[legacy] Fix Huffman jump table reads in v01 and v05
Nick Terrell
parent
785331acbd
commit
2536771134
|
|
@ -1073,99 +1073,102 @@ static size_t HUF_decompress_usingDTable( /* -3% slower when non static */
|
|||
const void* cSrc, size_t cSrcSize,
|
||||
const U16* DTable)
|
||||
{
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* op = ostart;
|
||||
BYTE* const omax = op + maxDstSize;
|
||||
BYTE* const olimit = omax-15;
|
||||
|
||||
const void* ptr = DTable;
|
||||
const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
|
||||
const U32 dtLog = DTable[0];
|
||||
size_t errorCode;
|
||||
U32 reloadStatus;
|
||||
|
||||
/* Init */
|
||||
|
||||
const U16* jumpTable = (const U16*)cSrc;
|
||||
const size_t length1 = FSE_readLE16(jumpTable);
|
||||
const size_t length2 = FSE_readLE16(jumpTable+1);
|
||||
const size_t length3 = FSE_readLE16(jumpTable+2);
|
||||
const size_t length4 = cSrcSize - 6 - length1 - length2 - length3; // check coherency !!
|
||||
const char* const start1 = (const char*)(cSrc) + 6;
|
||||
const char* const start2 = start1 + length1;
|
||||
const char* const start3 = start2 + length2;
|
||||
const char* const start4 = start3 + length3;
|
||||
FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
|
||||
|
||||
if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
|
||||
|
||||
errorCode = FSE_initDStream(&bitD1, start1, length1);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD2, start2, length2);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD3, start3, length3);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD4, start4, length4);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
|
||||
reloadStatus=FSE_reloadDStream(&bitD2);
|
||||
|
||||
/* 16 symbols per loop */
|
||||
for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit); /* D2-3-4 are supposed to be synchronized and finish together */
|
||||
op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
|
||||
if (cSrcSize < 6) return (size_t)-FSE_ERROR_srcSize_wrong;
|
||||
{
|
||||
#define HUF_DECODE_SYMBOL_0(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* op = ostart;
|
||||
BYTE* const omax = op + maxDstSize;
|
||||
BYTE* const olimit = omax-15;
|
||||
|
||||
#define HUF_DECODE_SYMBOL_1(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
|
||||
if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
|
||||
const void* ptr = DTable;
|
||||
const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
|
||||
const U32 dtLog = DTable[0];
|
||||
size_t errorCode;
|
||||
U32 reloadStatus;
|
||||
|
||||
#define HUF_DECODE_SYMBOL_2(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
|
||||
if (FSE_32bits()) FSE_reloadDStream(&Dstream)
|
||||
/* Init */
|
||||
|
||||
HUF_DECODE_SYMBOL_1( 0, bitD1);
|
||||
HUF_DECODE_SYMBOL_1( 1, bitD2);
|
||||
HUF_DECODE_SYMBOL_1( 2, bitD3);
|
||||
HUF_DECODE_SYMBOL_1( 3, bitD4);
|
||||
HUF_DECODE_SYMBOL_2( 4, bitD1);
|
||||
HUF_DECODE_SYMBOL_2( 5, bitD2);
|
||||
HUF_DECODE_SYMBOL_2( 6, bitD3);
|
||||
HUF_DECODE_SYMBOL_2( 7, bitD4);
|
||||
HUF_DECODE_SYMBOL_1( 8, bitD1);
|
||||
HUF_DECODE_SYMBOL_1( 9, bitD2);
|
||||
HUF_DECODE_SYMBOL_1(10, bitD3);
|
||||
HUF_DECODE_SYMBOL_1(11, bitD4);
|
||||
HUF_DECODE_SYMBOL_0(12, bitD1);
|
||||
HUF_DECODE_SYMBOL_0(13, bitD2);
|
||||
HUF_DECODE_SYMBOL_0(14, bitD3);
|
||||
HUF_DECODE_SYMBOL_0(15, bitD4);
|
||||
}
|
||||
const U16* jumpTable = (const U16*)cSrc;
|
||||
const size_t length1 = FSE_readLE16(jumpTable);
|
||||
const size_t length2 = FSE_readLE16(jumpTable+1);
|
||||
const size_t length3 = FSE_readLE16(jumpTable+2);
|
||||
const size_t length4 = cSrcSize - 6 - length1 - length2 - length3; // check coherency !!
|
||||
const char* const start1 = (const char*)(cSrc) + 6;
|
||||
const char* const start2 = start1 + length1;
|
||||
const char* const start3 = start2 + length2;
|
||||
const char* const start4 = start3 + length3;
|
||||
FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
|
||||
|
||||
if (reloadStatus!=FSE_DStream_completed) /* not complete : some bitStream might be FSE_DStream_unfinished */
|
||||
return (size_t)-FSE_ERROR_corruptionDetected;
|
||||
if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
|
||||
|
||||
/* tail */
|
||||
{
|
||||
// bitTail = bitD1; // *much* slower : -20% !??!
