2_count == r2){
+ n1_count ++;
+ n2_count = 1;
+ data_pos += r3;
+ }
+ offset_count_2 = (n1_count + n2_count) % sampleDistance;
+ data_pos += (r3 + sampleDistance - offset_count) + (sampleDistance - offset_count_2);
+ offset_count = (sampleDistance - offset_count_2);
+ if(offset_count == 0) offset_count ++;
+ }
+ else data_pos += sampleDistance;
+ sample_count ++;
+ }
+ *max_freq = freq_count * 1.0/ sample_count;
+
+ //compute the appropriate number
+ size_t targetCount = sample_count*predThreshold;
+ size_t sum = 0;
+ for(i=0;i<maxRangeRadius;i++)
+ {
+ sum += intervals[i];
+ if(sum>targetCount)
+ break;
+ }
+ if(i>=maxRangeRadius)
+ i = maxRangeRadius-1;
+ unsigned int accIntervals = 2*(i+1);
+ unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
+
+ if(powerOf2<32)
+ powerOf2 = 32;
+ // collect frequency
+ size_t max_sum = 0;
+ size_t max_index = 0;
+ size_t tmp_sum;
+ size_t * freq_pos = freq_intervals + 1;
+ for(size_t i=1; i<range-2; i++){
+ tmp_sum = freq_pos[0] + freq_pos[1];
+ if(tmp_sum > max_sum){
+ max_sum = tmp_sum;
+ max_index = i;
+ }
+ freq_pos ++;
+ }
+ *dense_pos = mean + realPrecision * (ptrdiff_t)(max_index + 1 - radius);
+ *mean_freq = max_sum * 1.0 / sample_count;
+
+ free(freq_intervals);
+ free(intervals);
+ return powerOf2;
+}
+
+#define MIN(a, b) a<b? a : b
+unsigned char * SZ_compress_double_2D_MDQ_nonblocked_with_blocked_regression(double *oriData, size_t r1, size_t r2, double realPrecision, size_t * comp_size){
+
+ unsigned int quantization_intervals;
+ double sz_sample_correct_freq = -1;//0.5; //-1
+ double dense_pos;
+ double mean_flush_freq;
+ unsigned char use_mean = 0;
+
+ if(exe_params->optQuantMode==1)
+ {
+ quantization_intervals = optimize_intervals_double_2D_with_freq_and_dense_pos(oriData, r1, r2, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
+ if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
+ updateQuantizationInfo(quantization_intervals);
+ }
+ else{
+ quantization_intervals = exe_params->intvCapacity;
+ }
+
+ // calculate block dims
+ size_t num_x, num_y;
+ size_t block_size = 16;
+
+ SZ_COMPUTE_2D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_2D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+
+ size_t split_index_x, split_index_y;
+ size_t early_blockcount_x, early_blockcount_y;
+ size_t late_blockcount_x, late_blockcount_y;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+
+ size_t max_num_block_elements = early_blockcount_x * early_blockcount_y;
+ size_t num_blocks = num_x * num_y;
+ size_t num_elements = r1 * r2;
+
+ size_t dim0_offset = r2;
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ size_t unpred_data_max_size = max_num_block_elements;
+ double * result_unpredictable_data = (double *) malloc(unpred_data_max_size * sizeof(double) * num_blocks);
+ size_t total_unpred = 0;
+ size_t unpredictable_count;
+ double * data_pos = oriData;
+ int * type = result_type;
+ size_t offset_x, offset_y;
+ size_t current_blockcount_x, current_blockcount_y;
+
+ double * reg_params = (double *) malloc(num_blocks * 4 * sizeof(double));
+ double * reg_params_pos = reg_params;
+ // move regression part out
+ size_t params_offset_b = num_blocks;
+ size_t params_offset_c = 2*num_blocks;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+
+ data_pos = oriData + offset_x * dim0_offset + offset_y;
+
+ {
+ double * cur_data_pos = data_pos;
+ double fx = 0.0;
+ double fy = 0.0;
+ double f = 0;
+ double sum_x;
+ double curData;
+ for(size_t i=0; i<current_blockcount_x; i++){
+ sum_x = 0;
+ for(size_t j=0; j<current_blockcount_y; j++){
+ curData = *cur_data_pos;
+ sum_x += curData;
+ fy += curData * j;
+ cur_data_pos ++;
+ }
+ fx += sum_x * i;
+ f += sum_x;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ double coeff = 1.0 / (current_blockcount_x * current_blockcount_y);
+ reg_params_pos[0] = (2 * fx / (current_blockcount_x - 1) - f) * 6 * coeff / (current_blockcount_x + 1);
+ reg_params_pos[params_offset_b] = (2 * fy / (current_blockcount_y - 1) - f) * 6 * coeff / (current_blockcount_y + 1);
+ reg_params_pos[params_offset_c] = f * coeff - ((current_blockcount_x - 1) * reg_params_pos[0] / 2 + (current_blockcount_y - 1) * reg_params_pos[params_offset_b] / 2);
+ }
+
+ reg_params_pos ++;
+ }
+ }
+
+ //Compress coefficient arrays
+ double precision_a, precision_b, precision_c;
+ double rel_param_err = 0.15/3;
+ precision_a = rel_param_err * realPrecision / late_blockcount_x;
+ precision_b = rel_param_err * realPrecision / late_blockcount_y;
+ precision_c = rel_param_err * realPrecision;
+
+ double mean = 0;
+ use_mean = 0;
+ if(use_mean){
+ // compute mean
+ double sum = 0.0;
+ size_t mean_count = 0;
+ for(size_t i=0; i<num_elements; i++){
+ if(fabs(oriData[i] - dense_pos) < realPrecision){
+ sum += oriData[i];
+ mean_count ++;
+ }
+ }
+ if(mean_count > 0) mean = sum / mean_count;
+ }
+
+
+ double tmp_realPrecision = realPrecision;
+
+ // use two prediction buffers for higher performance
+ double * unpredictable_data = result_unpredictable_data;
+ unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
+ memset(indicator, 0, num_blocks * sizeof(unsigned char));
+ size_t reg_count = 0;
+ size_t strip_dim_0 = early_blockcount_x + 1;
+ size_t strip_dim_1 = r2 + 1;
+ size_t strip_dim0_offset = strip_dim_1;
+ unsigned char * indicator_pos = indicator;
+ size_t prediction_buffer_size = strip_dim_0 * strip_dim0_offset * sizeof(double);
+ double * prediction_buffer_1 = (double *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_1, 0, prediction_buffer_size);
+ double * prediction_buffer_2 = (double *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_2, 0, prediction_buffer_size);
+ double * cur_pb_buf = prediction_buffer_1;
+ double * next_pb_buf = prediction_buffer_2;
+ double * cur_pb_buf_pos;
+ double * next_pb_buf_pos;
+ int intvCapacity = exe_params->intvCapacity;
+ int intvRadius = exe_params->intvRadius;
+ int use_reg = 0;
+
+ reg_params_pos = reg_params;
+ // compress the regression coefficients on the fly
+ double last_coeffcients[3] = {0.0};
+ int coeff_intvCapacity_sz = 65536;
+ int coeff_intvRadius = coeff_intvCapacity_sz / 2;
+ int * coeff_type[3];
+ int * coeff_result_type = (int *) malloc(num_blocks*3*sizeof(int));
+ double * coeff_unpred_data[3];
+ double * coeff_unpredictable_data = (double *) malloc(num_blocks*3*sizeof(double));
+ double precision[3];
+ precision[0] = precision_a, precision[1] = precision_b, precision[2] = precision_c;
+ for(int i=0; i<3; i++){
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ coeff_unpred_data[i] = coeff_unpredictable_data + i * num_blocks;
+ }
+ int coeff_index = 0;
+ unsigned int coeff_unpredictable_count[3] = {0};
+ if(use_mean){
+ type = result_type;
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = oriData + offset_x * dim0_offset;
+
+ cur_pb_buf_pos = cur_pb_buf + strip_dim0_offset + 1;
+ next_pb_buf_pos = next_pb_buf + 1;
+ double * pb_pos = cur_pb_buf_pos;
+ double * next_pb_pos = next_pb_buf_pos;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ /*sampling: decide which predictor to use (regression or lorenzo)*/
+ {
+ double * cur_data_pos;
+ double curData;
+ double pred_reg, pred_sz;
+ double err_sz = 0.0, err_reg = 0.0;
+ // [1, 1] [3, 3] [5, 5] [7, 7] [9, 9]
+ // [1, 9] [3, 7] [7, 3] [9, 1]
+ int count = 0;
+ for(int i=1; i<current_blockcount_x; i+=2){
+ cur_data_pos = data_pos + i * dim0_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c];
+
+ err_sz += MIN(fabs(pred_sz - curData) + realPrecision*0.81, fabs(mean - curData));
+
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i * dim0_offset + (block_size - i);
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * (block_size - i) + reg_params_pos[params_offset_c];
+ err_sz += MIN(fabs(pred_sz - curData) + realPrecision*0.81, fabs(mean - curData));
+
+ err_reg += fabs(pred_reg - curData);
+
+ count += 2;
+ }
+
+ use_reg = (err_reg < err_sz);
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ double cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<3; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ double curData;
+ double pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ double * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+
+ index ++;
+ cur_data_pos ++;
+ }
+ } // end ii == -1
+ unpredictable_count = block_unpredictable_count;
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }// end use_reg
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ double * cur_pb_pos = pb_pos;
+ double * cur_data_pos = data_pos;
+ double curData;
+ double pred2D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ next_pb_pos[jj] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[j] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ data_pos += current_blockcount_y;
+ pb_pos += current_blockcount_y;
+ next_pb_pos += current_blockcount_y;
+ type += current_blockcount_x * current_blockcount_y;
+ }// end j
+ indicator_pos += num_y;
+ double * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }// end use mean
+ else{
+ type = result_type;
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = oriData + offset_x * dim0_offset;
+
+ cur_pb_buf_pos = cur_pb_buf + strip_dim0_offset + 1;
+ next_pb_buf_pos = next_pb_buf + 1;
+ double * pb_pos = cur_pb_buf_pos;
+ double * next_pb_pos = next_pb_buf_pos;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ /*sampling*/
+ {
+ // sample [2i + 1, 2i + 1] [2i + 1, bs - 2i]
+ double * cur_data_pos;
+ double curData;
+ double pred_reg, pred_sz;
+ double err_sz = 0.0, err_reg = 0.0;
+ // [1, 1] [3, 3] [5, 5] [7, 7] [9, 9]
+ // [1, 9] [3, 7] [7, 3] [9, 1]
+ int count = 0;
+ for(int i=1; i<current_blockcount_x; i+=2){
+ cur_data_pos = data_pos + i * dim0_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c];
+ err_sz += fabs(pred_sz - curData);
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i * dim0_offset + (block_size - i);
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * (block_size - i) + reg_params_pos[params_offset_c];
+ err_sz += fabs(pred_sz - curData);
+ err_reg += fabs(pred_reg - curData);
+
+ count += 2;
+ }
+ err_sz += realPrecision * count * 0.81;
+ use_reg = (err_reg < err_sz);
+
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ double cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<3; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ double curData;
+ double pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ double * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ // jj == current_blockcount_y - 1
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ // jj == current_blockcount_y - 1
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+
+ index ++;
+ cur_data_pos ++;
+ }
+ } // end ii == -1
+ unpredictable_count = block_unpredictable_count;
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }// end use_reg
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ double * cur_pb_pos = pb_pos;
+ double * cur_data_pos = data_pos;
+ double curData;
+ double pred2D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ next_pb_pos[jj] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[j] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ data_pos += current_blockcount_y;
+ pb_pos += current_blockcount_y;
+ next_pb_pos += current_blockcount_y;
+ type += current_blockcount_x * current_blockcount_y;
+ }// end j
+ indicator_pos += num_y;
+ double * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }
+ free(prediction_buffer_1);
+ free(prediction_buffer_2);
+
+ int stateNum = 2*quantization_intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ size_t nodeCount = 0;
+ size_t i = 0;
+ init(huffmanTree, result_type, num_elements);
+ for (i = 0; i < stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+
+ unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
+ // total size metadata # elements real precision intervals nodeCount huffman block index unpredicatable count mean unpred size elements
+ unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + treeByteSize + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(double) + total_unpred * sizeof(double) + num_elements * sizeof(int), 1);
+ unsigned char * result_pos = result;
+ initRandomAccessBytes(result_pos);
+ result_pos += meta_data_offset;
+
+ sizeToBytes(result_pos, num_elements);
+ result_pos += exe_params->SZ_SIZE_TYPE;
+
+ intToBytes_bigEndian(result_pos, block_size);
+ result_pos += sizeof(int);
+ doubleToBytes(result_pos, realPrecision);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, quantization_intervals);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+
+ memcpy(result_pos, &use_mean, sizeof(unsigned char));
+ result_pos += sizeof(unsigned char);
+ memcpy(result_pos, &mean, sizeof(double));
+ result_pos += sizeof(double);
+
+ size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
+ result_pos += indicator_size;
+
+ //convert the lead/mid/resi to byte stream
+ if(reg_count>0){
+ for(int e=0; e<3; e++){
+ int stateNum = 2*coeff_intvCapacity_sz;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ size_t nodeCount = 0;
+ init(huffmanTree, coeff_type[e], reg_count);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+ doubleToBytes(result_pos, precision[e]);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, coeff_intvRadius);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
+ sizeToBytes(result_pos, typeArray_size);
+ result_pos += sizeof(size_t) + typeArray_size;
+ intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
+ result_pos += sizeof(int);
+ memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(double));
+ result_pos += coeff_unpredictable_count[e]*sizeof(double);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ free(coeff_result_type);
+ free(coeff_unpredictable_data);
+
+ //record the number of unpredictable data and also store them
+ memcpy(result_pos, &total_unpred, sizeof(size_t));
+ result_pos += sizeof(size_t);
+ memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(double));
+ result_pos += total_unpred * sizeof(double);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, result_type, num_elements, result_pos, &typeArray_size);
+ result_pos += typeArray_size;
+
+ size_t totalEncodeSize = result_pos - result;
+ free(indicator);
+ free(result_unpredictable_data);
+ free(result_type);
+ free(reg_params);
+
+ SZ_ReleaseHuffman(huffmanTree);
+ *comp_size = totalEncodeSize;
+
+ return result;
+}
+
+unsigned char * SZ_compress_double_3D_MDQ_nonblocked_with_blocked_regression(double *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size){
+
+ unsigned int quantization_intervals;
+ double sz_sample_correct_freq = -1;//0.5; //-1
+ double dense_pos;
+ double mean_flush_freq;
+ unsigned char use_mean = 0;
+
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ size_t block_size = 6;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r3, num_z, block_size);
+
+ size_t split_index_x, split_index_y, split_index_z;
+ size_t early_blockcount_x, early_blockcount_y, early_blockcount_z;
+ size_t late_blockcount_x, late_blockcount_y, late_blockcount_z;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+ SZ_COMPUTE_BLOCKCOUNT(r3, num_z, split_index_z, early_blockcount_z, late_blockcount_z);
+
+ size_t max_num_block_elements = early_blockcount_x * early_blockcount_y * early_blockcount_z;
+ size_t num_blocks = num_x * num_y * num_z;
+ size_t num_elements = r1 * r2 * r3;
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ size_t unpred_data_max_size = max_num_block_elements;
+ double * result_unpredictable_data = (double *) malloc(unpred_data_max_size * sizeof(double) * num_blocks);
+ size_t total_unpred = 0;
+ size_t unpredictable_count;
+ size_t max_unpred_count = 0;
+ double * data_pos = oriData;
+ int * type = result_type;
+ size_t type_offset;
+ size_t offset_x, offset_y, offset_z;
+ size_t current_blockcount_x, current_blockcount_y, current_blockcount_z;
+
+ double * reg_params = (double *) malloc(num_blocks * 4 * sizeof(double));
+ double * reg_params_pos = reg_params;
+ // move regression part out
+ size_t params_offset_b = num_blocks;
+ size_t params_offset_c = 2*num_blocks;
+ size_t params_offset_d = 3*num_blocks;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+ /*Calculate regression coefficients*/
+ {
+ double * cur_data_pos = data_pos;
+ double fx = 0.0;
+ double fy = 0.0;
+ double fz = 0.0;
+ double f = 0;
+ double sum_x, sum_y;
+ double curData;
+ for(size_t i=0; i<current_blockcount_x; i++){
+ sum_x = 0;
+ for(size_t j=0; j<current_blockcount_y; j++){
+ sum_y = 0;
+ for(size_t k=0; k<current_blockcount_z; k++){
+ curData = *cur_data_pos;
+ // f += curData;
+ // fx += curData * i;
+ // fy += curData * j;
+ // fz += curData * k;
+ sum_y += curData;
+ fz += curData * k;
+ cur_data_pos ++;
+ }
+ fy += sum_y * j;
+ sum_x += sum_y;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ fx += sum_x * i;
+ f += sum_x;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ double coeff = 1.0 / (current_blockcount_x * current_blockcount_y * current_blockcount_z);
+ reg_params_pos[0] = (2 * fx / (current_blockcount_x - 1) - f) * 6 * coeff / (current_blockcount_x + 1);
+ reg_params_pos[params_offset_b] = (2 * fy / (current_blockcount_y - 1) - f) * 6 * coeff / (current_blockcount_y + 1);
+ reg_params_pos[params_offset_c] = (2 * fz / (current_blockcount_z - 1) - f) * 6 * coeff / (current_blockcount_z + 1);
+ reg_params_pos[params_offset_d] = f * coeff - ((current_blockcount_x - 1) * reg_params_pos[0] / 2 + (current_blockcount_y - 1) * reg_params_pos[params_offset_b] / 2 + (current_blockcount_z - 1) * reg_params_pos[params_offset_c] / 2);
+ }
+ reg_params_pos ++;
+ }
+ }
+ }
+
+ //Compress coefficient arrays
+ double precision_a, precision_b, precision_c, precision_d;
+ double rel_param_err = 0.025;
+ precision_a = rel_param_err * realPrecision / late_blockcount_x;
+ precision_b = rel_param_err * realPrecision / late_blockcount_y;
+ precision_c = rel_param_err * realPrecision / late_blockcount_z;
+ precision_d = rel_param_err * realPrecision;
+
+ if(exe_params->optQuantMode==1)
+ {
+ quantization_intervals = optimize_intervals_double_3D_with_freq_and_dense_pos(oriData, r1, r2, r3, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
+ if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
+ updateQuantizationInfo(quantization_intervals);
+ }
+ else{
+ quantization_intervals = exe_params->intvCapacity;
+ }
+
+ double mean = 0;
+ if(use_mean){
+ // compute mean
+ double sum = 0.0;
+ size_t mean_count = 0;
+ for(size_t i=0; i<num_elements; i++){
+ if(fabs(oriData[i] - dense_pos) < realPrecision){
+ sum += oriData[i];
+ mean_count ++;
+ }
+ }
+ if(mean_count > 0) mean = sum / mean_count;
+ }
+
+ double tmp_realPrecision = realPrecision;
+
+ // use two prediction buffers for higher performance
+ double * unpredictable_data = result_unpredictable_data;
+ unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
+ memset(indicator, 0, num_blocks * sizeof(unsigned char));
+ size_t reg_count = 0;
+ size_t strip_dim_0 = early_blockcount_x + 1;
+ size_t strip_dim_1 = r2 + 1;
+ size_t strip_dim_2 = r3 + 1;
+ size_t strip_dim0_offset = strip_dim_1 * strip_dim_2;
+ size_t strip_dim1_offset = strip_dim_2;
+ unsigned char * indicator_pos = indicator;
+
+ size_t prediction_buffer_size = strip_dim_0 * strip_dim0_offset * sizeof(double);
+ double * prediction_buffer_1 = (double *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_1, 0, prediction_buffer_size);
+ double * prediction_buffer_2 = (double *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_2, 0, prediction_buffer_size);
+ double * cur_pb_buf = prediction_buffer_1;
+ double * next_pb_buf = prediction_buffer_2;
+ double * cur_pb_buf_pos;
+ double * next_pb_buf_pos;
+ int intvCapacity = exe_params->intvCapacity;
+ int intvRadius = exe_params->intvRadius;
+ int use_reg = 0;
+ double noise = realPrecision * 1.22;
+
+ reg_params_pos = reg_params;
+ // compress the regression coefficients on the fly
+ double last_coeffcients[4] = {0.0};
+ int coeff_intvCapacity_sz = 65536;
+ int coeff_intvRadius = coeff_intvCapacity_sz / 2;
+ int * coeff_type[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ double * coeff_unpred_data[4];
+ double * coeff_unpredictable_data = (double *) malloc(num_blocks*4*sizeof(double));
+ double precision[4];
+ precision[0] = precision_a, precision[1] = precision_b, precision[2] = precision_c, precision[3] = precision_d;
+ for(int i=0; i<4; i++){
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ coeff_unpred_data[i] = coeff_unpredictable_data + i * num_blocks;
+ }
+ int coeff_index = 0;
+ unsigned int coeff_unpredictable_count[4] = {0};
+
+ if(use_mean){
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset;
+ type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset;
+ type = result_type + type_offset;
+
+ // prediction buffer is (current_block_count_x + 1) * (current_block_count_y + 1) * (current_block_count_z + 1)
+ cur_pb_buf_pos = cur_pb_buf + offset_y * strip_dim1_offset + strip_dim0_offset + strip_dim1_offset + 1;
+ next_pb_buf_pos = next_pb_buf + offset_y * strip_dim1_offset + strip_dim1_offset + 1;
+
+ size_t current_blockcount_z;
+ double * pb_pos = cur_pb_buf_pos;
+ double * next_pb_pos = next_pb_buf_pos;
+ size_t strip_unpredictable_count = 0;
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ /*sampling and decide which predictor*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ double * cur_data_pos;
+ double curData;
+ double pred_reg, pred_sz;
+ double err_sz = 0.0, err_reg = 0.0;
+ int bmi = 0;
+ if(i>0 && j>0 && k>0){
+ for(int i=0; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ else{
+ for(int i=1; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ }
+ }
+ use_reg = (err_reg < err_sz);
+ }
+ if(use_reg){
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ double cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ double curData;
+ double pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ double * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ unpredictable_count = block_unpredictable_count;
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ double * cur_pb_pos = pb_pos;
+ double * cur_data_pos = data_pos;
+ double curData;
+ double pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y * strip_dim1_offset;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ next_pb_pos[jj * strip_dim1_offset + kk] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+
+ reg_params_pos ++;
+ data_pos += current_blockcount_z;
+ pb_pos += current_blockcount_z;
+ next_pb_pos += current_blockcount_z;
+ type += current_blockcount_x * current_blockcount_y * current_blockcount_z;
+
+ } // end k
+
+ if(strip_unpredictable_count > max_unpred_count){
+ max_unpred_count = strip_unpredictable_count;
+ }
+ total_unpred += strip_unpredictable_count;
+ indicator_pos += num_z;
+ }// end j
+ double * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }
+ else{
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset;
+ // copy bottom plane from plane buffer
+ // memcpy(prediction_buffer, bottom_buffer + offset_y * strip_dim1_offset, (current_blockcount_y + 1) * strip_dim1_offset * sizeof(double));
+ type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset;
+ type = result_type + type_offset;
+
+ // prediction buffer is (current_block_count_x + 1) * (current_block_count_y + 1) * (current_block_count_z + 1)
+ cur_pb_buf_pos = cur_pb_buf + offset_y * strip_dim1_offset + strip_dim0_offset + strip_dim1_offset + 1;
+ next_pb_buf_pos = next_pb_buf + offset_y * strip_dim1_offset + strip_dim1_offset + 1;
+
+ size_t current_blockcount_z;
+ double * pb_pos = cur_pb_buf_pos;
+ double * next_pb_pos = next_pb_buf_pos;
+ size_t strip_unpredictable_count = 0;
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+ /*sampling*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ double * cur_data_pos;
+ double curData;
+ double pred_reg, pred_sz;
+ double err_sz = 0.0, err_reg = 0.0;
+ int bmi;
+ if(i>0 && j>0 && k>0){
+ for(int i=0; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ else{
+ for(int i=1; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ use_reg = (err_reg < err_sz);
+
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ double cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ double curData;
+ double pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ double * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ unpredictable_count = block_unpredictable_count;
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ double * cur_pb_pos = pb_pos;
+ double * cur_data_pos = data_pos;
+ double curData;
+ double pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y * strip_dim1_offset;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+
+ reg_params_pos ++;
+ data_pos += current_blockcount_z;
+ pb_pos += current_blockcount_z;
+ next_pb_pos += current_blockcount_z;
+ type += current_blockcount_x * current_blockcount_y * current_blockcount_z;
+
+ }
+
+ if(strip_unpredictable_count > max_unpred_count){
+ max_unpred_count = strip_unpredictable_count;
+ }
+ total_unpred += strip_unpredictable_count;
+ indicator_pos += num_z;
+ }
+ double * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }
+ }
+
+ free(prediction_buffer_1);
+ free(prediction_buffer_2);
+
+ int stateNum = 2*quantization_intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ size_t nodeCount = 0;
+ init(huffmanTree, result_type, num_elements);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+
+ unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
+ // total size metadata # elements real precision intervals nodeCount huffman block index unpredicatable count mean unpred size elements
+ unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + treeByteSize + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(double) + total_unpred * sizeof(double) + num_elements * sizeof(int), 1);
+ unsigned char * result_pos = result;
+ initRandomAccessBytes(result_pos);
+
+ result_pos += meta_data_offset;
+
+ sizeToBytes(result_pos,num_elements); //SZ_SIZE_TYPE: 4 or 8
+ result_pos += exe_params->SZ_SIZE_TYPE;
+
+ intToBytes_bigEndian(result_pos, block_size);
+ result_pos += sizeof(int);
+ doubleToBytes(result_pos, realPrecision);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, quantization_intervals);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+
+ memcpy(result_pos, &use_mean, sizeof(unsigned char));
+ result_pos += sizeof(unsigned char);
+ memcpy(result_pos, &mean, sizeof(double));
+ result_pos += sizeof(double);
+ size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
+ result_pos += indicator_size;
+
+ //convert the lead/mid/resi to byte stream
+ if(reg_count > 0){
+ for(int e=0; e<4; e++){
+ int stateNum = 2*coeff_intvCapacity_sz;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ size_t nodeCount = 0;
+ init(huffmanTree, coeff_type[e], reg_count);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+ doubleToBytes(result_pos, precision[e]);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, coeff_intvRadius);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
+ sizeToBytes(result_pos, typeArray_size);
+ result_pos += sizeof(size_t) + typeArray_size;
+ intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
+ result_pos += sizeof(int);
+ memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(double));
+ result_pos += coeff_unpredictable_count[e]*sizeof(double);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ free(coeff_result_type);
+ free(coeff_unpredictable_data);
+
+ //record the number of unpredictable data and also store them
+ memcpy(result_pos, &total_unpred, sizeof(size_t));
+ result_pos += sizeof(size_t);
+ memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(double));
+ result_pos += total_unpred * sizeof(double);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, result_type, num_elements, result_pos, &typeArray_size);
+ result_pos += typeArray_size;
+ size_t totalEncodeSize = result_pos - result;
+ free(indicator);
+ free(result_unpredictable_data);
+ free(result_type);
+ free(reg_params);
+
+
+ SZ_ReleaseHuffman(huffmanTree);
+ *comp_size = totalEncodeSize;
+ return result;
+}
diff --git a/thirdparty/SZ/sz/src/sz_double_pwr.c b/thirdparty/SZ/sz/src/sz_double_pwr.c
index 59be38c..881d495 100644
--- a/thirdparty/SZ/sz/src/sz_double_pwr.c
+++ b/thirdparty/SZ/sz/src/sz_double_pwr.c
@@ -23,6 +23,7 @@
#include "sz_double_pwr.h"
#include "zlib.h"
#include "rw.h"
+#include "utility.h"
void compute_segment_precisions_double_1D(double *oriData, size_t dataLength, double* pwrErrBound, unsigned char* pwrErrBoundBytes, double globalPrecision)
{
@@ -1772,3 +1773,188 @@ size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio,
free_TightDataPointStorageD(tdps);
}
+
+#include <stdbool.h>
+
+void SZ_compress_args_double_NoCkRngeNoGzip_1D_pwr_pre_log(unsigned char** newByteData, double *oriData, double pwrErrRatio, size_t dataLength, size_t *outSize, double min, double max){
+
+ double * log_data = (double *) malloc(dataLength * sizeof(double));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ double max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ double min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ double valueRangeSize, medianValue_f;
+ computeRangeSize_double(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 2.23e-16;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+ TightDataPointStorageD* tdps = SZ_compress_double_1D_MDQ(log_data, dataLength, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_double(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(double))
+ SZ_compress_args_double_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageD(tdps);
+}
+
+void SZ_compress_args_double_NoCkRngeNoGzip_2D_pwr_pre_log(unsigned char** newByteData, double *oriData, double pwrErrRatio, size_t r1, size_t r2, size_t *outSize, double min, double max){
+
+ size_t dataLength = r1 * r2;
+ double * log_data = (double *) malloc(dataLength * sizeof(double));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ double max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ double min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ double valueRangeSize, medianValue_f;
+ computeRangeSize_double(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 2.23e-16;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+ TightDataPointStorageD* tdps = SZ_compress_double_2D_MDQ(log_data, r1, r2, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_double(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(double))
+ SZ_compress_args_double_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageD(tdps);
+}
+
+void SZ_compress_args_double_NoCkRngeNoGzip_3D_pwr_pre_log(unsigned char** newByteData, double *oriData, double pwrErrRatio, size_t r1, size_t r2, size_t r3, size_t *outSize, double min, double max){
+
+ size_t dataLength = r1 * r2 * r3;
+ double * log_data = (double *) malloc(dataLength * sizeof(double));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ double max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ double min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ double valueRangeSize, medianValue_f;
+ computeRangeSize_double(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 2.23e-16;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+ TightDataPointStorageD* tdps = SZ_compress_double_3D_MDQ(log_data, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_double(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(double))
+ SZ_compress_args_double_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageD(tdps);
+}
diff --git a/thirdparty/SZ/sz/src/sz_double_ts.c b/thirdparty/SZ/sz/src/sz_double_ts.c
index b83562e..de9f7cd 100644
--- a/thirdparty/SZ/sz/src/sz_double_ts.c
+++ b/thirdparty/SZ/sz/src/sz_double_ts.c
@@ -66,7 +66,7 @@ unsigned int optimize_intervals_double_1D_ts(double *oriData, size_t dataLength,
TightDataPointStorageD* SZ_compress_double_1D_MDQ_ts(double *oriData, size_t dataLength, sz_multisteps* multisteps,
double realPrecision, double valueRangeSize, double medianValue_d)
{
-double* preStepData = (double*)(multisteps->hist_data);
+ double* preStepData = (double*)(multisteps->hist_data);
//store the decompressed data
double* decData = (double*)malloc(sizeof(double)*dataLength);
memset(decData, 0, sizeof(double)*dataLength);
diff --git a/thirdparty/SZ/sz/src/sz_float.c b/thirdparty/SZ/sz/src/sz_float.c
index c0a2a18..74d7f20 100644
--- a/thirdparty/SZ/sz/src/sz_float.c
+++ b/thirdparty/SZ/sz/src/sz_float.c
@@ -1,6 +1,6 @@
/**
* @file sz_float.c
- * @author Sheng Di and Dingwen Tao
+ * @author Sheng Di, Dingwen Tao, Xin Liang
* @date Aug, 2016
* @brief SZ_Init, Compression and Decompression functions
* (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory.
@@ -25,6 +25,7 @@
#include "zlib.h"
#include "rw.h"
#include "sz_float_ts.h"
+#include "utility.h"
unsigned char* SZ_skip_compress_float(float* data, size_t dataLength, size_t* outSize)
{
@@ -405,7 +406,7 @@ size_t dataLength, double realPrecision, float valueRangeSize, float medianValue
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = fabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1356,8 +1357,11 @@ char SZ_compress_args_float_NoCkRngeNoGzip_3D(unsigned char** newByteData, float
compressionType = 1; //time-series based compression
}
else
- {
- tdps = SZ_compress_float_3D_MDQ(oriData, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ tdps = SZ_compress_float_3D_MDQ(oriData, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f);
+ else
+ *newByteData = SZ_compress_float_3D_MDQ_nonblocked_with_blocked_regression(oriData, r1, r2, r3, realPrecision, outSize);
compressionType = 0; //snapshot-based compression
multisteps->lastSnapshotStep = timestep;
}
@@ -1366,14 +1370,14 @@ char SZ_compress_args_float_NoCkRngeNoGzip_3D(unsigned char** newByteData, float
#endif
tdps = SZ_compress_float_3D_MDQ(oriData, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f);
+ if(tdps!=NULL)
+ {
+ convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(float))
+ SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize);
+ free_TightDataPointStorageF(tdps);
+ }
- convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
-
- if(*outSize>dataLength*sizeof(float))
- SZ_compress_args_float_StoreOriData(oriData, dataLength, tdps, newByteData, outSize);
-
- free_TightDataPointStorageF(tdps);
-
return compressionType;
}
@@ -1770,8 +1774,8 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwrErrRatio)
{
if(errBoundMode>=PW_REL)
{
- //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(newByteData, oriData, realPrecision, r1, outSize, min, max);
- SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(newByteData, oriData, r1, absErr_Bound, relBoundRatio, pwrErrRatio, valueRangeSize, medianValue, outSize);
+ SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log(newByteData, oriData, pwrErrRatio, r1, outSize, min, max);
+ //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(newByteData, oriData, r1, absErr_Bound, relBoundRatio, pwrErrRatio, valueRangeSize, medianValue, outSize);
}
else
SZ_compress_args_float_NoCkRngeNoGzip_1D(newByteData, oriData, r1, realPrecision, outSize, valueRangeSize, medianValue);
@@ -1779,21 +1783,21 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwrErrRatio)
else if(r5==0&&r4==0&&r3==0)
{
if(errBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr(newByteData, oriData, realPrecision, r2, r1, outSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr_pre_log(newByteData, oriData, pwrErrRatio, r2, r1, outSize, min, max);
else
SZ_compress_args_float_NoCkRngeNoGzip_2D(newByteData, oriData, r2, r1, realPrecision, outSize, valueRangeSize, medianValue);
}
else if(r5==0&&r4==0)
{
if(errBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(newByteData, oriData, realPrecision, r3, r2, r1, outSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_pre_log(newByteData, oriData, pwrErrRatio, r3, r2, r1, outSize, min, max);
else
SZ_compress_args_float_NoCkRngeNoGzip_3D(newByteData, oriData, r3, r2, r1, realPrecision, outSize, valueRangeSize, medianValue);
}
else if(r5==0)
{
if(errBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(newByteData, oriData, realPrecision, r4*r3, r2, r1, outSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_pre_log(newByteData, oriData, pwrErrRatio, r4*r3, r2, r1, outSize, min, max);
else
SZ_compress_args_float_NoCkRngeNoGzip_3D(newByteData, oriData, r4*r3, r2, r1, realPrecision, outSize, valueRangeSize, medianValue);
}
@@ -1854,9 +1858,8 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
{
if(confparams_cpr->errorBoundMode>=PW_REL)
{
- //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr(&tmpByteData, oriData, realPrecision, r1, &tmpOutSize, min, max);
- SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio,
- valueRangeSize, medianValue, &tmpOutSize);
+ SZ_compress_args_float_NoCkRngeNoGzip_1D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r1, &tmpOutSize, min, max);
+ //SZ_compress_args_float_NoCkRngeNoGzip_1D_pwrgroup(&tmpByteData, oriData, r1, absErr_Bound, relBoundRatio, pwRelBoundRatio, valueRangeSize, medianValue, &tmpOutSize);
}
else
#ifdef HAVE_TIMECMPR
@@ -1870,33 +1873,43 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
if (r3==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr(&tmpByteData, oriData, realPrecision, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_2D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r2, r1, &tmpOutSize, min, max);
else
#ifdef HAVE_TIMECMPR
if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
- SZ_compress_args_float_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_float_NoCkRngeNoGzip_2D(&tmpByteData, oriData, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_float_2D_MDQ_nonblocked_with_blocked_regression(oriData, r2, r1, realPrecision, &tmpOutSize);
+ }
}
else
if (r4==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r3, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r3, r2, r1, &tmpOutSize, min, max);
else
#ifdef HAVE_TIMECMPR
if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
- multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
- SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_float_NoCkRngeNoGzip_3D(&tmpByteData, oriData, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_float_3D_MDQ_nonblocked_with_blocked_regression(oriData, r3, r2, r1, realPrecision, &tmpOutSize);
+ }
}
else
if (r5==0)
{
if(confparams_cpr->errorBoundMode>=PW_REL)
- SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr(&tmpByteData, oriData, realPrecision, r4*r3, r2, r1, &tmpOutSize, min, max);
+ SZ_compress_args_float_NoCkRngeNoGzip_3D_pwr_pre_log(&tmpByteData, oriData, pwRelBoundRatio, r4*r3, r2, r1, &tmpOutSize, min, max);
//ToDO
//SZ_compress_args_float_NoCkRngeNoGzip_4D_pwr(&tmpByteData, oriData, r4, r3, r2, r1, &tmpOutSize, min, max);
else
@@ -1905,7 +1918,12 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
multisteps->compressionType = SZ_compress_args_float_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
else
#endif
- SZ_compress_args_float_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ {
+ if(sz_with_regression == SZ_NO_REGRESSION)
+ SZ_compress_args_float_NoCkRngeNoGzip_4D(&tmpByteData, oriData, r4, r3, r2, r1, realPrecision, &tmpOutSize, valueRangeSize, medianValue);
+ else
+ tmpByteData = SZ_compress_float_3D_MDQ_nonblocked_with_blocked_regression(oriData, r4*r3, r2, r1, realPrecision, &tmpOutSize);
+ }
}
else
{
@@ -1920,7 +1938,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio, double pwRelBoundRa
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION || confparams_cpr->szMode==SZ_TEMPORAL_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
@@ -3374,7 +3392,7 @@ unsigned int optimize_intervals_float_3D_opt(float *oriData, size_t r1, size_t r
float pred_value = 0, pred_err;
size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
- size_t totalSampleSize = 0;//(r1-1)*(r2-1)*(r3-1)/confparams_cpr->sampleDistance;
+ size_t totalSampleSize = 0;
size_t offset_count = confparams_cpr->sampleDistance - 2; // count r3 offset
size_t offset_count_2;
@@ -3389,11 +3407,8 @@ unsigned int optimize_intervals_float_3D_opt(float *oriData, size_t r1, size_t r
if(radiusIndex>=confparams_cpr->maxRangeRadius)
{
radiusIndex = confparams_cpr->maxRangeRadius - 1;
- //printf("radiusIndex=%d\n", radiusIndex);
}
intervals[radiusIndex]++;
- // printf("TEST: %ld, i: %ld\tj: %ld\tk: %ld\n", data_pos - oriData);
- // fflush(stdout);
offset_count += confparams_cpr->sampleDistance;
if(offset_count >= r3){
n2_count ++;
@@ -3409,9 +3424,6 @@ unsigned int optimize_intervals_float_3D_opt(float *oriData, size_t r1, size_t r
}
else data_pos += confparams_cpr->sampleDistance;
}
- // printf("sample_count: %ld\n", sample_count);
- // fflush(stdout);
- // if(*max_freq < 0.15) *max_freq *= 2;
//compute the appropriate number
size_t targetCount = totalSampleSize*confparams_cpr->predThreshold;
size_t sum = 0;
@@ -3429,7 +3441,6 @@ unsigned int optimize_intervals_float_3D_opt(float *oriData, size_t r1, size_t r
if(powerOf2<32)
powerOf2 = 32;
free(intervals);
- //printf("targetCount=%d, sum=%d, totalSampleSize=%d, ratio=%f, accIntervals=%d, powerOf2=%d\n", targetCount, sum, totalSampleSize, (double)sum/(double)totalSampleSize, accIntervals, powerOf2);
return powerOf2;
}
@@ -3749,10 +3760,7 @@ unsigned int optimize_intervals_float_2D_opt(float *oriData, size_t r1, size_t r
float pred_value = 0, pred_err;
size_t *intervals = (size_t*)malloc(confparams_cpr->maxRangeRadius*sizeof(size_t));
memset(intervals, 0, confparams_cpr->maxRangeRadius*sizeof(size_t));
- size_t totalSampleSize = 0;//(r1-1)*(r2-1)/confparams_cpr->sampleDistance;
-
- //float max = oriData[0];
- //float min = oriData[0];
+ size_t totalSampleSize = 0;
size_t offset_count = confparams_cpr->sampleDistance - 1; // count r2 offset
size_t offset_count_2;
@@ -3811,7 +3819,6 @@ unsigned int optimize_intervals_float_1D_opt(float *oriData, size_t dataLength,
float * data_pos = oriData + 2;
while(data_pos - oriData < dataLength){
totalSampleSize++;
- //pred_value = 2*data_pos[-1] - data_pos[-2];
pred_value = data_pos[-1];
pred_err = fabs(pred_value - *data_pos);
radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
@@ -3840,7 +3847,6 @@ unsigned int optimize_intervals_float_1D_opt(float *oriData, size_t dataLength,
powerOf2 = 32;
free(intervals);
- //printf("accIntervals=%d, powerOf2=%d\n", accIntervals, powerOf2);
return powerOf2;
}
@@ -4036,3 +4042,2790 @@ size_t SZ_compress_float_2D_MDQ_RA_block(float * block_ori_data, float * mean, s
return unpredictable_count;
}
+/*The above code is for sz 1.4.13; the following code is for sz 2.0*/
+
+unsigned int optimize_intervals_float_2D_with_freq_and_dense_pos(float *oriData, size_t r1, size_t r2, double realPrecision, float * dense_pos, float * max_freq, float * mean_freq)
+{
+ float mean = 0.0;
+ size_t len = r1 * r2;
+ size_t mean_distance = (int) (sqrt(len));
+
+ float * data_pos = oriData;
+ size_t mean_count = 0;
+ while(data_pos - oriData < len){
+ mean += *data_pos;
+ mean_count ++;
+ data_pos += mean_distance;
+ }
+ if(mean_count > 0) mean /= mean_count;
+ size_t range = 8192;
+ size_t radius = 4096;
+ size_t * freq_intervals = (size_t *) malloc(range*sizeof(size_t));
+ memset(freq_intervals, 0, range*sizeof(size_t));
+
+ unsigned int maxRangeRadius = confparams_cpr->maxRangeRadius;
+ int sampleDistance = confparams_cpr->sampleDistance;