|
||||
FSE_DStream_t bitTail;
|
||||
bitTail.ptr = bitD1.ptr;
|
||||
bitTail.bitsConsumed = bitD1.bitsConsumed;
|
||||
bitTail.bitContainer = bitD1.bitContainer; // required in case of FSE_DStream_endOfBuffer
|
||||
bitTail.start = start1;
|
||||
for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
|
||||
errorCode = FSE_initDStream(&bitD1, start1, length1);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD2, start2, length2);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD3, start3, length3);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
errorCode = FSE_initDStream(&bitD4, start4, length4);
|
||||
if (FSE_isError(errorCode)) return errorCode;
|
||||
|
||||
reloadStatus=FSE_reloadDStream(&bitD2);
|
||||
|
||||
/* 16 symbols per loop */
|
||||
for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit); /* D2-3-4 are supposed to be synchronized and finish together */
|
||||
op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
|
||||
{
|
||||
HUF_DECODE_SYMBOL_0(0, bitTail);
|
||||
#define HUF_DECODE_SYMBOL_0(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
|
||||
|
||||
#define HUF_DECODE_SYMBOL_1(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
|
||||
if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
|
||||
|
||||
#define HUF_DECODE_SYMBOL_2(n, Dstream) \
|
||||
op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
|
||||
if (FSE_32bits()) FSE_reloadDStream(&Dstream)
|
||||
|
||||
HUF_DECODE_SYMBOL_1( 0, bitD1);
|
||||
HUF_DECODE_SYMBOL_1( 1, bitD2);
|
||||
HUF_DECODE_SYMBOL_1( 2, bitD3);
|
||||
HUF_DECODE_SYMBOL_1( 3, bitD4);
|
||||
HUF_DECODE_SYMBOL_2( 4, bitD1);
|
||||
HUF_DECODE_SYMBOL_2( 5, bitD2);
|
||||
HUF_DECODE_SYMBOL_2( 6, bitD3);
|
||||
HUF_DECODE_SYMBOL_2( 7, bitD4);
|
||||
HUF_DECODE_SYMBOL_1( 8, bitD1);
|
||||
HUF_DECODE_SYMBOL_1( 9, bitD2);
|
||||
HUF_DECODE_SYMBOL_1(10, bitD3);
|
||||
HUF_DECODE_SYMBOL_1(11, bitD4);
|
||||
HUF_DECODE_SYMBOL_0(12, bitD1);
|
||||
HUF_DECODE_SYMBOL_0(13, bitD2);
|
||||
HUF_DECODE_SYMBOL_0(14, bitD3);
|
||||
HUF_DECODE_SYMBOL_0(15, bitD4);
|
||||
}
|
||||
|
||||
if (FSE_endOfDStream(&bitTail))
|
||||
return op-ostart;
|
||||
if (reloadStatus!=FSE_DStream_completed) /* not complete : some bitStream might be FSE_DStream_unfinished */
|
||||
return (size_t)-FSE_ERROR_corruptionDetected;
|
||||
|
||||
/* tail */
|
||||
{
|
||||
// bitTail = bitD1; // *much* slower : -20% !??!
|
||||
FSE_DStream_t bitTail;
|
||||
bitTail.ptr = bitD1.ptr;
|
||||
bitTail.bitsConsumed = bitD1.bitsConsumed;
|
||||
bitTail.bitContainer = bitD1.bitContainer; // required in case of FSE_DStream_endOfBuffer
|
||||
bitTail.start = start1;
|
||||
for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
|
||||
{
|
||||
HUF_DECODE_SYMBOL_0(0, bitTail);
|
||||
}
|
||||
|
||||
if (FSE_endOfDStream(&bitTail))
|
||||
return op-ostart;
|
||||
}
|
||||
|
||||
if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall; /* dst buffer is full, but cSrc unfinished */
|
||||
|
||||
return (size_t)-FSE_ERROR_corruptionDetected;
|
||||
}
|
||||
|
||||
if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall; /* dst buffer is full, but cSrc unfinished */
|
||||
|
||||
return (size_t)-FSE_ERROR_corruptionDetected;
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -1998,91 +1998,92 @@ size_t HUFv05_decompress4X2_usingDTable(
|
|||
const void* cSrc, size_t cSrcSize,
|
||||
const U16* DTable)
|
||||
{
|
||||
const BYTE* const istart = (const BYTE*) cSrc;
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
const void* const dtPtr = DTable;
|
||||
const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
|
||||
const U32 dtLog = DTable[0];
|
||||
size_t errorCode;
|
||||
|
||||
/* Init */
|
||||
BITv05_DStream_t bitD1;
|
||||
BITv05_DStream_t bitD2;
|
||||
BITv05_DStream_t bitD3;
|
||||
BITv05_DStream_t bitD4;
|
||||
const size_t length1 = MEM_readLE16(istart);
|
||||
const size_t length2 = MEM_readLE16(istart+2);
|
||||
const size_t length3 = MEM_readLE16(istart+4);
|
||||
size_t length4;
|
||||
const BYTE* const istart1 = istart + 6; /* jumpTable */
|
||||
const BYTE* const istart2 = istart1 + length1;
|
||||
const BYTE* const istart3 = istart2 + length2;
|
||||
const BYTE* const istart4 = istart3 + length3;
|
||||
const size_t segmentSize = (dstSize+3) / 4;
|
||||
BYTE* const opStart2 = ostart + segmentSize;
|
||||
BYTE* const opStart3 = opStart2 + segmentSize;
|
||||
BYTE* const opStart4 = opStart3 + segmentSize;
|
||||
BYTE* op1 = ostart;