+ float predThreshold = confparams_cpr->predThreshold;
+
+ size_t i;
+ size_t radiusIndex;
+ float pred_value = 0, pred_err;
+ size_t *intervals = (size_t*)malloc(maxRangeRadius*sizeof(size_t));
+ memset(intervals, 0, maxRangeRadius*sizeof(size_t));
+
+ float mean_diff;
+ ptrdiff_t freq_index;
+ size_t freq_count = 0;
+ size_t n1_count = 1;
+ size_t offset_count = sampleDistance - 1;
+ size_t offset_count_2 = 0;
+ size_t sample_count = 0;
+ data_pos = oriData + r2 + offset_count;
+ while(data_pos - oriData < len){
+ pred_value = data_pos[-1] + data_pos[-r2] - data_pos[-r2-1];
+ pred_err = fabs(pred_value - *data_pos);
+ if(pred_err < realPrecision) freq_count ++;
+ radiusIndex = (unsigned long)((pred_err/realPrecision+1)/2);
+ if(radiusIndex>=maxRangeRadius)
+ radiusIndex = maxRangeRadius - 1;
+ intervals[radiusIndex]++;
+
+ mean_diff = *data_pos - mean;
+ if(mean_diff > 0) freq_index = (ptrdiff_t)(mean_diff/realPrecision) + radius;
+ else freq_index = (ptrdiff_t)(mean_diff/realPrecision) - 1 + radius;
+ if(freq_index <= 0){
+ freq_intervals[0] ++;
+ }
+ else if(freq_index >= range){
+ freq_intervals[range - 1] ++;
+ }
+ else{
+ freq_intervals[freq_index] ++;
+ }
+ offset_count += sampleDistance;
+ if(offset_count >= r2){
+ n1_count ++;
+ offset_count_2 = n1_count % sampleDistance;
+ data_pos += (r2 + sampleDistance - offset_count) + (sampleDistance - offset_count_2);
+ offset_count = (sampleDistance - offset_count_2);
+ if(offset_count == 0) offset_count ++;
+ }
+ else data_pos += sampleDistance;
+ sample_count ++;
+ }
+ *max_freq = freq_count * 1.0/ sample_count;
+
+ //compute the appropriate number
+ size_t targetCount = sample_count*predThreshold;
+ size_t sum = 0;
+ for(i=0;i<maxRangeRadius;i++)
+ {
+ sum += intervals[i];
+ if(sum>targetCount)
+ break;
+ }
+ if(i>=maxRangeRadius)
+ i = maxRangeRadius-1;
+ unsigned int accIntervals = 2*(i+1);
+ unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
+
+ if(powerOf2<32)
+ powerOf2 = 32;
+
+ // collect frequency
+ size_t max_sum = 0;
+ size_t max_index = 0;
+ size_t tmp_sum;
+ size_t * freq_pos = freq_intervals + 1;
+ for(size_t i=1; i<range-2; i++){
+ tmp_sum = freq_pos[0] + freq_pos[1];
+ if(tmp_sum > max_sum){
+ max_sum = tmp_sum;
+ max_index = i;
+ }
+ freq_pos ++;
+ }
+ *dense_pos = mean + realPrecision * (ptrdiff_t)(max_index + 1 - radius);
+ *mean_freq = max_sum * 1.0 / sample_count;
+
+ free(freq_intervals);
+ free(intervals);
+ return powerOf2;
+}
+
+// 2D: modified for higher performance
+#define MIN(a, b) a<b? a : b
+unsigned char * SZ_compress_float_2D_MDQ_nonblocked_with_blocked_regression(float *oriData, size_t r1, size_t r2, double realPrecision, size_t * comp_size){
+
+ unsigned int quantization_intervals;
+ float sz_sample_correct_freq = -1;//0.5; //-1
+ float dense_pos;
+ float mean_flush_freq;
+ unsigned char use_mean = 0;
+
+ if(exe_params->optQuantMode==1)
+ {
+ quantization_intervals = optimize_intervals_float_2D_with_freq_and_dense_pos(oriData, r1, r2, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
+ if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
+ updateQuantizationInfo(quantization_intervals);
+ }
+ else{
+ quantization_intervals = exe_params->intvCapacity;
+ }
+
+ // calculate block dims
+ size_t num_x, num_y;
+ size_t block_size = 16;
+
+ SZ_COMPUTE_2D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_2D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+
+ size_t split_index_x, split_index_y;
+ size_t early_blockcount_x, early_blockcount_y;
+ size_t late_blockcount_x, late_blockcount_y;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+
+ size_t max_num_block_elements = early_blockcount_x * early_blockcount_y;
+ size_t num_blocks = num_x * num_y;
+ size_t num_elements = r1 * r2;
+
+ size_t dim0_offset = r2;
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ size_t unpred_data_max_size = max_num_block_elements;
+ float * result_unpredictable_data = (float *) malloc(unpred_data_max_size * sizeof(float) * num_blocks);
+ size_t total_unpred = 0;
+ size_t unpredictable_count;
+ float * data_pos = oriData;
+ int * type = result_type;
+ size_t offset_x, offset_y;
+ size_t current_blockcount_x, current_blockcount_y;
+
+ float * reg_params = (float *) malloc(num_blocks * 4 * sizeof(float));
+ float * reg_params_pos = reg_params;
+ // move regression part out
+ size_t params_offset_b = num_blocks;
+ size_t params_offset_c = 2*num_blocks;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+
+ data_pos = oriData + offset_x * dim0_offset + offset_y;
+
+ {
+ float * cur_data_pos = data_pos;
+ float fx = 0.0;
+ float fy = 0.0;
+ float f = 0;
+ double sum_x;
+ float curData;
+ for(size_t i=0; i<current_blockcount_x; i++){
+ sum_x = 0;
+ for(size_t j=0; j<current_blockcount_y; j++){
+ curData = *cur_data_pos;
+ sum_x += curData;
+ fy += curData * j;
+ cur_data_pos ++;
+ }
+ fx += sum_x * i;
+ f += sum_x;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ float coeff = 1.0 / (current_blockcount_x * current_blockcount_y);
+ reg_params_pos[0] = (2 * fx / (current_blockcount_x - 1) - f) * 6 * coeff / (current_blockcount_x + 1);
+ reg_params_pos[params_offset_b] = (2 * fy / (current_blockcount_y - 1) - f) * 6 * coeff / (current_blockcount_y + 1);
+ reg_params_pos[params_offset_c] = f * coeff - ((current_blockcount_x - 1) * reg_params_pos[0] / 2 + (current_blockcount_y - 1) * reg_params_pos[params_offset_b] / 2);
+ }
+
+ reg_params_pos ++;
+ }
+ }
+
+ //Compress coefficient arrays
+ double precision_a, precision_b, precision_c;
+ float rel_param_err = 0.15/3;
+ precision_a = rel_param_err * realPrecision / late_blockcount_x;
+ precision_b = rel_param_err * realPrecision / late_blockcount_y;
+ precision_c = rel_param_err * realPrecision;
+
+ float mean = 0;
+ use_mean = 0;
+ if(use_mean){
+ // compute mean
+ double sum = 0.0;
+ size_t mean_count = 0;
+ for(size_t i=0; i<num_elements; i++){
+ if(fabs(oriData[i] - dense_pos) < realPrecision){
+ sum += oriData[i];
+ mean_count ++;
+ }
+ }
+ if(mean_count > 0) mean = sum / mean_count;
+ }
+
+
+ double tmp_realPrecision = realPrecision;
+
+ // use two prediction buffers for higher performance
+ float * unpredictable_data = result_unpredictable_data;
+ unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
+ memset(indicator, 0, num_blocks * sizeof(unsigned char));
+ size_t reg_count = 0;
+ size_t strip_dim_0 = early_blockcount_x + 1;
+ size_t strip_dim_1 = r2 + 1;
+ size_t strip_dim0_offset = strip_dim_1;
+ unsigned char * indicator_pos = indicator;
+ size_t prediction_buffer_size = strip_dim_0 * strip_dim0_offset * sizeof(float);
+ float * prediction_buffer_1 = (float *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_1, 0, prediction_buffer_size);
+ float * prediction_buffer_2 = (float *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_2, 0, prediction_buffer_size);
+ float * cur_pb_buf = prediction_buffer_1;
+ float * next_pb_buf = prediction_buffer_2;
+ float * cur_pb_buf_pos;
+ float * next_pb_buf_pos;
+ int intvCapacity = exe_params->intvCapacity;
+ int intvRadius = exe_params->intvRadius;
+ int use_reg = 0;
+
+ reg_params_pos = reg_params;
+ // compress the regression coefficients on the fly
+ float last_coeffcients[3] = {0.0};
+ int coeff_intvCapacity_sz = 65536;
+ int coeff_intvRadius = coeff_intvCapacity_sz / 2;
+ int * coeff_type[3];
+ int * coeff_result_type = (int *) malloc(num_blocks*3*sizeof(int));
+ float * coeff_unpred_data[3];
+ float * coeff_unpredictable_data = (float *) malloc(num_blocks*3*sizeof(float));
+ double precision[3];
+ precision[0] = precision_a, precision[1] = precision_b, precision[2] = precision_c;
+ for(int i=0; i<3; i++){
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ coeff_unpred_data[i] = coeff_unpredictable_data + i * num_blocks;
+ }
+ int coeff_index = 0;
+ unsigned int coeff_unpredictable_count[3] = {0};
+ if(use_mean){
+ type = result_type;
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = oriData + offset_x * dim0_offset;
+
+ cur_pb_buf_pos = cur_pb_buf + strip_dim0_offset + 1;
+ next_pb_buf_pos = next_pb_buf + 1;
+ float * pb_pos = cur_pb_buf_pos;
+ float * next_pb_pos = next_pb_buf_pos;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ /*sampling: decide which predictor to use (regression or lorenzo)*/
+ {
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ // [1, 1] [3, 3] [5, 5] [7, 7] [9, 9]
+ // [1, 9] [3, 7] [7, 3] [9, 1]
+ int count = 0;
+ for(int i=1; i<current_blockcount_x; i+=2){
+ cur_data_pos = data_pos + i * dim0_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c];
+
+ err_sz += MIN(fabs(pred_sz - curData) + realPrecision*0.81, fabs(mean - curData));
+
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i * dim0_offset + (block_size - i);
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * (block_size - i) + reg_params_pos[params_offset_c];
+ err_sz += MIN(fabs(pred_sz - curData) + realPrecision*0.81, fabs(mean - curData));
+
+ err_reg += fabs(pred_reg - curData);
+
+ count += 2;
+ }
+
+ use_reg = (err_reg < err_sz);
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<3; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+
+ index ++;
+ cur_data_pos ++;
+ }
+ } // end ii == -1
+ unpredictable_count = block_unpredictable_count;
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }// end use_reg
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_pb_pos = pb_pos;
+ float * cur_data_pos = data_pos;
+ float curData;
+ float pred2D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ next_pb_pos[jj] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[j] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ data_pos += current_blockcount_y;
+ pb_pos += current_blockcount_y;
+ next_pb_pos += current_blockcount_y;
+ type += current_blockcount_x * current_blockcount_y;
+ }// end j
+ indicator_pos += num_y;
+ float * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }// end use mean
+ else{
+ type = result_type;
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = oriData + offset_x * dim0_offset;
+
+ cur_pb_buf_pos = cur_pb_buf + strip_dim0_offset + 1;
+ next_pb_buf_pos = next_pb_buf + 1;
+ float * pb_pos = cur_pb_buf_pos;
+ float * next_pb_pos = next_pb_buf_pos;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ /*sampling*/
+ {
+ // sample [2i + 1, 2i + 1] [2i + 1, bs - 2i]
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ // [1, 1] [3, 3] [5, 5] [7, 7] [9, 9]
+ // [1, 9] [3, 7] [7, 3] [9, 1]
+ int count = 0;
+ for(int i=1; i<current_blockcount_x; i+=2){
+ cur_data_pos = data_pos + i * dim0_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c];
+ err_sz += fabs(pred_sz - curData);
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i * dim0_offset + (block_size - i);
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim0_offset] - cur_data_pos[-dim0_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * (block_size - i) + reg_params_pos[params_offset_c];
+ err_sz += fabs(pred_sz - curData);
+ err_reg += fabs(pred_reg - curData);
+
+ count += 2;
+ }
+ err_sz += realPrecision * count * 0.81;
+ use_reg = (err_reg < err_sz);
+
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<3; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ // jj == current_blockcount_y - 1
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y - 1; jj++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ /*dealing with the last jj (boundary)*/
+ {
+ // jj == current_blockcount_y - 1
+ size_t jj = current_blockcount_y - 1;
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2];
+ diff = curData - pred;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj] = pred;
+ // assign value to next prediction buffer
+ next_pb_pos[jj] = pred;
+
+ index ++;
+ cur_data_pos ++;
+ }
+ } // end ii == -1
+ unpredictable_count = block_unpredictable_count;
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }// end use_reg
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_pb_pos = pb_pos;
+ float * cur_data_pos = data_pos;
+ float curData;
+ float pred2D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y;
+ cur_data_pos += dim0_offset - current_blockcount_y;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ curData = *cur_data_pos;
+
+ pred2D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim0_offset - 1];
+ diff = curData - pred2D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred2D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ next_pb_pos[jj] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[j] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ data_pos += current_blockcount_y;
+ pb_pos += current_blockcount_y;
+ next_pb_pos += current_blockcount_y;
+ type += current_blockcount_x * current_blockcount_y;
+ }// end j
+ indicator_pos += num_y;
+ float * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }
+ free(prediction_buffer_1);
+ free(prediction_buffer_2);
+
+ int stateNum = 2*quantization_intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ size_t nodeCount = 0;
+ size_t i = 0;
+ init(huffmanTree, result_type, num_elements);
+ for (i = 0; i < stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+
+ unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
+ // total size metadata # elements real precision intervals nodeCount huffman block index unpredicatable count mean unpred size elements
+ unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + treeByteSize + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(float) + total_unpred * sizeof(float) + num_elements * sizeof(int), 1);
+ unsigned char * result_pos = result;
+ initRandomAccessBytes(result_pos);
+ result_pos += meta_data_offset;
+
+ sizeToBytes(result_pos, num_elements);
+ result_pos += exe_params->SZ_SIZE_TYPE;
+
+ intToBytes_bigEndian(result_pos, block_size);
+ result_pos += sizeof(int);
+ doubleToBytes(result_pos, realPrecision);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, quantization_intervals);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+
+ memcpy(result_pos, &use_mean, sizeof(unsigned char));
+ result_pos += sizeof(unsigned char);
+ memcpy(result_pos, &mean, sizeof(float));
+ result_pos += sizeof(float);
+
+ size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
+ result_pos += indicator_size;
+
+ //convert the lead/mid/resi to byte stream
+ if(reg_count>0){
+ for(int e=0; e<3; e++){
+ int stateNum = 2*coeff_intvCapacity_sz;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ size_t nodeCount = 0;
+ init(huffmanTree, coeff_type[e], reg_count);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+ doubleToBytes(result_pos, precision[e]);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, coeff_intvRadius);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
+ sizeToBytes(result_pos, typeArray_size);
+ result_pos += sizeof(size_t) + typeArray_size;
+ intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
+ result_pos += sizeof(int);
+ memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(float));
+ result_pos += coeff_unpredictable_count[e]*sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ free(coeff_result_type);
+ free(coeff_unpredictable_data);
+
+ //record the number of unpredictable data and also store them
+ memcpy(result_pos, &total_unpred, sizeof(size_t));
+ result_pos += sizeof(size_t);
+ memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(float));
+ result_pos += total_unpred * sizeof(float);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, result_type, num_elements, result_pos, &typeArray_size);
+ result_pos += typeArray_size;
+
+ size_t totalEncodeSize = result_pos - result;
+ free(indicator);
+ free(result_unpredictable_data);
+ free(result_type);
+ free(reg_params);
+
+ SZ_ReleaseHuffman(huffmanTree);
+ *comp_size = totalEncodeSize;
+
+ return result;
+}
+
+unsigned int optimize_intervals_float_3D_with_freq_and_dense_pos(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, float * dense_pos, float * max_freq, float * mean_freq)
+{
+ float mean = 0.0;
+ size_t len = r1 * r2 * r3;
+ size_t mean_distance = (int) (sqrt(len));
+ float * data_pos = oriData;
+ size_t offset_count = 0;
+ size_t offset_count_2 = 0;
+ size_t mean_count = 0;
+ while(data_pos - oriData < len){
+ mean += *data_pos;
+ mean_count ++;
+ data_pos += mean_distance;
+ offset_count += mean_distance;
+ offset_count_2 += mean_distance;
+ if(offset_count >= r3){
+ offset_count = 0;
+ data_pos -= 1;
+ }
+ if(offset_count_2 >= r2 * r3){
+ offset_count_2 = 0;
+ data_pos -= 1;
+ }
+ }
+ if(mean_count > 0) mean /= mean_count;
+ size_t range = 8192;
+ size_t radius = 4096;
+ size_t * freq_intervals = (size_t *) malloc(range*sizeof(size_t));
+ memset(freq_intervals, 0, range*sizeof(size_t));
+
+ unsigned int maxRangeRadius = confparams_cpr->maxRangeRadius;
+ int sampleDistance = confparams_cpr->sampleDistance;
+ float predThreshold = confparams_cpr->predThreshold;
+
+ size_t i;
+ size_t radiusIndex;
+ size_t r23=r2*r3;
+ float pred_value = 0, pred_err;
+ size_t *intervals = (size_t*)malloc(maxRangeRadius*sizeof(size_t));
+ memset(intervals, 0, maxRangeRadius*sizeof(size_t));
+
+ float mean_diff;
+ ptrdiff_t freq_index;
+ size_t freq_count = 0;
+ size_t sample_count = 0;
+
+ offset_count = confparams_cpr->sampleDistance - 2; // count r3 offset
+ data_pos = oriData + r23 + r3 + offset_count;
+ size_t n1_count = 1, n2_count = 1; // count i,j sum
+
+ while(data_pos - oriData < len){
+
+ 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];
+ pred_err = fabs(pred_value - *data_pos);
+ if(pred_err < realPrecision) freq_count ++;
+ radiusIndex = (pred_err/realPrecision+1)/2;
+ if(radiusIndex>=maxRangeRadius)
+ {
+ radiusIndex = maxRangeRadius - 1;
+ }
+ intervals[radiusIndex]++;
+
+ mean_diff = *data_pos - mean;
+ if(mean_diff > 0) freq_index = (ptrdiff_t)(mean_diff/realPrecision) + radius;
+ else freq_index = (ptrdiff_t)(mean_diff/realPrecision) - 1 + radius;
+ if(freq_index <= 0){
+ freq_intervals[0] ++;
+ }
+ else if(freq_index >= range){
+ freq_intervals[range - 1] ++;
+ }
+ else{
+ freq_intervals[freq_index] ++;
+ }
+ offset_count += sampleDistance;
+ if(offset_count >= r3){
+ n2_count ++;
+ if(n2_count == r2){
+ n1_count ++;
+ n2_count = 1;
+ data_pos += r3;
+ }
+ offset_count_2 = (n1_count + n2_count) % sampleDistance;
+ data_pos += (r3 + sampleDistance - offset_count) + (sampleDistance - offset_count_2);
+ offset_count = (sampleDistance - offset_count_2);
+ if(offset_count == 0) offset_count ++;
+ }
+ else data_pos += sampleDistance;
+ sample_count ++;
+ }
+ *max_freq = freq_count * 1.0/ sample_count;
+
+ //compute the appropriate number
+ size_t targetCount = sample_count*predThreshold;
+ size_t sum = 0;
+ for(i=0;i<maxRangeRadius;i++)
+ {
+ sum += intervals[i];
+ if(sum>targetCount)
+ break;
+ }
+ if(i>=maxRangeRadius)
+ i = maxRangeRadius-1;
+ unsigned int accIntervals = 2*(i+1);
+ unsigned int powerOf2 = roundUpToPowerOf2(accIntervals);
+
+ if(powerOf2<32)
+ powerOf2 = 32;
+ // collect frequency
+ size_t max_sum = 0;
+ size_t max_index = 0;
+ size_t tmp_sum;
+ size_t * freq_pos = freq_intervals + 1;
+ for(size_t i=1; i<range-2; i++){
+ tmp_sum = freq_pos[0] + freq_pos[1];
+ if(tmp_sum > max_sum){
+ max_sum = tmp_sum;
+ max_index = i;
+ }
+ freq_pos ++;
+ }
+ *dense_pos = mean + realPrecision * (ptrdiff_t)(max_index + 1 - radius);
+ *mean_freq = max_sum * 1.0 / sample_count;
+
+ free(freq_intervals);
+ free(intervals);
+ return powerOf2;
+}
+
+
+// 3D: modified for higher performance
+unsigned char * SZ_compress_float_3D_MDQ_nonblocked_with_blocked_regression(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size){
+
+#ifdef HAVE_TIMECMPR
+ float* decData = NULL;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData = (float*)(multisteps->hist_data);
+#endif
+
+ unsigned int quantization_intervals;
+ float sz_sample_correct_freq = -1;//0.5; //-1
+ float dense_pos;
+ float mean_flush_freq;
+ unsigned char use_mean = 0;
+
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ size_t block_size = 6;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r3, num_z, block_size);
+
+ size_t split_index_x, split_index_y, split_index_z;
+ size_t early_blockcount_x, early_blockcount_y, early_blockcount_z;
+ size_t late_blockcount_x, late_blockcount_y, late_blockcount_z;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+ SZ_COMPUTE_BLOCKCOUNT(r3, num_z, split_index_z, early_blockcount_z, late_blockcount_z);
+
+ size_t max_num_block_elements = early_blockcount_x * early_blockcount_y * early_blockcount_z;
+ size_t num_blocks = num_x * num_y * num_z;
+ size_t num_elements = r1 * r2 * r3;
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ size_t unpred_data_max_size = max_num_block_elements;
+ float * result_unpredictable_data = (float *) malloc(unpred_data_max_size * sizeof(float) * num_blocks);
+ size_t total_unpred = 0;
+ size_t unpredictable_count;
+ size_t max_unpred_count = 0;
+ float * data_pos = oriData;
+ int * type = result_type;
+ size_t type_offset;
+ size_t offset_x, offset_y, offset_z;
+ size_t current_blockcount_x, current_blockcount_y, current_blockcount_z;
+
+ float * reg_params = (float *) malloc(num_blocks * 4 * sizeof(float));
+ float * reg_params_pos = reg_params;
+ // move regression part out
+ size_t params_offset_b = num_blocks;
+ size_t params_offset_c = 2*num_blocks;
+ size_t params_offset_d = 3*num_blocks;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+ /*Calculate regression coefficients*/
+ {
+ float * cur_data_pos = data_pos;
+ float fx = 0.0;
+ float fy = 0.0;
+ float fz = 0.0;
+ float f = 0;
+ float sum_x, sum_y;
+ float curData;
+ for(size_t i=0; i<current_blockcount_x; i++){
+ sum_x = 0;
+ for(size_t j=0; j<current_blockcount_y; j++){
+ sum_y = 0;
+ for(size_t k=0; k<current_blockcount_z; k++){
+ curData = *cur_data_pos;
+ // f += curData;
+ // fx += curData * i;
+ // fy += curData * j;
+ // fz += curData * k;
+ sum_y += curData;
+ fz += curData * k;
+ cur_data_pos ++;
+ }
+ fy += sum_y * j;
+ sum_x += sum_y;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ fx += sum_x * i;
+ f += sum_x;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ float coeff = 1.0 / (current_blockcount_x * current_blockcount_y * current_blockcount_z);
+ reg_params_pos[0] = (2 * fx / (current_blockcount_x - 1) - f) * 6 * coeff / (current_blockcount_x + 1);
+ reg_params_pos[params_offset_b] = (2 * fy / (current_blockcount_y - 1) - f) * 6 * coeff / (current_blockcount_y + 1);
+ reg_params_pos[params_offset_c] = (2 * fz / (current_blockcount_z - 1) - f) * 6 * coeff / (current_blockcount_z + 1);
+ reg_params_pos[params_offset_d] = f * coeff - ((current_blockcount_x - 1) * reg_params_pos[0] / 2 + (current_blockcount_y - 1) * reg_params_pos[params_offset_b] / 2 + (current_blockcount_z - 1) * reg_params_pos[params_offset_c] / 2);
+ }
+ reg_params_pos ++;
+ }
+ }
+ }
+
+ //Compress coefficient arrays
+ double precision_a, precision_b, precision_c, precision_d;
+ float rel_param_err = 0.025;
+ precision_a = rel_param_err * realPrecision / late_blockcount_x;
+ precision_b = rel_param_err * realPrecision / late_blockcount_y;
+ precision_c = rel_param_err * realPrecision / late_blockcount_z;
+ precision_d = rel_param_err * realPrecision;
+
+ if(exe_params->optQuantMode==1)
+ {
+ quantization_intervals = optimize_intervals_float_3D_with_freq_and_dense_pos(oriData, r1, r2, r3, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
+ if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
+ updateQuantizationInfo(quantization_intervals);
+ }
+ else{
+ quantization_intervals = exe_params->intvCapacity;
+ }
+
+ float mean = 0;
+ if(use_mean){
+ // compute mean
+ double sum = 0.0;
+ size_t mean_count = 0;
+ for(size_t i=0; i<num_elements; i++){
+ if(fabs(oriData[i] - dense_pos) < realPrecision){
+ sum += oriData[i];
+ mean_count ++;
+ }
+ }
+ if(mean_count > 0) mean = sum / mean_count;
+ }
+
+ double tmp_realPrecision = realPrecision;
+
+ // use two prediction buffers for higher performance
+ float * unpredictable_data = result_unpredictable_data;
+ unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
+ memset(indicator, 0, num_blocks * sizeof(unsigned char));
+ size_t reg_count = 0;
+ size_t strip_dim_0 = early_blockcount_x + 1;
+ size_t strip_dim_1 = r2 + 1;
+ size_t strip_dim_2 = r3 + 1;
+ size_t strip_dim0_offset = strip_dim_1 * strip_dim_2;
+ size_t strip_dim1_offset = strip_dim_2;
+ unsigned char * indicator_pos = indicator;
+
+ size_t prediction_buffer_size = strip_dim_0 * strip_dim0_offset * sizeof(float);
+ float * prediction_buffer_1 = (float *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_1, 0, prediction_buffer_size);
+ float * prediction_buffer_2 = (float *) malloc(prediction_buffer_size);
+ memset(prediction_buffer_2, 0, prediction_buffer_size);
+ float * cur_pb_buf = prediction_buffer_1;
+ float * next_pb_buf = prediction_buffer_2;
+ float * cur_pb_buf_pos;
+ float * next_pb_buf_pos;
+ int intvCapacity = exe_params->intvCapacity;
+ int intvRadius = exe_params->intvRadius;
+ int use_reg = 0;
+ float noise = realPrecision * 1.22;
+
+ reg_params_pos = reg_params;
+ // compress the regression coefficients on the fly
+ float last_coeffcients[4] = {0.0};
+ int coeff_intvCapacity_sz = 65536;
+ int coeff_intvRadius = coeff_intvCapacity_sz / 2;
+ int * coeff_type[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ float * coeff_unpred_data[4];
+ float * coeff_unpredictable_data = (float *) malloc(num_blocks*4*sizeof(float));
+ double precision[4];
+ precision[0] = precision_a, precision[1] = precision_b, precision[2] = precision_c, precision[3] = precision_d;
+ for(int i=0; i<4; i++){
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ coeff_unpred_data[i] = coeff_unpredictable_data + i * num_blocks;
+ }
+ int coeff_index = 0;
+ unsigned int coeff_unpredictable_count[4] = {0};
+
+ if(use_mean){
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset;
+ type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset;
+ type = result_type + type_offset;
+
+ // prediction buffer is (current_block_count_x + 1) * (current_block_count_y + 1) * (current_block_count_z + 1)
+ cur_pb_buf_pos = cur_pb_buf + offset_y * strip_dim1_offset + strip_dim0_offset + strip_dim1_offset + 1;
+ next_pb_buf_pos = next_pb_buf + offset_y * strip_dim1_offset + strip_dim1_offset + 1;
+
+ size_t current_blockcount_z;
+ float * pb_pos = cur_pb_buf_pos;
+ float * next_pb_pos = next_pb_buf_pos;
+ size_t strip_unpredictable_count = 0;
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+#ifdef HAVE_TIMECMPR
+ size_t offset_z = 0;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ size_t block_offset = offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+#endif
+ /*sampling and decide which predictor*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ int bmi = 0;
+ if(i>0 && j>0 && k>0){
+ for(int i=0; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ else{
+ for(int i=1; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ }
+ }
+ use_reg = (err_reg < err_sz);
+ }
+ if(use_reg){
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = pred;
+#endif
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = pred;
+#endif
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ unpredictable_count = block_unpredictable_count;
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_pb_pos = pb_pos;
+ float * cur_data_pos = data_pos;
+ float curData;
+ float pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = *cur_pb_pos;
+#endif
+
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y * strip_dim1_offset;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ // adjust type[index] to intvRadius for coherence with freq in reg
+ type[index] = intvRadius;
+ *cur_pb_pos = mean;
+ }
+ else
+ {
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ if(type[index] <= intvRadius) type[index] -= 1;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+#ifdef HAVE_TIMECMPR
+ size_t ii = current_blockcount_x - 1;
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = *cur_pb_pos;
+#endif
+
+ next_pb_pos[jj * strip_dim1_offset + kk] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+
+ reg_params_pos ++;
+ data_pos += current_blockcount_z;
+ pb_pos += current_blockcount_z;
+ next_pb_pos += current_blockcount_z;
+ type += current_blockcount_x * current_blockcount_y * current_blockcount_z;
+
+ } // end k
+
+ if(strip_unpredictable_count > max_unpred_count){
+ max_unpred_count = strip_unpredictable_count;
+ }
+ total_unpred += strip_unpredictable_count;
+ indicator_pos += num_z;
+ }// end j
+ float * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }// end i
+ }
+ else{
+ int intvCapacity_sz = intvCapacity - 2;
+ for(size_t i=0; i<num_x; i++){
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+
+ for(size_t j=0; j<num_y; j++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ data_pos = oriData + offset_x * dim0_offset + offset_y * dim1_offset;
+ // copy bottom plane from plane buffer
+ // memcpy(prediction_buffer, bottom_buffer + offset_y * strip_dim1_offset, (current_blockcount_y + 1) * strip_dim1_offset * sizeof(float));
+ type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset;
+ type = result_type + type_offset;
+
+ // prediction buffer is (current_block_count_x + 1) * (current_block_count_y + 1) * (current_block_count_z + 1)
+ cur_pb_buf_pos = cur_pb_buf + offset_y * strip_dim1_offset + strip_dim0_offset + strip_dim1_offset + 1;
+ next_pb_buf_pos = next_pb_buf + offset_y * strip_dim1_offset + strip_dim1_offset + 1;
+
+ size_t current_blockcount_z;
+ float * pb_pos = cur_pb_buf_pos;
+ float * next_pb_pos = next_pb_buf_pos;
+ size_t strip_unpredictable_count = 0;
+ for(size_t k=0; k<num_z; k++){
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+#ifdef HAVE_TIMECMPR
+ size_t offset_z = 0;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ size_t block_offset = offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+#endif
+ /*sampling*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ int bmi;
+ if(i>0 && j>0 && k>0){
+ for(int i=0; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ else{
+ for(int i=1; i<block_size; i++){
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = data_pos + i*dim0_offset + i*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = data_pos + i*dim0_offset + bmi*dim1_offset + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-dim1_offset]+ cur_data_pos[-dim0_offset] - cur_data_pos[-dim1_offset - 1] - cur_data_pos[-dim0_offset - 1] - cur_data_pos[-dim0_offset - dim1_offset] + cur_data_pos[-dim0_offset - dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+ }
+ }
+ use_reg = (err_reg < err_sz);
+
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = data_pos;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = pred;
+#endif
+
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ size_t ii = current_blockcount_x - 1;
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = pred;
+#endif
+
+ if((jj == current_blockcount_y - 1) || (kk == current_blockcount_z - 1)){
+ // assign value to block surfaces
+ pb_pos[ii * strip_dim0_offset + jj * strip_dim1_offset + kk] = pred;
+ }
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = pred;
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ unpredictable_count = block_unpredictable_count;
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_pb_pos = pb_pos;
+ float * cur_data_pos = data_pos;
+ float curData;
+ float pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<current_blockcount_x - 1; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = *cur_pb_pos;
+#endif
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ cur_pb_pos += strip_dim0_offset - current_blockcount_y * strip_dim1_offset;
+ cur_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ /*dealing with the last ii (boundary)*/
+ {
+ // ii == current_blockcount_x - 1
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+
+ curData = *cur_data_pos;
+ pred3D = cur_pb_pos[-1] + cur_pb_pos[-strip_dim1_offset]+ cur_pb_pos[-strip_dim0_offset] - cur_pb_pos[-strip_dim1_offset - 1]
+ - cur_pb_pos[-strip_dim0_offset - 1] - cur_pb_pos[-strip_dim0_offset - strip_dim1_offset] + cur_pb_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_pb_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_pb_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_pb_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+
+#ifdef HAVE_TIMECMPR
+ size_t ii = current_blockcount_x - 1;
+ size_t point_offset = ii*dim0_offset + jj*dim1_offset + kk;
+ if(confparams_cpr->szMode == SZ_TEMPORAL_COMPRESSION)
+ decData[block_offset + point_offset] = *cur_pb_pos;
+#endif
+
+ // assign value to next prediction buffer
+ next_pb_pos[jj * strip_dim1_offset + kk] = *cur_pb_pos;
+ index ++;
+ cur_pb_pos ++;
+ cur_data_pos ++;
+ }
+ cur_pb_pos += strip_dim1_offset - current_blockcount_z;
+ cur_data_pos += dim1_offset - current_blockcount_z;
+ }
+ }
+ strip_unpredictable_count += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+
+ reg_params_pos ++;
+ data_pos += current_blockcount_z;
+ pb_pos += current_blockcount_z;
+ next_pb_pos += current_blockcount_z;
+ type += current_blockcount_x * current_blockcount_y * current_blockcount_z;
+
+ }
+
+ if(strip_unpredictable_count > max_unpred_count){
+ max_unpred_count = strip_unpredictable_count;
+ }
+ total_unpred += strip_unpredictable_count;
+ indicator_pos += num_z;
+ }
+ float * tmp;
+ tmp = cur_pb_buf;
+ cur_pb_buf = next_pb_buf;
+ next_pb_buf = tmp;
+ }
+ }
+
+ free(prediction_buffer_1);
+ free(prediction_buffer_2);
+
+ int stateNum = 2*quantization_intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ size_t nodeCount = 0;
+ init(huffmanTree, result_type, num_elements);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+
+ unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
+ // total size metadata # elements real precision intervals nodeCount huffman block index unpredicatable count mean unpred size elements
+ unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + treeByteSize + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(float) + total_unpred * sizeof(float) + num_elements * sizeof(int), 1);
+ unsigned char * result_pos = result;
+ initRandomAccessBytes(result_pos);
+
+ result_pos += meta_data_offset;
+
+ sizeToBytes(result_pos,num_elements); //SZ_SIZE_TYPE: 4 or 8
+ result_pos += exe_params->SZ_SIZE_TYPE;
+
+ intToBytes_bigEndian(result_pos, block_size);
+ result_pos += sizeof(int);
+ doubleToBytes(result_pos, realPrecision);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, quantization_intervals);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+
+ memcpy(result_pos, &use_mean, sizeof(unsigned char));
+ result_pos += sizeof(unsigned char);
+ memcpy(result_pos, &mean, sizeof(float));
+ result_pos += sizeof(float);
+ size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
+ result_pos += indicator_size;
+
+ //convert the lead/mid/resi to byte stream
+ if(reg_count > 0){
+ for(int e=0; e<4; e++){
+ int stateNum = 2*coeff_intvCapacity_sz;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ size_t nodeCount = 0;
+ init(huffmanTree, coeff_type[e], reg_count);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+ doubleToBytes(result_pos, precision[e]);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, coeff_intvRadius);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
+ sizeToBytes(result_pos, typeArray_size);
+ result_pos += sizeof(size_t) + typeArray_size;
+ intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
+ result_pos += sizeof(int);
+ memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(float));
+ result_pos += coeff_unpredictable_count[e]*sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ free(coeff_result_type);
+ free(coeff_unpredictable_data);
+
+ //record the number of unpredictable data and also store them
+ memcpy(result_pos, &total_unpred, sizeof(size_t));
+ result_pos += sizeof(size_t);
+ memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(float));
+ result_pos += total_unpred * sizeof(float);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, result_type, num_elements, result_pos, &typeArray_size);
+ result_pos += typeArray_size;
+ size_t totalEncodeSize = result_pos - result;
+ free(indicator);
+ free(result_unpredictable_data);
+ free(result_type);
+ free(reg_params);
+
+
+ SZ_ReleaseHuffman(huffmanTree);
+ *comp_size = totalEncodeSize;
+ return result;
+}
+
+unsigned char * SZ_compress_float_3D_MDQ_random_access_with_blocked_regression(float *oriData, size_t r1, size_t r2, size_t r3, double realPrecision, size_t * comp_size){
+
+ unsigned int quantization_intervals;
+ float sz_sample_correct_freq = -1;//0.5; //-1
+ float dense_pos;
+ float mean_flush_freq;
+ unsigned char use_mean = 0;
+
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ size_t block_size = 6;
+ num_x = (r1 - 1) / block_size + 1;
+ num_y = (r2 - 1) / block_size + 1;
+ num_z = (r3 - 1) / block_size + 1;
+
+ size_t max_num_block_elements = block_size * block_size * block_size;
+ size_t num_blocks = num_x * num_y * num_z;
+ size_t num_elements = r1 * r2 * r3;
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+
+ int * result_type = (int *) malloc(num_blocks*max_num_block_elements * sizeof(int));
+ size_t unpred_data_max_size = max_num_block_elements;
+ float * result_unpredictable_data = (float *) malloc(unpred_data_max_size * sizeof(float) * num_blocks);
+ size_t total_unpred = 0;
+ size_t unpredictable_count;
+ float * data_pos = oriData;
+ int * type = result_type;
+ float * reg_params = (float *) malloc(num_blocks * 4 * sizeof(float));
+ float * reg_params_pos = reg_params;
+ // move regression part out
+ size_t params_offset_b = num_blocks;
+ size_t params_offset_c = 2*num_blocks;
+ size_t params_offset_d = 3*num_blocks;
+ float * pred_buffer = (float *) malloc((block_size+1)*(block_size+1)*(block_size+1)*sizeof(float));
+ float * pred_buffer_pos = NULL;
+ float * block_data_pos_x = NULL;
+ float * block_data_pos_y = NULL;
+ float * block_data_pos_z = NULL;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ data_pos = oriData + i*block_size * dim0_offset + j*block_size * dim1_offset + k*block_size;
+ pred_buffer_pos = pred_buffer;
+ block_data_pos_x = data_pos;
+ // use the buffer as block_size*block_size*block_size
+ for(int ii=0; ii<block_size; ii++){
+ block_data_pos_y = block_data_pos_x;
+ for(int jj=0; jj<block_size; jj++){
+ block_data_pos_z = block_data_pos_y;
+ for(int kk=0; kk<block_size; kk++){
+ *pred_buffer_pos = *block_data_pos_z;
+ if(k*block_size + kk + 1 < r3) block_data_pos_z ++;
+ pred_buffer_pos ++;
+ }
+ if(j*block_size + jj + 1 < r2) block_data_pos_y += dim1_offset;
+ }
+ if(i*block_size + ii + 1 < r1) block_data_pos_x += dim0_offset;
+ }
+ /*Calculate regression coefficients*/
+ {
+ float * cur_data_pos = pred_buffer;
+ float fx = 0.0;
+ float fy = 0.0;
+ float fz = 0.0;
+ float f = 0;
+ float sum_x, sum_y;
+ float curData;
+ for(size_t i=0; i<block_size; i++){
+ sum_x = 0;
+ for(size_t j=0; j<block_size; j++){
+ sum_y = 0;
+ for(size_t k=0; k<block_size; k++){
+ curData = *cur_data_pos;
+ sum_y += curData;
+ fz += curData * k;
+ cur_data_pos ++;
+ }
+ fy += sum_y * j;
+ sum_x += sum_y;
+ }
+ fx += sum_x * i;
+ f += sum_x;
+ }
+ float coeff = 1.0 / (block_size * block_size * block_size);
+ reg_params_pos[0] = (2 * fx / (block_size - 1) - f) * 6 * coeff / (block_size + 1);
+ reg_params_pos[params_offset_b] = (2 * fy / (block_size - 1) - f) * 6 * coeff / (block_size + 1);
+ reg_params_pos[params_offset_c] = (2 * fz / (block_size - 1) - f) * 6 * coeff / (block_size + 1);
+ reg_params_pos[params_offset_d] = f * coeff - ((block_size - 1) * reg_params_pos[0] / 2 + (block_size - 1) * reg_params_pos[params_offset_b] / 2 + (block_size - 1) * reg_params_pos[params_offset_c] / 2);
+ }
+ reg_params_pos ++;
+ }
+ }
+ }
+
+ //Compress coefficient arrays
+ double precision_a, precision_b, precision_c, precision_d;
+ float rel_param_err = 0.025;
+ precision_a = rel_param_err * realPrecision / block_size;
+ precision_b = rel_param_err * realPrecision / block_size;
+ precision_c = rel_param_err * realPrecision / block_size;
+ precision_d = rel_param_err * realPrecision;
+
+ if(exe_params->optQuantMode==1)
+ {
+ quantization_intervals = optimize_intervals_float_3D_with_freq_and_dense_pos(oriData, r1, r2, r3, realPrecision, &dense_pos, &sz_sample_correct_freq, &mean_flush_freq);
+ if(mean_flush_freq > 0.5 || mean_flush_freq > sz_sample_correct_freq) use_mean = 1;
+ updateQuantizationInfo(quantization_intervals);
+ }
+ else{
+ quantization_intervals = exe_params->intvCapacity;
+ }
+
+ float mean = 0;
+ if(use_mean){
+ // compute mean
+ double sum = 0.0;
+ size_t mean_count = 0;
+ for(size_t i=0; i<num_elements; i++){
+ if(fabs(oriData[i] - dense_pos) < realPrecision){
+ sum += oriData[i];
+ mean_count ++;
+ }
+ }
+ if(mean_count > 0) mean = sum / mean_count;
+ }
+
+ double tmp_realPrecision = realPrecision;
+
+ // use two prediction buffers for higher performance
+ float * unpredictable_data = result_unpredictable_data;
+ unsigned char * indicator = (unsigned char *) malloc(num_blocks * sizeof(unsigned char));
+ memset(indicator, 0, num_blocks * sizeof(unsigned char));
+ size_t reg_count = 0;
+ unsigned char * indicator_pos = indicator;
+
+ int intvCapacity = exe_params->intvCapacity;
+ int intvRadius = exe_params->intvRadius;
+ int use_reg = 0;
+ float noise = realPrecision * 1.22;
+
+ reg_params_pos = reg_params;
+ // compress the regression coefficients on the fly
+ float last_coeffcients[4] = {0.0};
+ int coeff_intvCapacity_sz = 65536;
+ int coeff_intvRadius = coeff_intvCapacity_sz / 2;
+ int * coeff_type[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ float * coeff_unpred_data[4];
+ float * coeff_unpredictable_data = (float *) malloc(num_blocks*4*sizeof(float));
+ double precision[4];
+ precision[0] = precision_a, precision[1] = precision_b, precision[2] = precision_c, precision[3] = precision_d;
+ for(int i=0; i<4; i++){
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ coeff_unpred_data[i] = coeff_unpredictable_data + i * num_blocks;
+ }
+ int coeff_index = 0;
+ unsigned int coeff_unpredictable_count[4] = {0};
+
+ memset(pred_buffer, 0, (block_size+1)*(block_size+1)*(block_size+1)*sizeof(float));
+ int pred_buffer_block_size = block_size + 1;
+ int strip_dim0_offset = pred_buffer_block_size * pred_buffer_block_size;
+ int strip_dim1_offset = pred_buffer_block_size;
+
+ if(use_mean){
+ int intvCapacity_sz = intvCapacity - 2;
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ data_pos = oriData + i*block_size * dim0_offset + j*block_size * dim1_offset + k*block_size;
+ // add 1 in x, y, z offset
+ pred_buffer_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ block_data_pos_x = data_pos;
+ for(int ii=0; ii<block_size; ii++){
+ block_data_pos_y = block_data_pos_x;
+ for(int jj=0; jj<block_size; jj++){
+ block_data_pos_z = block_data_pos_y;
+ for(int kk=0; kk<block_size; kk++){
+ *pred_buffer_pos = *block_data_pos_z;
+ if(k*block_size + kk + 1< r3) block_data_pos_z ++;
+ pred_buffer_pos ++;
+ }
+ // add 1 in z offset
+ pred_buffer_pos ++;
+ if(j*block_size + jj + 1< r2) block_data_pos_y += dim1_offset;
+ }
+ // add 1 in y offset
+ pred_buffer_pos += pred_buffer_block_size;
+ if(i*block_size + ii + 1< r1) block_data_pos_x += dim0_offset;
+ }
+ /*sampling and decide which predictor*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ int bmi = 0;
+ for(int i=2; i<=block_size; i++){
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + i*pred_buffer_block_size + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + i*pred_buffer_block_size + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + bmi*pred_buffer_block_size + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + bmi*pred_buffer_block_size + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += MIN(fabs(pred_sz - curData) + noise, fabs(mean - curData));
+ err_reg += fabs(pred_reg - curData);
+ }
+
+ use_reg = (err_reg < err_sz);
+ }
+ if(use_reg){
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos ++;
+ }
+ cur_data_pos += pred_buffer_block_size;
+ }
+
+ total_unpred += block_unpredictable_count;
+ unpredictable_data += block_unpredictable_count;
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_data_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ float curData;
+ float pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+
+ curData = *cur_data_pos;
+ if(fabs(curData - mean) <= realPrecision){
+ type[index] = 1;
+ *cur_data_pos = mean;
+ }
+ else
+ {
+ pred3D = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1]
+ - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_data_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_data_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_data_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_data_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos ++;
+ }
+ cur_data_pos += pred_buffer_block_size;
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ type += block_size * block_size * block_size;
+ } // end k
+ indicator_pos += num_z;
+ }// end j
+ }// end i
+ }
+ else{
+ int intvCapacity_sz = intvCapacity - 2;
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ data_pos = oriData + i*block_size * dim0_offset + j*block_size * dim1_offset + k*block_size;
+ // add 1 in x, y, z offset
+ pred_buffer_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ block_data_pos_x = data_pos;
+ for(int ii=0; ii<block_size; ii++){
+ block_data_pos_y = block_data_pos_x;
+ for(int jj=0; jj<block_size; jj++){
+ block_data_pos_z = block_data_pos_y;
+ for(int kk=0; kk<block_size; kk++){
+ *pred_buffer_pos = *block_data_pos_z;
+ if(k*block_size + kk < r3) block_data_pos_z ++;
+ pred_buffer_pos ++;
+ }
+ // add 1 in z offset
+ pred_buffer_pos ++;
+ if(j*block_size + jj < r2) block_data_pos_y += dim1_offset;
+ }
+ // add 1 in y offset
+ pred_buffer_pos += pred_buffer_block_size;
+ if(i*block_size + ii < r1) block_data_pos_x += dim0_offset;
+ }
+ /*sampling*/
+ {
+ // sample point [1, 1, 1] [1, 1, 4] [1, 4, 1] [1, 4, 4] [4, 1, 1] [4, 1, 4] [4, 4, 1] [4, 4, 4]
+ float * cur_data_pos;
+ float curData;
+ float pred_reg, pred_sz;
+ float err_sz = 0.0, err_reg = 0.0;
+ int bmi;
+ for(int i=2; i<=block_size; i++){
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + i*pred_buffer_block_size + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ bmi = block_size - i;
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + i*pred_buffer_block_size + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * i + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + bmi*pred_buffer_block_size + i;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * i + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+
+ cur_data_pos = pred_buffer + i*pred_buffer_block_size*pred_buffer_block_size + bmi*pred_buffer_block_size + bmi;
+ curData = *cur_data_pos;
+ pred_sz = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1] - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ pred_reg = reg_params_pos[0] * i + reg_params_pos[params_offset_b] * bmi + reg_params_pos[params_offset_c] * bmi + reg_params_pos[params_offset_d];
+ err_sz += fabs(pred_sz - curData) + noise;
+ err_reg += fabs(pred_reg - curData);
+ }
+
+ use_reg = (err_reg < err_sz);
+
+ }
+ if(use_reg)
+ {
+ {
+ /*predict coefficients in current block via previous reg_block*/
+ float cur_coeff;
+ double diff, itvNum;
+ for(int e=0; e<4; e++){
+ cur_coeff = reg_params_pos[e*num_blocks];
+ diff = cur_coeff - last_coeffcients[e];
+ itvNum = fabs(diff)/precision[e] + 1;
+ if (itvNum < coeff_intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ coeff_type[e][coeff_index] = (int) (itvNum/2) + coeff_intvRadius;
+ last_coeffcients[e] = last_coeffcients[e] + 2 * (coeff_type[e][coeff_index] - coeff_intvRadius) * precision[e];
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(cur_coeff - last_coeffcients[e])>precision[e]){
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ else{
+ coeff_type[e][coeff_index] = 0;
+ last_coeffcients[e] = cur_coeff;
+ coeff_unpred_data[e][coeff_unpredictable_count[e] ++] = cur_coeff;
+ }
+ }
+ coeff_index ++;
+ }
+ float curData;
+ float pred;
+ double itvNum;
+ double diff;
+ size_t index = 0;
+ size_t block_unpredictable_count = 0;
+ float * cur_data_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ curData = *cur_data_pos;
+ pred = last_coeffcients[0] * ii + last_coeffcients[1] * jj + last_coeffcients[2] * kk + last_coeffcients[3];
+ diff = curData - pred;
+ itvNum = fabs(diff)/tmp_realPrecision + 1;
+ if (itvNum < intvCapacity){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ pred = pred + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - pred)>tmp_realPrecision){
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ pred = curData;
+ unpredictable_data[block_unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos ++;
+ }
+ cur_data_pos += pred_buffer_block_size;
+ }
+ total_unpred += block_unpredictable_count;
+ unpredictable_data += block_unpredictable_count;
+ reg_count ++;
+ }
+ else{
+ // use SZ
+ // SZ predication
+ unpredictable_count = 0;
+ float * cur_data_pos = pred_buffer + pred_buffer_block_size*pred_buffer_block_size + pred_buffer_block_size + 1;
+ float curData;
+ float pred3D;
+ double itvNum, diff;
+ size_t index = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ curData = *cur_data_pos;
+ pred3D = cur_data_pos[-1] + cur_data_pos[-strip_dim1_offset]+ cur_data_pos[-strip_dim0_offset] - cur_data_pos[-strip_dim1_offset - 1]
+ - cur_data_pos[-strip_dim0_offset - 1] - cur_data_pos[-strip_dim0_offset - strip_dim1_offset] + cur_data_pos[-strip_dim0_offset - strip_dim1_offset - 1];
+ diff = curData - pred3D;
+ itvNum = fabs(diff)/realPrecision + 1;
+ if (itvNum < intvCapacity_sz){
+ if (diff < 0) itvNum = -itvNum;
+ type[index] = (int) (itvNum/2) + intvRadius;
+ *cur_data_pos = pred3D + 2 * (type[index] - intvRadius) * tmp_realPrecision;
+ //ganrantee comporession error against the case of machine-epsilon
+ if(fabs(curData - *cur_data_pos)>tmp_realPrecision){
+ type[index] = 0;
+ *cur_data_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ }
+ else{
+ type[index] = 0;
+ *cur_data_pos = curData;
+ unpredictable_data[unpredictable_count ++] = curData;
+ }
+ index ++;
+ cur_data_pos ++;
+ }
+ cur_data_pos ++;
+ }
+ cur_data_pos += pred_buffer_block_size;
+ }
+ total_unpred += unpredictable_count;
+ unpredictable_data += unpredictable_count;
+ // change indicator
+ indicator_pos[k] = 1;
+ }// end SZ
+ reg_params_pos ++;
+ type += block_size * block_size * block_size;
+ }
+ indicator_pos += num_z;
+ }
+ }
+ }
+ free(pred_buffer);
+ int stateNum = 2*quantization_intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ size_t nodeCount = 0;
+ init(huffmanTree, result_type, num_blocks*max_num_block_elements);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+
+ unsigned int meta_data_offset = 3 + 1 + MetaDataByteLength;
+ // total size metadata # elements real precision intervals nodeCount huffman block index unpredicatable count mean unpred size elements
+ unsigned char * result = (unsigned char *) calloc(meta_data_offset + exe_params->SZ_SIZE_TYPE + sizeof(double) + sizeof(int) + sizeof(int) + treeByteSize + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(unsigned short) + num_blocks * sizeof(float) + total_unpred * sizeof(float) + num_elements * sizeof(int), 1);
+ unsigned char * result_pos = result;
+ initRandomAccessBytes(result_pos);
+
+ result_pos += meta_data_offset;
+
+ sizeToBytes(result_pos,num_elements); //SZ_SIZE_TYPE: 4 or 8
+ result_pos += exe_params->SZ_SIZE_TYPE;
+
+ intToBytes_bigEndian(result_pos, block_size);
+ result_pos += sizeof(int);
+ doubleToBytes(result_pos, realPrecision);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, quantization_intervals);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+
+ memcpy(result_pos, &use_mean, sizeof(unsigned char));
+ result_pos += sizeof(unsigned char);
+ memcpy(result_pos, &mean, sizeof(float));
+ result_pos += sizeof(float);
+ size_t indicator_size = convertIntArray2ByteArray_fast_1b_to_result(indicator, num_blocks, result_pos);
+ result_pos += indicator_size;
+
+ //convert the lead/mid/resi to byte stream
+ if(reg_count > 0){
+ for(int e=0; e<4; e++){
+ int stateNum = 2*coeff_intvCapacity_sz;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ size_t nodeCount = 0;
+ init(huffmanTree, coeff_type[e], reg_count);
+ size_t i = 0;
+ for (i = 0; i < huffmanTree->stateNum; i++)
+ if (huffmanTree->code[i]) nodeCount++;
+ nodeCount = nodeCount*2-1;
+ unsigned char *treeBytes;
+ unsigned int treeByteSize = convert_HuffTree_to_bytes_anyStates(huffmanTree, nodeCount, &treeBytes);
+ doubleToBytes(result_pos, precision[e]);
+ result_pos += sizeof(double);
+ intToBytes_bigEndian(result_pos, coeff_intvRadius);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, treeByteSize);
+ result_pos += sizeof(int);
+ intToBytes_bigEndian(result_pos, nodeCount);
+ result_pos += sizeof(int);
+ memcpy(result_pos, treeBytes, treeByteSize);
+ result_pos += treeByteSize;
+ free(treeBytes);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, coeff_type[e], reg_count, result_pos + sizeof(size_t), &typeArray_size);
+ sizeToBytes(result_pos, typeArray_size);
+ result_pos += sizeof(size_t) + typeArray_size;
+ intToBytes_bigEndian(result_pos, coeff_unpredictable_count[e]);
+ result_pos += sizeof(int);
+ memcpy(result_pos, coeff_unpred_data[e], coeff_unpredictable_count[e]*sizeof(float));
+ result_pos += coeff_unpredictable_count[e]*sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ free(coeff_result_type);
+ free(coeff_unpredictable_data);
+
+ //record the number of unpredictable data and also store them
+ memcpy(result_pos, &total_unpred, sizeof(size_t));
+ result_pos += sizeof(size_t);
+ memcpy(result_pos, result_unpredictable_data, total_unpred * sizeof(float));
+ result_pos += total_unpred * sizeof(float);
+ size_t typeArray_size = 0;
+ encode(huffmanTree, result_type, num_blocks*max_num_block_elements, result_pos, &typeArray_size);
+ result_pos += typeArray_size;
+ size_t totalEncodeSize = result_pos - result;
+ free(indicator);
+ free(result_unpredictable_data);
+ free(result_type);
+ free(reg_params);
+
+
+ SZ_ReleaseHuffman(huffmanTree);
+ *comp_size = totalEncodeSize;
+ return result;
+}
diff --git a/thirdparty/SZ/sz/src/sz_float_pwr.c b/thirdparty/SZ/sz/src/sz_float_pwr.c
index 644afdd..92a449f 100644
--- a/thirdparty/SZ/sz/src/sz_float_pwr.c
+++ b/thirdparty/SZ/sz/src/sz_float_pwr.c
@@ -23,6 +23,7 @@
#include "sz_float_pwr.h"
#include "zlib.h"
#include "rw.h"
+#include "utility.h"
void compute_segment_precisions_float_1D(float *oriData, size_t dataLength, float* pwrErrBound, unsigned char* pwrErrBoundBytes, double globalPrecision)
{
@@ -1780,3 +1781,190 @@ size_t dataLength, double absErrBound, double relBoundRatio, double pwrErrRatio,
free_TightDataPointStorageF(tdps);
}
+
+#include <stdbool.h>
+
+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){
+
+ float * log_data = (float *) malloc(dataLength * sizeof(float));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ float max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ float min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ float valueRangeSize, medianValue_f;
+ computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+
+ TightDataPointStorageF* tdps = SZ_compress_float_1D_MDQ(log_data, dataLength, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(float))
+ SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageF(tdps);
+}
+
+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){
+
+ size_t dataLength = r1 * r2;
+ float * log_data = (float *) malloc(dataLength * sizeof(float));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ float max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ float min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ float valueRangeSize, medianValue_f;
+ computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+
+ TightDataPointStorageF* tdps = SZ_compress_float_2D_MDQ(log_data, r1, r2, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(float))
+ SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageF(tdps);
+}
+
+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){
+
+ size_t dataLength = r1 * r2 * r3;
+ float * log_data = (float *) malloc(dataLength * sizeof(float));
+
+ unsigned char * signs = (unsigned char *) malloc(dataLength);
+ memset(signs, 0, dataLength);
+ // preprocess
+ float max_abs_log_data;
+ if(min == 0) max_abs_log_data = fabs(log2(fabs(max)));
+ else if(max == 0) max_abs_log_data = fabs(log2(fabs(min)));
+ else max_abs_log_data = fabs(log2(fabs(min))) > fabs(log2(fabs(max))) ? fabs(log2(fabs(min))) : fabs(log2(fabs(max)));
+ float min_log_data = max_abs_log_data;
+ bool positive = true;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] < 0){
+ signs[i] = 1;
+ log_data[i] = -oriData[i];
+ positive = false;
+ }
+ else
+ log_data[i] = oriData[i];
+ if(log_data[i] > 0){
+ log_data[i] = log2(log_data[i]);
+ if(log_data[i] > max_abs_log_data) max_abs_log_data = log_data[i];
+ if(log_data[i] < min_log_data) min_log_data = log_data[i];
+ }
+ }
+
+ float valueRangeSize, medianValue_f;
+ computeRangeSize_float(log_data, dataLength, &valueRangeSize, &medianValue_f);
+ if(fabs(min_log_data) > max_abs_log_data) max_abs_log_data = fabs(min_log_data);
+ double realPrecision = log2(1.0 + pwrErrRatio) - max_abs_log_data * 1.2e-7;
+ for(size_t i=0; i<dataLength; i++){
+ if(oriData[i] == 0){
+ log_data[i] = min_log_data - 2.0001*realPrecision;
+ }
+ }
+
+ TightDataPointStorageF* tdps = SZ_compress_float_3D_MDQ(log_data, r1, r2, r3, realPrecision, valueRangeSize, medianValue_f);
+ tdps->minLogValue = min_log_data - 1.0001*realPrecision;
+ free(log_data);
+ if(!positive){
+ unsigned char * comp_signs;
+ // compress signs
+ unsigned long signSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, signs, dataLength, &comp_signs);
+ tdps->pwrErrBoundBytes = comp_signs;
+ tdps->pwrErrBoundBytes_size = signSize;
+ }
+ else{
+ tdps->pwrErrBoundBytes = NULL;
+ tdps->pwrErrBoundBytes_size = 0;
+ }
+ free(signs);
+
+ convertTDPStoFlatBytes_float(tdps, newByteData, outSize);
+ if(*outSize>dataLength*sizeof(float))
+ SZ_compress_args_float_StoreOriData(oriData, dataLength+2, tdps, newByteData, outSize);
+
+ free_TightDataPointStorageF(tdps);
+}
diff --git a/thirdparty/SZ/sz/src/sz_int16.c b/thirdparty/SZ/sz/src/sz_int16.c
index fc91dd1..8226234 100644
--- a/thirdparty/SZ/sz/src/sz_int16.c
+++ b/thirdparty/SZ/sz/src/sz_int16.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_int16.h"
+#include "utility.h"
unsigned int optimize_intervals_int16_1D(int16_t *oriData, size_t dataLength, double realPrecision)
{
@@ -266,7 +267,7 @@ TightDataPointStorageI* SZ_compress_int16_1D_MDQ(int16_t *oriData, size_t dataLe
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1369,7 +1370,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_int32.c b/thirdparty/SZ/sz/src/sz_int32.c
index bcd9783..6c5a66e 100644
--- a/thirdparty/SZ/sz/src/sz_int32.c
+++ b/thirdparty/SZ/sz/src/sz_int32.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_int32.h"
+#include "utility.h"
unsigned int optimize_intervals_int32_1D(int32_t *oriData, size_t dataLength, double realPrecision)
{
@@ -268,7 +269,7 @@ TightDataPointStorageI* SZ_compress_int32_1D_MDQ(int32_t *oriData, size_t dataLe
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1253,7 +1254,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_int64.c b/thirdparty/SZ/sz/src/sz_int64.c
index eb97377..065fb16 100644
--- a/thirdparty/SZ/sz/src/sz_int64.c
+++ b/thirdparty/SZ/sz/src/sz_int64.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_int64.h"
+#include "utility.h"
unsigned int optimize_intervals_int64_1D(int64_t *oriData, size_t dataLength, double realPrecision)
{
@@ -269,7 +270,7 @@ TightDataPointStorageI* SZ_compress_int64_1D_MDQ(int64_t *oriData, size_t dataLe
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1254,7 +1255,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_int8.c b/thirdparty/SZ/sz/src/sz_int8.c
index c869681..83febd0 100644
--- a/thirdparty/SZ/sz/src/sz_int8.c
+++ b/thirdparty/SZ/sz/src/sz_int8.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_int8.h"
+#include "utility.h"
unsigned int optimize_intervals_int8_1D(int8_t *oriData, size_t dataLength, double realPrecision)
{
@@ -266,7 +267,7 @@ TightDataPointStorageI* SZ_compress_int8_1D_MDQ(int8_t *oriData, size_t dataLeng
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1370,7 +1371,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_uint16.c b/thirdparty/SZ/sz/src/sz_uint16.c
index 4200b31..b55f007 100644
--- a/thirdparty/SZ/sz/src/sz_uint16.c
+++ b/thirdparty/SZ/sz/src/sz_uint16.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_uint16.h"
+#include "utility.h"
unsigned int optimize_intervals_uint16_1D(uint16_t *oriData, size_t dataLength, double realPrecision)
{
@@ -266,7 +267,7 @@ TightDataPointStorageI* SZ_compress_uint16_1D_MDQ(uint16_t *oriData, size_t data
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1369,7 +1370,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_uint32.c b/thirdparty/SZ/sz/src/sz_uint32.c
index 29d5961..6f27510 100644
--- a/thirdparty/SZ/sz/src/sz_uint32.c
+++ b/thirdparty/SZ/sz/src/sz_uint32.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_uint32.h"
+#include "utility.h"
unsigned int optimize_intervals_uint32_1D(uint32_t *oriData, size_t dataLength, double realPrecision)
{
@@ -268,7 +269,7 @@ TightDataPointStorageI* SZ_compress_uint32_1D_MDQ(uint32_t *oriData, size_t data
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1253,7 +1254,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_uint64.c b/thirdparty/SZ/sz/src/sz_uint64.c
index b8cb8bc..7d2eca8 100644
--- a/thirdparty/SZ/sz/src/sz_uint64.c
+++ b/thirdparty/SZ/sz/src/sz_uint64.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_uint64.h"
+#include "utility.h"
unsigned int optimize_intervals_uint64_1D(uint64_t *oriData, size_t dataLength, double realPrecision)
{
@@ -268,7 +269,7 @@ TightDataPointStorageI* SZ_compress_uint64_1D_MDQ(uint64_t *oriData, size_t data
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1253,7 +1254,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/sz_uint8.c b/thirdparty/SZ/sz/src/sz_uint8.c
index 6ca4ae4..6865564 100644
--- a/thirdparty/SZ/sz/src/sz_uint8.c
+++ b/thirdparty/SZ/sz/src/sz_uint8.c
@@ -21,6 +21,7 @@
#include "rw.h"
#include "TightDataPointStorageI.h"
#include "sz_uint8.h"
+#include "utility.h"
unsigned int optimize_intervals_uint8_1D(uint8_t *oriData, size_t dataLength, double realPrecision)
{
@@ -266,7 +267,7 @@ TightDataPointStorageI* SZ_compress_uint8_1D_MDQ(uint8_t *oriData, size_t dataLe
//pred = 2*last3CmprsData[0] - last3CmprsData[1];
pred = last3CmprsData[0];
predAbsErr = llabs(curData - pred);
- if(predAbsErr<=checkRadius)
+ if(predAbsErr<checkRadius)
{
state = (predAbsErr/realPrecision+1)/2;
if(curData>=pred)
@@ -1370,7 +1371,7 @@ int errBoundMode, double absErr_Bound, double relBoundRatio)
}
else if(confparams_cpr->szMode==SZ_BEST_COMPRESSION || confparams_cpr->szMode==SZ_DEFAULT_COMPRESSION)
{
- *outSize = zlib_compress5(tmpByteData, tmpOutSize, newByteData, confparams_cpr->gzipMode);
+ *outSize = sz_lossless_compress(confparams_cpr->losslessCompressor, confparams_cpr->gzipMode, tmpByteData, tmpOutSize, newByteData);
free(tmpByteData);
}
else
diff --git a/thirdparty/SZ/sz/src/szd_double.c b/thirdparty/SZ/sz/src/szd_double.c
index 1440e2d..09d5851 100644
--- a/thirdparty/SZ/sz/src/szd_double.c
+++ b/thirdparty/SZ/sz/src/szd_double.c
@@ -16,6 +16,7 @@
#include "Huffman.h"
#include "szd_double_pwr.h"
#include "szd_double_ts.h"
+#include "utility.h"
int SZ_decompress_args_double(double** newData, size_t r5, size_t r4, size_t r3, size_t r2, size_t r1, unsigned char* cmpBytes, size_t cmpSize)
{
@@ -29,10 +30,10 @@ int SZ_decompress_args_double(double** newData, size_t r5, size_t r4, size_t r3,
unsigned char* szTmpBytes;
if(cmpSize!=12+4+MetaDataByteLength && cmpSize!=12+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
if(confparams_dec->szMode!=SZ_TEMPORAL_COMPRESSION)
{
- if(isZlib)
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -46,7 +47,7 @@ int SZ_decompress_args_double(double** newData, size_t r5, size_t r4, size_t r3,
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
@@ -80,22 +81,45 @@ int SZ_decompress_args_double(double** newData, size_t r5, size_t r4, size_t r3,
(*newData)[i] = bytesToDouble(p);
}
}
- else if (dim == 1)
- getSnapshotData_double_1D(newData,r1,tdps, errBoundMode);
- else
- if (dim == 2)
- getSnapshotData_double_2D(newData,r2,r1,tdps, errBoundMode);
- else
- if (dim == 3)
- getSnapshotData_double_3D(newData,r3,r2,r1,tdps, errBoundMode);
- else
- if (dim == 4)
- getSnapshotData_double_4D(newData,r4,r3,r2,r1,tdps, errBoundMode);
- else
+ else
{
- printf("Error: currently support only at most 4 dimensions!\n");
- status = SZ_DERR;
- }
+ if(tdps->raBytes_size > 0) //v2.0
+ {
+ if (dim == 1)
+ getSnapshotData_double_1D(newData,r1,tdps, errBoundMode);
+ else if(dim == 2)
+ decompressDataSeries_double_2D_nonblocked_with_blocked_regression(newData, r2, r1, tdps->raBytes);
+ else if(dim == 3)
+ decompressDataSeries_double_3D_nonblocked_with_blocked_regression(newData, r3, r2, r1, tdps->raBytes);
+ else if(dim == 4)
+ decompressDataSeries_double_3D_nonblocked_with_blocked_regression(newData, r4*r3, r2, r1, tdps->raBytes);
+ else
+ {
+ printf("Error: currently support only at most 4 dimensions!\n");
+ status = SZ_DERR;
+ }
+ }
+ else //1.4.13
+ {
+ if (dim == 1)
+ getSnapshotData_double_1D(newData,r1,tdps, errBoundMode);
+ else
+ if (dim == 2)
+ getSnapshotData_double_2D(newData,r2,r1,tdps, errBoundMode);
+ else
+ if (dim == 3)
+ getSnapshotData_double_3D(newData,r3,r2,r1,tdps, errBoundMode);
+ else
+ if (dim == 4)
+ getSnapshotData_double_4D(newData,r4,r3,r2,r1,tdps, errBoundMode);
+ else
+ {
+ printf("Error: currently support only at most 4 dimensions!\n");
+ status = SZ_DERR;
+ }
+ }
+ }
+
free_TightDataPointStorageD2(tdps);
if(confparams_dec->szMode!=SZ_BEST_SPEED && cmpSize!=12+MetaDataByteLength+exe_params->SZ_SIZE_TYPE)
free(szTmpBytes);
@@ -1647,8 +1671,8 @@ void getSnapshotData_double_1D(double** data, size_t dataSeriesLength, TightData
}
else
{
- //decompressDataSeries_double_1D_pwr(data, dataSeriesLength, tdps);
- decompressDataSeries_double_1D_pwrgroup(data, dataSeriesLength, tdps);
+ decompressDataSeries_double_1D_pwr_pre_log(data, dataSeriesLength, tdps);
+ //decompressDataSeries_double_1D_pwrgroup(data, dataSeriesLength, tdps);
}
return;
} else {
@@ -1671,7 +1695,8 @@ void getSnapshotData_double_1D(double** data, size_t dataSeriesLength, TightData
if(errBoundMode < PW_REL)
decompressDataSeries_double_1D(&decmpData, dataSeriesLength, tdps);
else
- decompressDataSeries_double_1D_pwr(&decmpData, dataSeriesLength, tdps);
+ //decompressDataSeries_double_1D_pwr(&decmpData, dataSeriesLength, tdps);
+ decompressDataSeries_double_1D_pwr_pre_log(&decmpData, dataSeriesLength, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1711,7 +1736,8 @@ void getSnapshotData_double_2D(double** data, size_t r1, size_t r2, TightDataPoi
decompressDataSeries_double_2D(data, r1, r2, tdps);
}
else
- decompressDataSeries_double_2D_pwr(data, r1, r2, tdps);
+ //decompressDataSeries_double_2D_pwr(data, r1, r2, tdps);
+ decompressDataSeries_double_2D_pwr_pre_log(data, r1, r2, tdps);
return;
} else {
*data = (double*)malloc(sizeof(double)*dataSeriesLength);
@@ -1733,7 +1759,8 @@ void getSnapshotData_double_2D(double** data, size_t r1, size_t r2, TightDataPoi
if(errBoundMode < PW_REL)
decompressDataSeries_double_2D(&decmpData, r1, r2, tdps);
else
- decompressDataSeries_double_2D_pwr(&decmpData, r1, r2, tdps);
+ //decompressDataSeries_double_2D_pwr(&decmpData, r1, r2, tdps);
+ decompressDataSeries_double_2D_pwr_pre_log(&decmpData, r1, r2, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1773,7 +1800,8 @@ void getSnapshotData_double_3D(double** data, size_t r1, size_t r2, size_t r3, T
decompressDataSeries_double_3D(data, r1, r2, r3, tdps);
}
else
- decompressDataSeries_double_3D_pwr(data, r1, r2, r3, tdps);
+ //decompressDataSeries_double_3D_pwr(data, r1, r2, r3, tdps);
+ decompressDataSeries_double_3D_pwr_pre_log(data, r1, r2, r3, tdps);
return;
} else {
*data = (double*)malloc(sizeof(double)*dataSeriesLength);
@@ -1795,7 +1823,8 @@ void getSnapshotData_double_3D(double** data, size_t r1, size_t r2, size_t r3, T
if(errBoundMode < PW_REL)
decompressDataSeries_double_3D(&decmpData, r1, r2, r3, tdps);
else
- decompressDataSeries_double_3D_pwr(&decmpData, r1, r2, r3, tdps);
+ //decompressDataSeries_double_3D_pwr(&decmpData, r1, r2, r3, tdps);
+ decompressDataSeries_double_3D_pwr_pre_log(&decmpData, r1, r2, r3, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1836,7 +1865,8 @@ void getSnapshotData_double_4D(double** data, size_t r1, size_t r2, size_t r3, s
}
else
{
- decompressDataSeries_double_3D_pwr(data, r1*r2, r3, r4, tdps);
+ //decompressDataSeries_double_3D_pwr(data, r1*r2, r3, r4, tdps);
+ decompressDataSeries_double_3D_pwr_pre_log(data, r1*r2, r3, r4, tdps);
//ToDO
//decompressDataSeries_double_4D_pwr(data, r1, r2, r3, r4, tdps);
}
@@ -1858,7 +1888,8 @@ void getSnapshotData_double_4D(double** data, size_t r1, size_t r2, size_t r3, s
if(errBoundMode < PW_REL)
decompressDataSeries_double_4D(&decmpData, r1, r2, r3, r4, tdps);
else
- decompressDataSeries_double_3D_pwr(&decmpData, r1*r2, r3, r4, tdps);
+ //decompressDataSeries_double_3D_pwr(&decmpData, r1*r2, r3, r4, tdps);
+ decompressDataSeries_double_3D_pwr_pre_log(&decmpData, r1*r2, r3, r4, tdps);
//ToDo
//decompressDataSeries_double_4D_pwr(&decmpData, r1, r2, r3, r4, tdps);
// insert the decompressed data
@@ -1873,3 +1904,2719 @@ void getSnapshotData_double_4D(double** data, size_t r1, size_t r2, size_t r3, s
}
}
}
+
+void decompressDataSeries_double_2D_nonblocked_with_blocked_regression(double** data, size_t r1, size_t r2, unsigned char* comp_data){
+
+ size_t dim0_offset = r2;
+ size_t num_elements = r1 * r2;
+
+ *data = (double*)malloc(sizeof(double)*num_elements);
+
+ unsigned char * comp_data_pos = comp_data;
+
+ size_t block_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ // calculate block dims
+ size_t num_x, num_y;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+
+ size_t split_index_x, split_index_y;
+ size_t early_blockcount_x, early_blockcount_y;
+ size_t late_blockcount_x, late_blockcount_y;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+
+ size_t num_blocks = num_x * num_y;
+
+ double realPrecision = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ unsigned int intervals = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ updateQuantizationInfo(intervals);
+
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ int stateNum = 2*intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ double mean;
+ unsigned char use_mean;
+ memcpy(&use_mean, comp_data_pos, sizeof(unsigned char));
+ comp_data_pos += sizeof(unsigned char);
+ memcpy(&mean, comp_data_pos, sizeof(double));
+ comp_data_pos += sizeof(double);
+ size_t reg_count = 0;
+
+ unsigned char * indicator;
+ size_t indicator_bitlength = (num_blocks - 1)/8 + 1;
+ convertByteArray2IntArray_fast_1b(num_blocks, comp_data_pos, indicator_bitlength, &indicator);
+ comp_data_pos += indicator_bitlength;
+ for(size_t i=0; i<num_blocks; i++){
+ if(!indicator[i]) reg_count ++;
+ }
+ //printf("reg_count: %ld\n", reg_count);
+
+ int coeff_intvRadius[3];
+ int * coeff_result_type = (int *) malloc(num_blocks*3*sizeof(int));
+ int * coeff_type[3];
+ double precision[3];
+ double * coeff_unpred_data[3];
+ if(reg_count > 0){
+ for(int i=0; i<3; i++){
+ precision[i] = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ coeff_intvRadius[i] = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ int stateNum = 2*coeff_intvRadius[i]*2;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree, comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ size_t typeArray_size = bytesToSize(comp_data_pos);
+ decode(comp_data_pos + sizeof(size_t), reg_count, root, coeff_type[i]);
+ comp_data_pos += sizeof(size_t) + typeArray_size;
+ int coeff_unpred_count = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ coeff_unpred_data[i] = (double *) comp_data_pos;
+ comp_data_pos += coeff_unpred_count * sizeof(double);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ double last_coefficients[3] = {0.0};
+ int coeff_unpred_data_count[3] = {0};
+ int coeff_index = 0;
+ updateQuantizationInfo(intervals);
+
+ size_t total_unpred;
+ memcpy(&total_unpred, comp_data_pos, sizeof(size_t));
+ comp_data_pos += sizeof(size_t);
+ double * unpred_data = (double *) comp_data_pos;
+ comp_data_pos += total_unpred * sizeof(double);
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ decode(comp_data_pos, num_elements, root, result_type);
+ SZ_ReleaseHuffman(huffmanTree);
+
+ int intvRadius = exe_params->intvRadius;
+
+ int * type;
+
+ double * data_pos = *data;
+ size_t offset_x, offset_y;
+ size_t current_blockcount_x, current_blockcount_y;
+ size_t cur_unpred_count;
+
+ unsigned char * indicator_pos = indicator;
+ if(use_mean){
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y;
+ if(*indicator_pos){
+ // decompress by SZ
+
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ // d11 is current data
+ size_t unpredictable_count = 0;
+ double d00, d01, d10;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ d00 = d01 = d10 = 1;
+ if(i == 0 && ii == 0){
+ d00 = d01 = 0;
+ }
+ if(j == 0 && jj == 0){
+ d00 = d10 = 0;
+ }
+ if(d00){
+ d00 = block_data_pos[- dim0_offset - 1];
+ }
+ if(d01){
+ d01 = block_data_pos[- dim0_offset];
+ }
+ if(d10){
+ d10 = block_data_pos[- 1];
+ }
+ if(type_ < intvRadius) type_ += 1;
+ pred = d10 + d01 - d00;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<3; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+
+ type += current_block_elements;
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ else{
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y;
+ if(*indicator_pos){
+ // decompress by SZ
+
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ // d11 is current data
+ size_t unpredictable_count = 0;
+ double d00, d01, d10;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ d00 = d01 = d10 = 1;
+ if(i == 0 && ii == 0){
+ d00 = d01 = 0;
+ }
+ if(j == 0 && jj == 0){
+ d00 = d10 = 0;
+ }
+ if(d00){
+ d00 = block_data_pos[- dim0_offset - 1];
+ }
+ if(d01){
+ d01 = block_data_pos[- dim0_offset];
+ }
+ if(d10){
+ d10 = block_data_pos[- 1];
+ }
+ pred = d10 + d01 - d00;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<3; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+
+ type += current_block_elements;
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ free(coeff_result_type);
+
+ free(indicator);
+ free(result_type);
+}
+
+
+void decompressDataSeries_double_3D_nonblocked_with_blocked_regression(double** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data){
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+ size_t num_elements = r1 * r2 * r3;
+
+ *data = (double*)malloc(sizeof(double)*num_elements);
+
+ unsigned char * comp_data_pos = comp_data;
+
+ size_t block_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r3, num_z, block_size);
+
+ size_t split_index_x, split_index_y, split_index_z;
+ size_t early_blockcount_x, early_blockcount_y, early_blockcount_z;
+ size_t late_blockcount_x, late_blockcount_y, late_blockcount_z;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+ SZ_COMPUTE_BLOCKCOUNT(r3, num_z, split_index_z, early_blockcount_z, late_blockcount_z);
+
+ size_t num_blocks = num_x * num_y * num_z;
+
+ double realPrecision = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ unsigned int intervals = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ updateQuantizationInfo(intervals);
+
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ int stateNum = 2*intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,comp_data_pos+4, nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ double mean;
+ unsigned char use_mean;
+ memcpy(&use_mean, comp_data_pos, sizeof(unsigned char));
+ comp_data_pos += sizeof(unsigned char);
+ memcpy(&mean, comp_data_pos, sizeof(double));
+ comp_data_pos += sizeof(double);
+ size_t reg_count = 0;
+
+ unsigned char * indicator;
+ size_t indicator_bitlength = (num_blocks - 1)/8 + 1;
+ convertByteArray2IntArray_fast_1b(num_blocks, comp_data_pos, indicator_bitlength, &indicator);
+ comp_data_pos += indicator_bitlength;
+ for(size_t i=0; i<num_blocks; i++){
+ if(!indicator[i]) reg_count ++;
+ }
+
+ int coeff_intvRadius[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ int * coeff_type[4];
+ double precision[4];
+ double * coeff_unpred_data[4];
+ if(reg_count > 0){
+ for(int i=0; i<4; i++){
+ precision[i] = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ coeff_intvRadius[i] = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ int stateNum = 2*coeff_intvRadius[i]*2;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree, comp_data_pos+4, nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ size_t typeArray_size = bytesToSize(comp_data_pos);
+ decode(comp_data_pos + sizeof(size_t), reg_count, root, coeff_type[i]);
+ comp_data_pos += sizeof(size_t) + typeArray_size;
+ int coeff_unpred_count = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ coeff_unpred_data[i] = (double *) comp_data_pos;
+ comp_data_pos += coeff_unpred_count * sizeof(double);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ double last_coefficients[4] = {0.0};
+ int coeff_unpred_data_count[4] = {0};
+ int coeff_index = 0;
+ updateQuantizationInfo(intervals);
+
+ size_t total_unpred;
+ memcpy(&total_unpred, comp_data_pos, sizeof(size_t));
+ comp_data_pos += sizeof(size_t);
+ double * unpred_data = (double *) comp_data_pos;
+ comp_data_pos += total_unpred * sizeof(double);
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ decode(comp_data_pos, num_elements, root, result_type);
+ SZ_ReleaseHuffman(huffmanTree);
+
+ int intvRadius = exe_params->intvRadius;
+
+ int * type;
+ double * data_pos = *data;
+ size_t offset_x, offset_y, offset_z;
+ size_t current_blockcount_x, current_blockcount_y, current_blockcount_z;
+ size_t cur_unpred_count;
+ unsigned char * indicator_pos = indicator;
+ if(use_mean){
+ // type = result_type;
+
+ // for(size_t i=0; i<num_x; i++){
+ // for(size_t j=0; j<num_y; j++){
+ // for(size_t k=0; k<num_z; k++){
+ // offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ // offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ // offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ // data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ // current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ // current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ // current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ // // type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset + offset_z * current_blockcount_x * current_blockcount_y;
+ // // type = result_type + type_offset;
+ // size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ // // index = i * num_y * num_z + j * num_z + k;
+
+ // // printf("i j k: %ld %ld %ld\toffset: %ld %ld %ld\tindicator: %ld\n", i, j, k, offset_x, offset_y, offset_z, indicator[index]);
+ // if(*indicator_pos){
+ // // decompress by SZ
+ // // cur_unpred_count = decompressDataSeries_double_3D_blocked_nonblock_pred(data_pos, r1, r2, r3, current_blockcount_x, current_blockcount_y, current_blockcount_z, i, j, k, realPrecision, type, unpred_data);
+ // double * block_data_pos = data_pos;
+ // double pred;
+ // size_t index = 0;
+ // int type_;
+ // // d111 is current data
+ // size_t unpredictable_count = 0;
+ // double d000, d001, d010, d011, d100, d101, d110;
+ // for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // for(size_t jj=0; jj<current_blockcount_y; jj++){
+ // for(size_t kk=0; kk<current_blockcount_z; kk++){
+ // type_ = type[index];
+ // if(type_ == intvRadius){
+ // *block_data_pos = mean;
+ // }
+ // else if(type_ == 0){
+ // *block_data_pos = unpred_data[unpredictable_count ++];
+ // }
+ // else{
+ // d000 = d001 = d010 = d011 = d100 = d101 = d110 = 1;
+ // if(i == 0 && ii == 0){
+ // d000 = d001 = d010 = d011 = 0;
+ // }
+ // if(j == 0 && jj == 0){
+ // d000 = d001 = d100 = d101 = 0;
+ // }
+ // if(k == 0 && kk == 0){
+ // d000 = d010 = d100 = d110 = 0;
+ // }
+ // if(d000){
+ // d000 = block_data_pos[- dim0_offset - dim1_offset - 1];
+ // }
+ // if(d001){
+ // d001 = block_data_pos[- dim0_offset - dim1_offset];
+ // }
+ // if(d010){
+ // d010 = block_data_pos[- dim0_offset - 1];
+ // }
+ // if(d011){
+ // d011 = block_data_pos[- dim0_offset];
+ // }
+ // if(d100){
+ // d100 = block_data_pos[- dim1_offset - 1];
+ // }
+ // if(d101){
+ // d101 = block_data_pos[- dim1_offset];
+ // }
+ // if(d110){
+ // d110 = block_data_pos[- 1];
+ // }
+ // if(type_ < intvRadius) type_ += 1;
+ // pred = d110 + d101 + d011 - d100 - d010 - d001 + d000;
+ // *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ // }
+ // index ++;
+ // block_data_pos ++;
+ // }
+ // block_data_pos += dim1_offset - current_blockcount_z;
+ // }
+ // block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ // }
+ // cur_unpred_count = unpredictable_count;
+ // }
+ // else{
+ // // decompress by regression
+ // {
+ // //restore regression coefficients
+ // double pred;
+ // int type_;
+ // for(int e=0; e<4; e++){
+ // // if(i == 0 && j == 0 && k == 19){
+ // // printf("~\n");
+ // // }
+ // type_ = coeff_type[e][coeff_index];
+ // if (type_ != 0){
+ // pred = last_coefficients[e];
+ // last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ // }
+ // else{
+ // last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ // coeff_unpred_data_count[e] ++;
+ // }
+ // if(fabs(last_coefficients[e]) > 10000){
+ // printf("%d %d %d-%d: pred %.4f type %d precision %.4g last_coefficients %.4g\n", i, j, k, e, pred, type_, precision[e], last_coefficients[e]);
+ // exit(0);
+ // }
+ // }
+ // coeff_index ++;
+ // }
+ // {
+ // double * block_data_pos = data_pos;
+ // double pred;
+ // int type_;
+ // size_t index = 0;
+ // size_t unpredictable_count = 0;
+ // for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // for(size_t jj=0; jj<current_blockcount_y; jj++){
+ // for(size_t kk=0; kk<current_blockcount_z; kk++){
+ // if(block_data_pos - (*data) == 19470788){
+ // printf("dec stop\n");
+ // }
+
+ // type_ = type[index];
+ // if (type_ != 0){
+ // pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ // *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ // }
+ // else{
+ // *block_data_pos = unpred_data[unpredictable_count ++];
+ // }
+ // index ++;
+ // block_data_pos ++;
+ // }
+ // block_data_pos += dim1_offset - current_blockcount_z;
+ // }
+ // block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ // }
+ // cur_unpred_count = unpredictable_count;
+ // }
+ // }
+
+ // type += current_block_elements;
+ // indicator_pos ++;
+ // unpred_data += cur_unpred_count;
+ // // decomp_unpred += cur_unpred_count;
+ // // printf("block comp done, data_offset from %ld to %ld: diff %ld\n", *data, data_pos, data_pos - *data);
+ // // fflush(stdout);
+ // }
+ // }
+ // }
+
+ type = result_type;
+ // i == 0
+ {
+ // j == 0
+ {
+ // k == 0
+ {
+ data_pos = *data;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = 0;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ // i == 0 j == 0 k != 0
+ for(size_t k=1; k<num_z; k++){
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j==0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_y * dim1_offset;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ } // end i==0
+ for(size_t i=1; i<num_x; i++){
+ // j == 0
+ {
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = *data + offset_x * dim0_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j = 0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ }
+ else{
+ type = result_type;
+ // i == 0
+ {
+ // j == 0
+ {
+ // k == 0
+ {
+ data_pos = *data;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = 0;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ // i == 0 j == 0 k != 0
+ for(size_t k=1; k<num_z; k++){
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j==0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_y * dim1_offset;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ } // end i==0
+ for(size_t i=1; i<num_x; i++){
+ // j == 0
+ {
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = *data + offset_x * dim0_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j = 0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ double * block_data_pos = data_pos;
+ double pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ double pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ double * block_data_pos = data_pos;
+ double pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ }
+
+ free(coeff_result_type);
+
+ free(indicator);
+ free(result_type);
+}
diff --git a/thirdparty/SZ/sz/src/szd_double_pwr.c b/thirdparty/SZ/sz/src/szd_double_pwr.c
index f4a6fd8..a3ec18e 100644
--- a/thirdparty/SZ/sz/src/szd_double_pwr.c
+++ b/thirdparty/SZ/sz/src/szd_double_pwr.c
@@ -13,6 +13,7 @@
#include "TightDataPointStorageD.h"
#include "sz.h"
#include "Huffman.h"
+#include "utility.h"
//#include "rw.h"
#pragma GCC diagnostic push
@@ -1347,4 +1348,77 @@ void decompressDataSeries_double_1D_pwrgroup(double** data, size_t dataSeriesLen
free(groupErrorBounds);
free(groupID);
}
+
+void decompressDataSeries_double_1D_pwr_pre_log(double** data, size_t dataSeriesLength, TightDataPointStorageD* tdps) {
+
+ decompressDataSeries_double_1D(data, dataSeriesLength, tdps);
+ double threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+
+}
+
+void decompressDataSeries_double_2D_pwr_pre_log(double** data, size_t r1, size_t r2, TightDataPointStorageD* tdps) {
+
+ size_t dataSeriesLength = r1 * r2;
+ decompressDataSeries_double_2D(data, r1, r2, tdps);
+ double threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+}
+
+void decompressDataSeries_double_3D_pwr_pre_log(double** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageD* tdps) {
+
+ size_t dataSeriesLength = r1 * r2 * r3;
+ decompressDataSeries_double_3D(data, r1, r2, r3, tdps);
+ double threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+}
+
#pragma GCC diagnostic pop
diff --git a/thirdparty/SZ/sz/src/szd_float.c b/thirdparty/SZ/sz/src/szd_float.c
index 5a420af..0c7df9e 100644
--- a/thirdparty/SZ/sz/src/szd_float.c
+++ b/thirdparty/SZ/sz/src/szd_float.c
@@ -1,7 +1,7 @@
/**
* @file szd_float.c
- * @author Sheng Di and Dingwen Tao
- * @date Aug, 2016
+ * @author Sheng Di, Dingwen Tao, Xin Liang
+ * @date Aug, 2018
* @brief
* (C) 2016 by Mathematics and Computer Science (MCS), Argonne National Laboratory.
* See COPYRIGHT in top-level directory.
@@ -16,6 +16,7 @@
#include "Huffman.h"
#include "szd_float_pwr.h"
#include "szd_float_ts.h"
+#include "utility.h"
/**
*
@@ -35,10 +36,10 @@ int SZ_decompress_args_float(float** newData, size_t r5, size_t r4, size_t r3, s
if(cmpSize!=8+4+MetaDataByteLength && cmpSize!=8+8+MetaDataByteLength) //4,8 means two posibilities of SZ_SIZE_TYPE
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
if(confparams_dec->szMode!=SZ_TEMPORAL_COMPRESSION)
{
- if(isZlib)
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -53,7 +54,7 @@ int SZ_decompress_args_float(float** newData, size_t r5, size_t r4, size_t r3, s
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
@@ -88,21 +89,40 @@ int SZ_decompress_args_float(float** newData, size_t r5, size_t r4, size_t r3, s
(*newData)[i] = bytesToFloat(p);
}
}
- else if (dim == 1)
- getSnapshotData_float_1D(newData,r1,tdps, errBoundMode);
- else
- if (dim == 2)
- getSnapshotData_float_2D(newData,r2,r1,tdps, errBoundMode);
- else
- if (dim == 3)
- getSnapshotData_float_3D(newData,r3,r2,r1,tdps, errBoundMode);
- else
- if (dim == 4)
- getSnapshotData_float_4D(newData,r4,r3,r2,r1,tdps, errBoundMode);
- else
+ else
{
- printf("Error: currently support only at most 4 dimensions!\n");
- status = SZ_DERR;
+ if(tdps->raBytes_size > 0) //v2.0
+ {
+ if (dim == 1)
+ getSnapshotData_float_1D(newData,r1,tdps, errBoundMode);
+ else if(dim == 2)
+ decompressDataSeries_float_2D_nonblocked_with_blocked_regression(newData, r2, r1, tdps->raBytes);
+ else if(dim == 3)
+ decompressDataSeries_float_3D_nonblocked_with_blocked_regression(newData, r3, r2, r1, tdps->raBytes);
+ else if(dim == 4)
+ decompressDataSeries_float_3D_nonblocked_with_blocked_regression(newData, r4*r3, r2, r1, tdps->raBytes);
+ else
+ {
+ printf("Error: currently support only at most 4 dimensions!\n");
+ status = SZ_DERR;
+ }
+ }
+ else //1.4.13
+ {
+ if (dim == 1)
+ getSnapshotData_float_1D(newData,r1,tdps, errBoundMode);
+ else if (dim == 2)
+ getSnapshotData_float_2D(newData,r2,r1,tdps, errBoundMode);
+ else if (dim == 3)
+ getSnapshotData_float_3D(newData,r3,r2,r1,tdps, errBoundMode);
+ else if (dim == 4)
+ getSnapshotData_float_4D(newData,r4,r3,r2,r1,tdps, errBoundMode);
+ else
+ {
+ printf("Error: currently support only at most 4 dimensions!\n");
+ status = SZ_DERR;
+ }
+ }
}
free_TightDataPointStorageF2(tdps);
if(confparams_dec->szMode!=SZ_BEST_SPEED && cmpSize!=8+MetaDataByteLength+exe_params->SZ_SIZE_TYPE)
@@ -1651,8 +1671,8 @@ void getSnapshotData_float_1D(float** data, size_t dataSeriesLength, TightDataPo
}
else
{
- //decompressDataSeries_float_1D_pwr(data, dataSeriesLength, tdps);
- decompressDataSeries_float_1D_pwrgroup(data, dataSeriesLength, tdps);
+ decompressDataSeries_float_1D_pwr_pre_log(data, dataSeriesLength, tdps);
+ //decompressDataSeries_float_1D_pwrgroup(data, dataSeriesLength, tdps);
}
return;
} else {
@@ -1675,7 +1695,8 @@ void getSnapshotData_float_1D(float** data, size_t dataSeriesLength, TightDataPo
if(errBoundMode < PW_REL)
decompressDataSeries_float_1D(&decmpData, dataSeriesLength, tdps);
else
- decompressDataSeries_float_1D_pwr(&decmpData, dataSeriesLength, tdps);
+ //decompressDataSeries_float_1D_pwr(&decmpData, dataSeriesLength, tdps);
+ decompressDataSeries_float_1D_pwr_pre_log(&decmpData, dataSeriesLength, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1716,7 +1737,8 @@ void getSnapshotData_float_2D(float** data, size_t r1, size_t r2, TightDataPoint
}
else
{
- decompressDataSeries_float_2D_pwr(data, r1, r2, tdps);
+ //decompressDataSeries_float_2D_pwr(data, r1, r2, tdps);
+ decompressDataSeries_float_2D_pwr_pre_log(data, r1, r2, tdps);
}
return;
@@ -1740,7 +1762,8 @@ void getSnapshotData_float_2D(float** data, size_t r1, size_t r2, TightDataPoint
if(errBoundMode < PW_REL)
decompressDataSeries_float_2D(&decmpData, r1, r2, tdps);
else
- decompressDataSeries_float_2D_pwr(&decmpData, r1, r2, tdps);
+ //decompressDataSeries_float_2D_pwr(&decmpData, r1, r2, tdps);
+ decompressDataSeries_float_2D_pwr_pre_log(&decmpData, r1, r2, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1773,7 +1796,7 @@ void getSnapshotData_float_3D(float** data, size_t r1, size_t r2, size_t r3, Tig
if(multisteps->compressionType == 0)
decompressDataSeries_float_3D(data, r1, r2, r3, tdps);
else
- decompressDataSeries_float_1D_ts(data, r1*r2*r3, multisteps, tdps);
+ decompressDataSeries_float_1D_ts(data, dataSeriesLength, multisteps, tdps);
}
else
#endif
@@ -1781,7 +1804,8 @@ void getSnapshotData_float_3D(float** data, size_t r1, size_t r2, size_t r3, Tig
}
else
{
- decompressDataSeries_float_3D_pwr(data, r1, r2, r3, tdps);
+ //decompressDataSeries_float_3D_pwr(data, r1, r2, r3, tdps);
+ decompressDataSeries_float_3D_pwr_pre_log(data, r1, r2, r3, tdps);
}
return;
@@ -1805,7 +1829,8 @@ void getSnapshotData_float_3D(float** data, size_t r1, size_t r2, size_t r3, Tig
if(errBoundMode < PW_REL)
decompressDataSeries_float_3D(&decmpData, r1, r2, r3, tdps);
else
- decompressDataSeries_float_3D_pwr(&decmpData, r1, r2, r3, tdps);
+ //decompressDataSeries_float_3D_pwr(&decmpData, r1, r2, r3, tdps);
+ decompressDataSeries_float_3D_pwr_pre_log(&decmpData, r1, r2, r3, tdps);
// insert the decompressed data
size_t k = 0;
for (i = 0; i < dataSeriesLength; i++) {
@@ -1846,7 +1871,8 @@ void getSnapshotData_float_4D(float** data, size_t r1, size_t r2, size_t r3, siz
}
else
{
- decompressDataSeries_float_3D_pwr(data, r1*r2, r3, r4, tdps);
+ //decompressDataSeries_float_3D_pwr(data, r1*r2, r3, r4, tdps);
+ decompressDataSeries_float_3D_pwr_pre_log(data, r1*r2, r3, r4, tdps);
//ToDO
//decompressDataSeries_float_4D_pwr(data, r1, r2, r3, r4, tdps);
}
@@ -1868,7 +1894,8 @@ void getSnapshotData_float_4D(float** data, size_t r1, size_t r2, size_t r3, siz
if(errBoundMode < PW_REL)
decompressDataSeries_float_4D(&decmpData, r1, r2, r3, r4, tdps);
else
- decompressDataSeries_float_3D_pwr(&decmpData, r1*r2, r3, r4, tdps);
+ //decompressDataSeries_float_3D_pwr(&decmpData, r1*r2, r3, r4, tdps);
+ decompressDataSeries_float_3D_pwr_pre_log(&decmpData, r1*r2, r3, r4, tdps);
//ToDO
//decompressDataSeries_float_4D_pwr(&decompData, r1, r2, r3, r4, tdps);
// insert the decompressed data
@@ -2177,3 +2204,3059 @@ size_t decompressDataSeries_float_2D_RA_block(float * data, float mean, size_t d
return unpredictable_count;
}
+void decompressDataSeries_float_2D_nonblocked_with_blocked_regression(float** data, size_t r1, size_t r2, unsigned char* comp_data){
+
+ size_t dim0_offset = r2;
+ size_t num_elements = r1 * r2;
+
+ *data = (float*)malloc(sizeof(float)*num_elements);
+
+ unsigned char * comp_data_pos = comp_data;
+
+ size_t block_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ // calculate block dims
+ size_t num_x, num_y;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+
+ size_t split_index_x, split_index_y;
+ size_t early_blockcount_x, early_blockcount_y;
+ size_t late_blockcount_x, late_blockcount_y;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+
+ size_t num_blocks = num_x * num_y;
+
+ double realPrecision = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ unsigned int intervals = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ updateQuantizationInfo(intervals);
+
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ int stateNum = 2*intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ float mean;
+ unsigned char use_mean;
+ memcpy(&use_mean, comp_data_pos, sizeof(unsigned char));
+ comp_data_pos += sizeof(unsigned char);
+ memcpy(&mean, comp_data_pos, sizeof(float));
+ comp_data_pos += sizeof(float);
+ size_t reg_count = 0;
+
+ unsigned char * indicator;
+ size_t indicator_bitlength = (num_blocks - 1)/8 + 1;
+ convertByteArray2IntArray_fast_1b(num_blocks, comp_data_pos, indicator_bitlength, &indicator);
+ comp_data_pos += indicator_bitlength;
+ for(size_t i=0; i<num_blocks; i++){
+ if(!indicator[i]) reg_count ++;
+ }
+ //printf("reg_count: %ld\n", reg_count);
+
+ int coeff_intvRadius[3];
+ int * coeff_result_type = (int *) malloc(num_blocks*3*sizeof(int));
+ int * coeff_type[3];
+ double precision[3];
+ float * coeff_unpred_data[3];
+ if(reg_count > 0){
+ for(int i=0; i<3; i++){
+ precision[i] = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ coeff_intvRadius[i] = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ int stateNum = 2*coeff_intvRadius[i]*2;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree, comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ size_t typeArray_size = bytesToSize(comp_data_pos);
+ decode(comp_data_pos + sizeof(size_t), reg_count, root, coeff_type[i]);
+ comp_data_pos += sizeof(size_t) + typeArray_size;
+ int coeff_unpred_count = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ coeff_unpred_data[i] = (float *) comp_data_pos;
+ comp_data_pos += coeff_unpred_count * sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ float last_coefficients[3] = {0.0};
+ int coeff_unpred_data_count[3] = {0};
+ int coeff_index = 0;
+ updateQuantizationInfo(intervals);
+
+ size_t total_unpred;
+ memcpy(&total_unpred, comp_data_pos, sizeof(size_t));
+ comp_data_pos += sizeof(size_t);
+ float * unpred_data = (float *) comp_data_pos;
+ comp_data_pos += total_unpred * sizeof(float);
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ decode(comp_data_pos, num_elements, root, result_type);
+ SZ_ReleaseHuffman(huffmanTree);
+
+ int intvRadius = exe_params->intvRadius;
+
+ int * type;
+
+ float * data_pos = *data;
+ size_t offset_x, offset_y;
+ size_t current_blockcount_x, current_blockcount_y;
+ size_t cur_unpred_count;
+
+ unsigned char * indicator_pos = indicator;
+ if(use_mean){
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y;
+ if(*indicator_pos){
+ // decompress by SZ
+
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ // d11 is current data
+ size_t unpredictable_count = 0;
+ float d00, d01, d10;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ d00 = d01 = d10 = 1;
+ if(i == 0 && ii == 0){
+ d00 = d01 = 0;
+ }
+ if(j == 0 && jj == 0){
+ d00 = d10 = 0;
+ }
+ if(d00){
+ d00 = block_data_pos[- dim0_offset - 1];
+ }
+ if(d01){
+ d01 = block_data_pos[- dim0_offset];
+ }
+ if(d10){
+ d10 = block_data_pos[- 1];
+ }
+ if(type_ < intvRadius) type_ += 1;
+ pred = d10 + d01 - d00;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<3; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+
+ type += current_block_elements;
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ else{
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y;
+ if(*indicator_pos){
+ // decompress by SZ
+
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ // d11 is current data
+ size_t unpredictable_count = 0;
+ float d00, d01, d10;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ d00 = d01 = d10 = 1;
+ if(i == 0 && ii == 0){
+ d00 = d01 = 0;
+ }
+ if(j == 0 && jj == 0){
+ d00 = d10 = 0;
+ }
+ if(d00){
+ d00 = block_data_pos[- dim0_offset - 1];
+ }
+ if(d01){
+ d01 = block_data_pos[- dim0_offset];
+ }
+ if(d10){
+ d10 = block_data_pos[- 1];
+ }
+ pred = d10 + d01 - d00;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<3; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+
+ type += current_block_elements;
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ free(coeff_result_type);
+
+ free(indicator);
+ free(result_type);
+}
+
+
+void decompressDataSeries_float_3D_nonblocked_with_blocked_regression(float** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data){
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+ size_t num_elements = r1 * r2 * r3;
+
+ *data = (float*)malloc(sizeof(float)*num_elements);
+
+ unsigned char * comp_data_pos = comp_data;
+
+ size_t block_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r1, num_x, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r2, num_y, block_size);
+ SZ_COMPUTE_3D_NUMBER_OF_BLOCKS(r3, num_z, block_size);
+
+ size_t split_index_x, split_index_y, split_index_z;
+ size_t early_blockcount_x, early_blockcount_y, early_blockcount_z;
+ size_t late_blockcount_x, late_blockcount_y, late_blockcount_z;
+ SZ_COMPUTE_BLOCKCOUNT(r1, num_x, split_index_x, early_blockcount_x, late_blockcount_x);
+ SZ_COMPUTE_BLOCKCOUNT(r2, num_y, split_index_y, early_blockcount_y, late_blockcount_y);
+ SZ_COMPUTE_BLOCKCOUNT(r3, num_z, split_index_z, early_blockcount_z, late_blockcount_z);
+
+ size_t num_blocks = num_x * num_y * num_z;
+
+ double realPrecision = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ unsigned int intervals = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ updateQuantizationInfo(intervals);
+
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ int stateNum = 2*intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ float mean;
+ unsigned char use_mean;
+ memcpy(&use_mean, comp_data_pos, sizeof(unsigned char));
+ comp_data_pos += sizeof(unsigned char);
+ memcpy(&mean, comp_data_pos, sizeof(float));
+ comp_data_pos += sizeof(float);
+ size_t reg_count = 0;
+
+ unsigned char * indicator;
+ size_t indicator_bitlength = (num_blocks - 1)/8 + 1;
+ convertByteArray2IntArray_fast_1b(num_blocks, comp_data_pos, indicator_bitlength, &indicator);
+ comp_data_pos += indicator_bitlength;
+ for(size_t i=0; i<num_blocks; i++){
+ if(!indicator[i]) reg_count ++;
+ }
+
+ int coeff_intvRadius[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ int * coeff_type[4];
+ double precision[4];
+ float * coeff_unpred_data[4];
+ if(reg_count > 0){
+ for(int i=0; i<4; i++){
+ precision[i] = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ coeff_intvRadius[i] = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ int stateNum = 2*coeff_intvRadius[i]*2;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree, comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ size_t typeArray_size = bytesToSize(comp_data_pos);
+ decode(comp_data_pos + sizeof(size_t), reg_count, root, coeff_type[i]);
+ comp_data_pos += sizeof(size_t) + typeArray_size;
+ int coeff_unpred_count = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ coeff_unpred_data[i] = (float *) comp_data_pos;
+ comp_data_pos += coeff_unpred_count * sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ float last_coefficients[4] = {0.0};
+ int coeff_unpred_data_count[4] = {0};
+ int coeff_index = 0;
+ updateQuantizationInfo(intervals);
+
+ size_t total_unpred;
+ memcpy(&total_unpred, comp_data_pos, sizeof(size_t));
+ comp_data_pos += sizeof(size_t);
+ float * unpred_data = (float *) comp_data_pos;
+ comp_data_pos += total_unpred * sizeof(float);
+
+ int * result_type = (int *) malloc(num_elements * sizeof(int));
+ decode(comp_data_pos, num_elements, root, result_type);
+ SZ_ReleaseHuffman(huffmanTree);
+
+ int intvRadius = exe_params->intvRadius;
+
+ int * type;
+ float * data_pos = *data;
+ size_t offset_x, offset_y, offset_z;
+ size_t current_blockcount_x, current_blockcount_y, current_blockcount_z;
+ size_t cur_unpred_count;
+ unsigned char * indicator_pos = indicator;
+ if(use_mean){
+ // type = result_type;
+
+ // for(size_t i=0; i<num_x; i++){
+ // for(size_t j=0; j<num_y; j++){
+ // for(size_t k=0; k<num_z; k++){
+ // offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ // offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ // offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ // data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ // current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ // current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ // current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ // // type_offset = offset_x * dim0_offset + offset_y * current_blockcount_x * dim1_offset + offset_z * current_blockcount_x * current_blockcount_y;
+ // // type = result_type + type_offset;
+ // size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ // // index = i * num_y * num_z + j * num_z + k;
+
+ // // printf("i j k: %ld %ld %ld\toffset: %ld %ld %ld\tindicator: %ld\n", i, j, k, offset_x, offset_y, offset_z, indicator[index]);
+ // if(*indicator_pos){
+ // // decompress by SZ
+ // // cur_unpred_count = decompressDataSeries_float_3D_blocked_nonblock_pred(data_pos, r1, r2, r3, current_blockcount_x, current_blockcount_y, current_blockcount_z, i, j, k, realPrecision, type, unpred_data);
+ // float * block_data_pos = data_pos;
+ // float pred;
+ // size_t index = 0;
+ // int type_;
+ // // d111 is current data
+ // size_t unpredictable_count = 0;
+ // float d000, d001, d010, d011, d100, d101, d110;
+ // for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // for(size_t jj=0; jj<current_blockcount_y; jj++){
+ // for(size_t kk=0; kk<current_blockcount_z; kk++){
+ // type_ = type[index];
+ // if(type_ == intvRadius){
+ // *block_data_pos = mean;
+ // }
+ // else if(type_ == 0){
+ // *block_data_pos = unpred_data[unpredictable_count ++];
+ // }
+ // else{
+ // d000 = d001 = d010 = d011 = d100 = d101 = d110 = 1;
+ // if(i == 0 && ii == 0){
+ // d000 = d001 = d010 = d011 = 0;
+ // }
+ // if(j == 0 && jj == 0){
+ // d000 = d001 = d100 = d101 = 0;
+ // }
+ // if(k == 0 && kk == 0){
+ // d000 = d010 = d100 = d110 = 0;
+ // }
+ // if(d000){
+ // d000 = block_data_pos[- dim0_offset - dim1_offset - 1];
+ // }
+ // if(d001){
+ // d001 = block_data_pos[- dim0_offset - dim1_offset];
+ // }
+ // if(d010){
+ // d010 = block_data_pos[- dim0_offset - 1];
+ // }
+ // if(d011){
+ // d011 = block_data_pos[- dim0_offset];
+ // }
+ // if(d100){
+ // d100 = block_data_pos[- dim1_offset - 1];
+ // }
+ // if(d101){
+ // d101 = block_data_pos[- dim1_offset];
+ // }
+ // if(d110){
+ // d110 = block_data_pos[- 1];
+ // }
+ // if(type_ < intvRadius) type_ += 1;
+ // pred = d110 + d101 + d011 - d100 - d010 - d001 + d000;
+ // *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ // }
+ // index ++;
+ // block_data_pos ++;
+ // }
+ // block_data_pos += dim1_offset - current_blockcount_z;
+ // }
+ // block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ // }
+ // cur_unpred_count = unpredictable_count;
+ // }
+ // else{
+ // // decompress by regression
+ // {
+ // //restore regression coefficients
+ // float pred;
+ // int type_;
+ // for(int e=0; e<4; e++){
+ // // if(i == 0 && j == 0 && k == 19){
+ // // printf("~\n");
+ // // }
+ // type_ = coeff_type[e][coeff_index];
+ // if (type_ != 0){
+ // pred = last_coefficients[e];
+ // last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ // }
+ // else{
+ // last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ // coeff_unpred_data_count[e] ++;
+ // }
+ // if(fabs(last_coefficients[e]) > 10000){
+ // printf("%d %d %d-%d: pred %.4f type %d precision %.4g last_coefficients %.4g\n", i, j, k, e, pred, type_, precision[e], last_coefficients[e]);
+ // exit(0);
+ // }
+ // }
+ // coeff_index ++;
+ // }
+ // {
+ // float * block_data_pos = data_pos;
+ // float pred;
+ // int type_;
+ // size_t index = 0;
+ // size_t unpredictable_count = 0;
+ // for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // for(size_t jj=0; jj<current_blockcount_y; jj++){
+ // for(size_t kk=0; kk<current_blockcount_z; kk++){
+ // if(block_data_pos - (*data) == 19470788){
+ // printf("dec stop\n");
+ // }
+
+ // type_ = type[index];
+ // if (type_ != 0){
+ // pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ // *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ // }
+ // else{
+ // *block_data_pos = unpred_data[unpredictable_count ++];
+ // }
+ // index ++;
+ // block_data_pos ++;
+ // }
+ // block_data_pos += dim1_offset - current_blockcount_z;
+ // }
+ // block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ // }
+ // cur_unpred_count = unpredictable_count;
+ // }
+ // }
+
+ // type += current_block_elements;
+ // indicator_pos ++;
+ // unpred_data += cur_unpred_count;
+ // // decomp_unpred += cur_unpred_count;
+ // // printf("block comp done, data_offset from %ld to %ld: diff %ld\n", *data, data_pos, data_pos - *data);
+ // // fflush(stdout);
+ // }
+ // }
+ // }
+
+ type = result_type;
+ // i == 0
+ {
+ // j == 0
+ {
+ // k == 0
+ {
+ data_pos = *data;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = 0;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ // i == 0 j == 0 k != 0
+ for(size_t k=1; k<num_z; k++){
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j==0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_y * dim1_offset;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ } // end i==0
+ for(size_t i=1; i<num_x; i++){
+ // j == 0
+ {
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = *data + offset_x * dim0_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j = 0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == intvRadius){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ if(type_ < intvRadius) type_ += 1;
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ }
+ else{
+ type = result_type;
+ // i == 0
+ {
+ // j == 0
+ {
+ // k == 0
+ {
+ data_pos = *data;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = 0;
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ // i == 0 j == 0 k != 0
+ for(size_t k=1; k<num_z; k++){
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j==0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_y * dim1_offset;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = early_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ // ii == 0
+ {
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] - block_data_pos[- dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ for(size_t ii=1; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ } // end i==0
+ for(size_t i=1; i<num_x; i++){
+ // j == 0
+ {
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ data_pos = *data + offset_x * dim0_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim0_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = early_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ // jj == 0
+ {
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ for(size_t jj=1; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }// end j = 0
+ for(size_t j=1; j<num_y; j++){
+ // k == 0
+ {
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = early_blockcount_z;
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ {
+ // kk == 0
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim0_offset - dim1_offset];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ for(size_t kk=1; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ } // end k == 0
+ for(size_t k=1; k<num_z; k++){
+ offset_x = (i < split_index_x) ? i * early_blockcount_x : i * late_blockcount_x + split_index_x;
+ offset_y = (j < split_index_y) ? j * early_blockcount_y : j * late_blockcount_y + split_index_y;
+ offset_z = (k < split_index_z) ? k * early_blockcount_z : k * late_blockcount_z + split_index_z;
+ data_pos = *data + offset_x * dim0_offset + offset_y * dim1_offset + offset_z;
+
+ current_blockcount_x = (i < split_index_x) ? early_blockcount_x : late_blockcount_x;
+ current_blockcount_y = (j < split_index_y) ? early_blockcount_y : late_blockcount_y;
+ current_blockcount_z = (k < split_index_z) ? early_blockcount_z : late_blockcount_z;
+
+ size_t current_block_elements = current_blockcount_x * current_blockcount_y * current_blockcount_z;
+ if(*indicator_pos){
+ // decompress by SZ
+ float * block_data_pos = data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[- 1] + block_data_pos[- dim1_offset] + block_data_pos[- dim0_offset] - block_data_pos[- dim1_offset - 1] - block_data_pos[- dim0_offset - 1] - block_data_pos[- dim0_offset - dim1_offset] + block_data_pos[- dim0_offset - dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float * block_data_pos = data_pos;
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<current_blockcount_x; ii++){
+ for(size_t jj=0; jj<current_blockcount_y; jj++){
+ for(size_t kk=0; kk<current_blockcount_z; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ block_data_pos ++;
+ }
+ block_data_pos += dim1_offset - current_blockcount_z;
+ }
+ block_data_pos += dim0_offset - current_blockcount_y * dim1_offset;
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ type += current_block_elements;
+ unpred_data += cur_unpred_count;
+ }
+ }
+ }
+ }
+
+#ifdef HAVE_TIMECMPR
+ if(confparams_dec->szMode == SZ_TEMPORAL_COMPRESSION)
+ memcpy(multisteps->hist_data, (*data), num_elements*sizeof(float));
+#endif
+
+ free(coeff_result_type);
+
+ free(indicator);
+ free(result_type);
+}
+
+void decompressDataSeries_float_3D_random_access_with_blocked_regression(float** data, size_t r1, size_t r2, size_t r3, unsigned char* comp_data){
+
+ size_t dim0_offset = r2 * r3;
+ size_t dim1_offset = r3;
+ size_t num_elements = r1 * r2 * r3;
+
+ *data = (float*)malloc(sizeof(float)*num_elements);
+
+ unsigned char * comp_data_pos = comp_data;
+
+ size_t block_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ // calculate block dims
+ size_t num_x, num_y, num_z;
+ num_x = (r1 - 1) / block_size + 1;
+ num_y = (r2 - 1) / block_size + 1;
+ num_z = (r3 - 1) / block_size + 1;
+
+ size_t max_num_block_elements = block_size * block_size * block_size;
+ size_t num_blocks = num_x * num_y * num_z;
+
+ double realPrecision = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ unsigned int intervals = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ updateQuantizationInfo(intervals);
+
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+
+ int stateNum = 2*intervals;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree,comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ float mean;
+ unsigned char use_mean;
+ memcpy(&use_mean, comp_data_pos, sizeof(unsigned char));
+ comp_data_pos += sizeof(unsigned char);
+ memcpy(&mean, comp_data_pos, sizeof(float));
+ comp_data_pos += sizeof(float);
+ size_t reg_count = 0;
+
+ unsigned char * indicator;
+ size_t indicator_bitlength = (num_blocks - 1)/8 + 1;
+ convertByteArray2IntArray_fast_1b(num_blocks, comp_data_pos, indicator_bitlength, &indicator);
+ comp_data_pos += indicator_bitlength;
+ for(size_t i=0; i<num_blocks; i++){
+ if(!indicator[i]) reg_count ++;
+ }
+
+ int coeff_intvRadius[4];
+ int * coeff_result_type = (int *) malloc(num_blocks*4*sizeof(int));
+ int * coeff_type[4];
+ double precision[4];
+ float * coeff_unpred_data[4];
+ if(reg_count > 0){
+ for(int i=0; i<4; i++){
+ precision[i] = bytesToDouble(comp_data_pos);
+ comp_data_pos += sizeof(double);
+ coeff_intvRadius[i] = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ unsigned int tree_size = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ int stateNum = 2*coeff_intvRadius[i]*2;
+ HuffmanTree* huffmanTree = createHuffmanTree(stateNum);
+ int nodeCount = bytesToInt_bigEndian(comp_data_pos);
+ node root = reconstruct_HuffTree_from_bytes_anyStates(huffmanTree, comp_data_pos+sizeof(int), nodeCount);
+ comp_data_pos += sizeof(int) + tree_size;
+
+ coeff_type[i] = coeff_result_type + i * num_blocks;
+ size_t typeArray_size = bytesToSize(comp_data_pos);
+ decode(comp_data_pos + sizeof(size_t), reg_count, root, coeff_type[i]);
+ comp_data_pos += sizeof(size_t) + typeArray_size;
+ int coeff_unpred_count = bytesToInt_bigEndian(comp_data_pos);
+ comp_data_pos += sizeof(int);
+ coeff_unpred_data[i] = (float *) comp_data_pos;
+ comp_data_pos += coeff_unpred_count * sizeof(float);
+ SZ_ReleaseHuffman(huffmanTree);
+ }
+ }
+ float last_coefficients[4] = {0.0};
+ int coeff_unpred_data_count[4] = {0};
+ int coeff_index = 0;
+ updateQuantizationInfo(intervals);
+
+ size_t total_unpred;
+ memcpy(&total_unpred, comp_data_pos, sizeof(size_t));
+ comp_data_pos += sizeof(size_t);
+ float * unpred_data = (float *) comp_data_pos;
+ comp_data_pos += total_unpred * sizeof(float);
+
+ int * result_type = (int *) malloc(num_blocks*max_num_block_elements * sizeof(int));
+ decode(comp_data_pos, num_blocks*max_num_block_elements, root, result_type);
+ SZ_ReleaseHuffman(huffmanTree);
+
+ int intvRadius = exe_params->intvRadius;
+
+ int * type;
+ float * data_pos = *data;
+ size_t cur_unpred_count;
+ unsigned char * indicator_pos = indicator;
+ int dec_buffer_size = block_size + 1;
+ float * dec_buffer = (float *) malloc(dec_buffer_size*dec_buffer_size*dec_buffer_size*sizeof(float));
+ memset(dec_buffer, 0, dec_buffer_size*dec_buffer_size*dec_buffer_size*sizeof(float));
+ float * block_data_pos_x = NULL;
+ float * block_data_pos_y = NULL;
+ float * block_data_pos_z = NULL;
+ int block_dim0_offset = dec_buffer_size*dec_buffer_size;
+ int block_dim1_offset = dec_buffer_size;
+ if(use_mean){
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ data_pos = dec_buffer + dec_buffer_size*dec_buffer_size + dec_buffer_size + 1;
+ if(*indicator_pos){
+ // decompress by SZ
+ // cur_unpred_count = decompressDataSeries_float_3D_blocked_nonblock_pred(data_pos, r1, r2, r3, current_blockcount_x, current_blockcount_y, current_blockcount_z, i, j, k, realPrecision, type, unpred_data);
+ float * block_data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ block_data_pos = data_pos + ii*block_dim0_offset + jj*block_dim1_offset + kk;
+ type_ = type[index];
+ if(type_ == 1){
+ *block_data_pos = mean;
+ }
+ else if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[-1] + block_data_pos[-block_dim1_offset]+ block_data_pos[-block_dim0_offset] - block_data_pos[-block_dim1_offset - 1]
+ - block_data_pos[-block_dim0_offset - 1] - block_data_pos[-block_dim0_offset - block_dim1_offset] + block_data_pos[-block_dim0_offset - block_dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ }
+ }
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ // if(i == 0 && j == 0 && k == 19){
+ // printf("~\n");
+ // }
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ data_pos[ii*block_dim0_offset + jj*block_dim1_offset + kk] = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ data_pos[ii*block_dim0_offset + jj*block_dim1_offset + kk] = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ }
+ }
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ // decomp_unpred += cur_unpred_count;
+ // printf("block comp done, data_offset from %ld to %ld: diff %ld\n", *data, data_pos, data_pos - *data);
+ // fflush(stdout);
+ type += block_size * block_size * block_size;
+
+ // mv data back
+ block_data_pos_x = *data + i*block_size * dim0_offset + j*block_size * dim1_offset + k*block_size;
+ for(int ii=0; ii<block_size; ii++){
+ if(i*block_size + ii >= r1) break;
+ block_data_pos_y = block_data_pos_x;
+ for(int jj=0; jj<block_size; jj++){
+ if(j*block_size + jj >= r2) break;
+ block_data_pos_z = block_data_pos_y;
+ for(int kk=0; kk<block_size; kk++){
+ if(k*block_size + kk >= r3) break;
+ *block_data_pos_z = data_pos[ii*dec_buffer_size*dec_buffer_size + jj*dec_buffer_size + kk];
+ block_data_pos_z ++;
+ }
+ block_data_pos_y += dim1_offset;
+ }
+ block_data_pos_x += dim0_offset;
+ }
+
+ }
+ }
+ }
+
+ }
+ else{
+ type = result_type;
+ for(size_t i=0; i<num_x; i++){
+ for(size_t j=0; j<num_y; j++){
+ for(size_t k=0; k<num_z; k++){
+ data_pos = dec_buffer + dec_buffer_size*dec_buffer_size + dec_buffer_size + 1;
+ if(*indicator_pos){
+ // decompress by SZ
+ // cur_unpred_count = decompressDataSeries_float_3D_blocked_nonblock_pred(data_pos, r1, r2, r3, current_blockcount_x, current_blockcount_y, current_blockcount_z, i, j, k, realPrecision, type, unpred_data);
+ float * block_data_pos;
+ float pred;
+ size_t index = 0;
+ int type_;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ block_data_pos = data_pos + ii*block_dim0_offset + jj*block_dim1_offset + kk;
+ type_ = type[index];
+ if(type_ == 0){
+ *block_data_pos = unpred_data[unpredictable_count ++];
+ }
+ else{
+ pred = block_data_pos[-1] + block_data_pos[-block_dim1_offset]+ block_data_pos[-block_dim0_offset] - block_data_pos[-block_dim1_offset - 1]
+ - block_data_pos[-block_dim0_offset - 1] - block_data_pos[-block_dim0_offset - block_dim1_offset] + block_data_pos[-block_dim0_offset - block_dim1_offset - 1];
+ *block_data_pos = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ index ++;
+ }
+ }
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ else{
+ // decompress by regression
+ {
+ //restore regression coefficients
+ float pred;
+ int type_;
+ for(int e=0; e<4; e++){
+ // if(i == 0 && j == 0 && k == 19){
+ // printf("~\n");
+ // }
+ type_ = coeff_type[e][coeff_index];
+ if (type_ != 0){
+ pred = last_coefficients[e];
+ last_coefficients[e] = pred + 2 * (type_ - coeff_intvRadius[e]) * precision[e];
+ }
+ else{
+ last_coefficients[e] = coeff_unpred_data[e][coeff_unpred_data_count[e]];
+ coeff_unpred_data_count[e] ++;
+ }
+ }
+ coeff_index ++;
+ }
+ {
+ float pred;
+ int type_;
+ size_t index = 0;
+ size_t unpredictable_count = 0;
+ for(size_t ii=0; ii<block_size; ii++){
+ for(size_t jj=0; jj<block_size; jj++){
+ for(size_t kk=0; kk<block_size; kk++){
+ type_ = type[index];
+ if (type_ != 0){
+ pred = last_coefficients[0] * ii + last_coefficients[1] * jj + last_coefficients[2] * kk + last_coefficients[3];
+ data_pos[ii*block_dim0_offset + jj*block_dim1_offset + kk] = pred + 2 * (type_ - intvRadius) * realPrecision;
+ }
+ else{
+ data_pos[ii*block_dim0_offset + jj*block_dim1_offset + kk] = unpred_data[unpredictable_count ++];
+ }
+ index ++;
+ }
+ }
+ }
+ cur_unpred_count = unpredictable_count;
+ }
+ }
+ indicator_pos ++;
+ unpred_data += cur_unpred_count;
+ // decomp_unpred += cur_unpred_count;
+ // printf("block comp done, data_offset from %ld to %ld: diff %ld\n", *data, data_pos, data_pos - *data);
+ // fflush(stdout);
+ type += block_size * block_size * block_size;
+ // mv data back
+ block_data_pos_x = *data + i*block_size * dim0_offset + j*block_size * dim1_offset + k*block_size;
+ for(int ii=0; ii<block_size; ii++){
+ if(i*block_size + ii >= r1) break;
+ block_data_pos_y = block_data_pos_x;
+ for(int jj=0; jj<block_size; jj++){
+ if(j*block_size + jj >= r2) break;
+ block_data_pos_z = block_data_pos_y;
+ for(int kk=0; kk<block_size; kk++){
+ if(k*block_size + kk >= r3) break;
+ *block_data_pos_z = data_pos[ii*dec_buffer_size*dec_buffer_size + jj*dec_buffer_size + kk];
+ block_data_pos_z ++;
+ }
+ block_data_pos_y += dim1_offset;
+ }
+ block_data_pos_x += dim0_offset;
+ }
+ }
+ }
+ }
+ }
+ free(dec_buffer);
+ free(coeff_result_type);
+
+ free(indicator);
+ free(result_type);
+}
diff --git a/thirdparty/SZ/sz/src/szd_float_pwr.c b/thirdparty/SZ/sz/src/szd_float_pwr.c
index 4ab1834..b761d4b 100644
--- a/thirdparty/SZ/sz/src/szd_float_pwr.c
+++ b/thirdparty/SZ/sz/src/szd_float_pwr.c
@@ -16,6 +16,7 @@
#include "sz.h"
#include "Huffman.h"
#include "sz_float_pwr.h"
+#include "utility.h"
//#include "rw.h"
//
#pragma GCC diagnostic push
@@ -1349,4 +1350,74 @@ void decompressDataSeries_float_1D_pwrgroup(float** data, size_t dataSeriesLengt
free(groupErrorBounds);
free(groupID);
}
+
+void decompressDataSeries_float_1D_pwr_pre_log(float** data, size_t dataSeriesLength, TightDataPointStorageF* tdps) {
+
+ decompressDataSeries_float_1D(data, dataSeriesLength, tdps);
+ float threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+
+}
+
+void decompressDataSeries_float_2D_pwr_pre_log(float** data, size_t r1, size_t r2, TightDataPointStorageF* tdps) {
+
+ size_t dataSeriesLength = r1 * r2;
+ decompressDataSeries_float_2D(data, r1, r2, tdps);
+ float threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+
+}
+
+void decompressDataSeries_float_3D_pwr_pre_log(float** data, size_t r1, size_t r2, size_t r3, TightDataPointStorageF* tdps) {
+
+ size_t dataSeriesLength = r1 * r2 * r3;
+ decompressDataSeries_float_3D(data, r1, r2, r3, tdps);
+ float threshold = tdps->minLogValue;
+ if(tdps->pwrErrBoundBytes_size > 0){
+ unsigned char * signs;
+ sz_lossless_decompress(confparams_dec->losslessCompressor, tdps->pwrErrBoundBytes, tdps->pwrErrBoundBytes_size, &signs, dataSeriesLength);
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ if(signs[i]) (*data)[i] = -((*data)[i]);
+ }
+ free(signs);
+ }
+ else{
+ for(size_t i=0; i<dataSeriesLength; i++){
+ if((*data)[i] < threshold) (*data)[i] = 0;
+ else (*data)[i] = exp2((*data)[i]);
+ }
+ }
+}
#pragma GCC diagnostic pop
diff --git a/thirdparty/SZ/sz/src/szd_int16.c b/thirdparty/SZ/sz/src/szd_int16.c
index 3c402dc..1198e05 100644
--- a/thirdparty/SZ/sz/src/szd_int16.c
+++ b/thirdparty/SZ/sz/src/szd_int16.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_int16.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -32,10 +33,10 @@ int SZ_decompress_args_int16(int16_t** newData, size_t r5, size_t r4, size_t r3,
size_t i, tmpSize = 3+MetaDataByteLength+1+sizeof(int16_t)+exe_params->SZ_SIZE_TYPE;
unsigned char* szTmpBytes;
- if(cmpSize!=4+2+4+MetaDataByteLength && cmpSize!=4+2+8+MetaDataByteLength)
+ if(cmpSize!=4+2+4+MetaDataByteLength && cmpSize!=4+2+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_int16(int16_t** newData, size_t r5, size_t r4, size_t r3,
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_int32.c b/thirdparty/SZ/sz/src/szd_int32.c
index 43dc74e..b5f31b0 100644
--- a/thirdparty/SZ/sz/src/szd_int32.c
+++ b/thirdparty/SZ/sz/src/szd_int32.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_int32.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_int32(int32_t** newData, size_t r5, size_t r4, size_t r3,
if(cmpSize!=4+4+4+MetaDataByteLength && cmpSize!=4+4+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_int32(int32_t** newData, size_t r5, size_t r4, size_t r3,
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_int64.c b/thirdparty/SZ/sz/src/szd_int64.c
index aaa4a53..07a054f 100644
--- a/thirdparty/SZ/sz/src/szd_int64.c
+++ b/thirdparty/SZ/sz/src/szd_int64.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_int64.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_int64(int64_t** newData, size_t r5, size_t r4, size_t r3,
if(cmpSize!=4+8+4+MetaDataByteLength && cmpSize!=4+8+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_int64(int64_t** newData, size_t r5, size_t r4, size_t r3,
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_int8.c b/thirdparty/SZ/sz/src/szd_int8.c
index 758e917..850b459 100644
--- a/thirdparty/SZ/sz/src/szd_int8.c
+++ b/thirdparty/SZ/sz/src/szd_int8.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_int8.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_int8(int8_t** newData, size_t r5, size_t r4, size_t r3, s
if(cmpSize!=4+1+4+MetaDataByteLength && cmpSize!=4+1+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_int8(int8_t** newData, size_t r5, size_t r4, size_t r3, s
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_uint16.c b/thirdparty/SZ/sz/src/szd_uint16.c
index bdc7469..551eecf 100644
--- a/thirdparty/SZ/sz/src/szd_uint16.c
+++ b/thirdparty/SZ/sz/src/szd_uint16.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_uint16.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_uint16(uint16_t** newData, size_t r5, size_t r4, size_t r
if(cmpSize!=4+2+4+MetaDataByteLength && cmpSize!=4+2+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_uint16(uint16_t** newData, size_t r5, size_t r4, size_t r
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_uint32.c b/thirdparty/SZ/sz/src/szd_uint32.c
index 795eabe..04e8049 100644
--- a/thirdparty/SZ/sz/src/szd_uint32.c
+++ b/thirdparty/SZ/sz/src/szd_uint32.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_uint32.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_uint32(uint32_t** newData, size_t r5, size_t r4, size_t r
if(cmpSize!=4+4+4+MetaDataByteLength && cmpSize!=4+4+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_uint32(uint32_t** newData, size_t r5, size_t r4, size_t r
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_uint64.c b/thirdparty/SZ/sz/src/szd_uint64.c
index df2b838..84d5716 100644
--- a/thirdparty/SZ/sz/src/szd_uint64.c
+++ b/thirdparty/SZ/sz/src/szd_uint64.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_uint64.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_uint64(uint64_t** newData, size_t r5, size_t r4, size_t r
if(cmpSize!=4+8+4+MetaDataByteLength && cmpSize!=4+8+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_uint64(uint64_t** newData, size_t r5, size_t r4, size_t r
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szd_uint8.c b/thirdparty/SZ/sz/src/szd_uint8.c
index cd61635..8b992bc 100644
--- a/thirdparty/SZ/sz/src/szd_uint8.c
+++ b/thirdparty/SZ/sz/src/szd_uint8.c
@@ -15,6 +15,7 @@
#include "sz.h"
#include "szd_uint8.h"
#include "Huffman.h"
+#include "utility.h"
/**
*
@@ -34,8 +35,8 @@ int SZ_decompress_args_uint8(uint8_t** newData, size_t r5, size_t r4, size_t r3,
if(cmpSize!=4+1+4+MetaDataByteLength && cmpSize!=4+1+8+MetaDataByteLength)
{
- int isZlib = isZlibFormat(cmpBytes[0], cmpBytes[1]);
- if(isZlib)
+ confparams_dec->losslessCompressor = is_lossless_compressed_data(cmpBytes, cmpSize);
+ if(confparams_dec->losslessCompressor!=-1)
confparams_dec->szMode = SZ_BEST_COMPRESSION;
else
confparams_dec->szMode = SZ_BEST_SPEED;
@@ -48,7 +49,7 @@ int SZ_decompress_args_uint8(uint8_t** newData, size_t r5, size_t r4, size_t r3,
{
if(targetUncompressSize<MIN_ZLIB_DEC_ALLOMEM_BYTES) //Considering the minimum size
targetUncompressSize = MIN_ZLIB_DEC_ALLOMEM_BYTES;
- tmpSize = zlib_uncompress5(cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
+ tmpSize = sz_lossless_decompress(confparams_dec->losslessCompressor, cmpBytes, (unsigned long)cmpSize, &szTmpBytes, (unsigned long)targetUncompressSize+4+MetaDataByteLength+exe_params->SZ_SIZE_TYPE);// (unsigned long)targetUncompressSize+8: consider the total length under lossless compression mode is actually 3+4+1+targetUncompressSize
//szTmpBytes = (unsigned char*)malloc(sizeof(unsigned char)*tmpSize);
//memcpy(szTmpBytes, tmpBytes, tmpSize);
//free(tmpBytes); //release useless memory
diff --git a/thirdparty/SZ/sz/src/szf.c b/thirdparty/SZ/sz/src/szf.c
index e3cca0b..43fe0b1 100644
--- a/thirdparty/SZ/sz/src/szf.c
+++ b/thirdparty/SZ/sz/src/szf.c
@@ -176,70 +176,70 @@ void sz_compress_d5_double_rev_(double* data, double *reservedValue, unsigned ch
void sz_compress_d1_float_args_(float* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, 0, 0, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, 0, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d2_float_args_(float* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, 0, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d3_float_args_(float* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d4_float_args_(float* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, *r4, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, *r4, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d5_float_args_(float* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4, size_t *r5)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, *r5, *r4, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_FLOAT, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, *r5, *r4, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d1_double_args_(double* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, 0, 0, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, 0, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d2_double_args_(double* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, 0, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d3_double_args_(double* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, 0, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d4_double_args_(double* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, 0, *r4, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, *r4, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
void sz_compress_d5_double_args_(double* data, unsigned char *bytes, size_t *outSize, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4, size_t *r5)
{
- unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 1, *r5, *r4, *r3, *r2, *r1);
+ unsigned char *tmp_bytes = SZ_compress_args(SZ_DOUBLE, data, outSize, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, *r5, *r4, *r3, *r2, *r1);
memcpy(bytes, tmp_bytes, *outSize);
free(tmp_bytes);
}
@@ -411,7 +411,7 @@ void sz_batchaddvar_d1_float_(char* varName, int *len, float* data, int *errBoun
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, 0, 0, *r1);
+ SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, 0, *r1);
}
void sz_batchaddvar_d2_float_(char* varName, int *len, float* data, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2)
{
@@ -420,7 +420,7 @@ void sz_batchaddvar_d2_float_(char* varName, int *len, float* data, int *errBoun
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, 0, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, *r2, *r1);
}
void sz_batchaddvar_d3_float_(char* varName, int *len, float* data, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3)
{
@@ -429,7 +429,7 @@ void sz_batchaddvar_d3_float_(char* varName, int *len, float* data, int *errBoun
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, *r3, *r2, *r1);
}
void sz_batchaddvar_d4_float_(char* varName, int *len, float* data, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4)
{
@@ -438,7 +438,7 @@ void sz_batchaddvar_d4_float_(char* varName, int *len, float* data, int *errBoun
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, *r4, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, *r4, *r3, *r2, *r1);
}
void sz_batchaddvar_d5_float_(char* varName, int *len, float* data, int *errBoundMode, float *absErrBound, float *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4, size_t *r5)
{
@@ -447,7 +447,7 @@ void sz_batchaddvar_d5_float_(char* varName, int *len, float* data, int *errBoun
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, *r5, *r4, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_FLOAT, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, *r5, *r4, *r3, *r2, *r1);
}
void sz_batchaddvar_d1_double_(char* varName, int *len, double* data, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1)
{
@@ -456,7 +456,7 @@ void sz_batchaddvar_d1_double_(char* varName, int *len, double* data, int *errBo
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, 0, 0, *r1);
+ SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, 0, *r1);
}
void sz_batchaddvar_d2_double_(char* varName, int *len, double* data, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2)
{
@@ -465,7 +465,7 @@ void sz_batchaddvar_d2_double_(char* varName, int *len, double* data, int *errBo
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, 0, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, 0, *r2, *r1);
}
void sz_batchaddvar_d3_double_(char* varName, int *len, double* data, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3)
{
@@ -474,7 +474,7 @@ void sz_batchaddvar_d3_double_(char* varName, int *len, double* data, int *errBo
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, 0, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, 0, *r3, *r2, *r1);
}
void sz_batchaddvar_d4_double_(char* varName, int *len, double* data, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4)
{
@@ -483,7 +483,7 @@ void sz_batchaddvar_d4_double_(char* varName, int *len, double* data, int *errBo
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0, *r4, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, 0, *r4, *r3, *r2, *r1);
}
void sz_batchaddvar_d5_double_(char* varName, int *len, double* data, int *errBoundMode, double *absErrBound, double *relBoundRatio, size_t *r1, size_t *r2, size_t *r3, size_t *r4, size_t *r5)
{
@@ -492,7 +492,7 @@ void sz_batchaddvar_d5_double_(char* varName, int *len, double* data, int *errBo
for(i=0;i<*len;i++)
s2[i]=varName[i];
s2[*len]='\0';
- SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, *r5, *r4, *r3, *r2, *r1);
+ SZ_batchAddVar(s2, SZ_DOUBLE, data, *errBoundMode, *absErrBound, *relBoundRatio, 0.1, *r5, *r4, *r3, *r2, *r1);
}
void sz_batchdelvar_c_(char* varName, int *len, int *errState)
{
@@ -503,15 +503,18 @@ void sz_batchdelvar_c_(char* varName, int *len, int *errState)
s2[*len]='\0';
*errState = SZ_batchDelVar(s2);
}
+
+/*@deprecated*/
void sz_batch_compress_c_(unsigned char* bytes, size_t *outSize)
{
- unsigned char* tmp_bytes = SZ_batch_compress(outSize);
- memcpy(bytes, tmp_bytes, *outSize);
- free(tmp_bytes);
+ //unsigned char* tmp_bytes = SZ_batch_compress(outSize);
+ //memcpy(bytes, tmp_bytes, *outSize);
+ //free(tmp_bytes);
}
+/*@deprecated*/
void sz_batch_decompress_c_(unsigned char* bytes, size_t *byteLength, int *ierr)
{
- SZ_batch_decompress(bytes, *byteLength, ierr);
+ //SZ_batch_decompress(bytes, *byteLength, ierr);
}
void sz_getvardim_c_(char* varName, int *len, int *dim, size_t *r1, size_t *r2, size_t *r3, size_t *r4, size_t *r5)
--
GitLab