|
||||
BYTE* op2 = opStart2;
|
||||
BYTE* op3 = opStart3;
|
||||
BYTE* op4 = opStart4;
|
||||
U32 endSignal;
|
||||
|
||||
/* Check */
|
||||
if (cSrcSize < 10) return ERROR(corruption_detected); /* strict minimum : jump table + 1 byte per stream */
|
||||
{
|
||||
const BYTE* const istart = (const BYTE*) cSrc;
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
const void* const dtPtr = DTable;
|
||||
const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
|
||||
const U32 dtLog = DTable[0];
|
||||
size_t errorCode;
|
||||
|
||||
length4 = cSrcSize - (length1 + length2 + length3 + 6);
|
||||
if (length4 > cSrcSize) return ERROR(corruption_detected); /* overflow */
|
||||
errorCode = BITv05_initDStream(&bitD1, istart1, length1);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD2, istart2, length2);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD3, istart3, length3);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD4, istart4, length4);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
/* Init */
|
||||
BITv05_DStream_t bitD1;
|
||||
BITv05_DStream_t bitD2;
|
||||
BITv05_DStream_t bitD3;
|
||||
BITv05_DStream_t bitD4;
|
||||
const size_t length1 = MEM_readLE16(istart);
|
||||
const size_t length2 = MEM_readLE16(istart+2);
|
||||
const size_t length3 = MEM_readLE16(istart+4);
|
||||
size_t length4;
|
||||
const BYTE* const istart1 = istart + 6; /* jumpTable */
|
||||
const BYTE* const istart2 = istart1 + length1;
|
||||
const BYTE* const istart3 = istart2 + length2;
|
||||
const BYTE* const istart4 = istart3 + length3;
|
||||
const size_t segmentSize = (dstSize+3) / 4;
|
||||
BYTE* const opStart2 = ostart + segmentSize;
|
||||
BYTE* const opStart3 = opStart2 + segmentSize;
|
||||
BYTE* const opStart4 = opStart3 + segmentSize;
|
||||
BYTE* op1 = ostart;
|
||||
BYTE* op2 = opStart2;
|
||||
BYTE* op3 = opStart3;
|
||||
BYTE* op4 = opStart4;
|
||||
U32 endSignal;
|
||||
|
||||
/* 16-32 symbols per loop (4-8 symbols per stream) */
|
||||
endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
|
||||
for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
|
||||
HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
|
||||
length4 = cSrcSize - (length1 + length2 + length3 + 6);
|
||||
if (length4 > cSrcSize) return ERROR(corruption_detected); /* overflow */
|
||||
errorCode = BITv05_initDStream(&bitD1, istart1, length1);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD2, istart2, length2);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD3, istart3, length3);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
errorCode = BITv05_initDStream(&bitD4, istart4, length4);
|
||||
if (HUFv05_isError(errorCode)) return errorCode;
|
||||
|
||||
/* 16-32 symbols per loop (4-8 symbols per stream) */
|
||||
endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
|
||||
for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
|
||||
HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
|
||||
HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
|
||||
endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
|
||||
}
|
||||
|
||||
/* check corruption */
|
||||
if (op1 > opStart2) return ERROR(corruption_detected);
|
||||
if (op2 > opStart3) return ERROR(corruption_detected);
|
||||
if (op3 > opStart4) return ERROR(corruption_detected);
|
||||
/* note : op4 supposed already verified within main loop */
|
||||
|
||||
/* finish bitStreams one by one */
|
||||
HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op4, &bitD4, oend, dt, dtLog);
|
||||
|
||||
/* check */
|
||||
endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
|
||||
if (!endSignal) return ERROR(corruption_detected);
|
||||
|
||||
/* decoded size */
|
||||
return dstSize;
|
||||
}
|
||||
|
||||
/* check corruption */
|
||||
if (op1 > opStart2) return ERROR(corruption_detected);
|
||||
if (op2 > opStart3) return ERROR(corruption_detected);
|
||||
if (op3 > opStart4) return ERROR(corruption_detected);
|
||||
/* note : op4 supposed already verified within main loop */
|
||||
|
||||
/* finish bitStreams one by one */
|
||||
HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
|
||||
HUFv05_decodeStreamX2(op4, &bitD4, oend, dt, dtLog);
|
||||
|
||||
/* check */
|
||||
endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
|
||||
if (!endSignal) return ERROR(corruption_detected);
|
||||
|
||||
/* decoded size */
|
||||
return dstSize;
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue