mirror of
https://github.com/hashcat/hashcat.git
synced 2024-11-25 01:18:15 +00:00
1105 lines
28 KiB
C
1105 lines
28 KiB
C
/* Ppmd7.c -- PPMdH codec
|
|
2021-04-13 : Igor Pavlov : Public domain
|
|
This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
|
|
|
|
#include "Precomp.h"
|
|
|
|
#include <string.h>
|
|
|
|
#include "Ppmd7.h"
|
|
|
|
/* define PPMD7_ORDER_0_SUPPPORT to suport order-0 mode, unsupported by orignal PPMd var.H. code */
|
|
// #define PPMD7_ORDER_0_SUPPPORT
|
|
|
|
MY_ALIGN(16)
|
|
static const Byte PPMD7_kExpEscape[16] = { 25, 14, 9, 7, 5, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 2 };
|
|
MY_ALIGN(16)
|
|
static const UInt16 kInitBinEsc[] = { 0x3CDD, 0x1F3F, 0x59BF, 0x48F3, 0x64A1, 0x5ABC, 0x6632, 0x6051};
|
|
|
|
#define MAX_FREQ 124
|
|
#define UNIT_SIZE 12
|
|
|
|
#define U2B(nu) ((UInt32)(nu) * UNIT_SIZE)
|
|
#define U2I(nu) (p->Units2Indx[(size_t)(nu) - 1])
|
|
#define I2U(indx) ((unsigned)p->Indx2Units[indx])
|
|
#define I2U_UInt16(indx) ((UInt16)p->Indx2Units[indx])
|
|
|
|
#define REF(ptr) Ppmd_GetRef(p, ptr)
|
|
|
|
#define STATS_REF(ptr) ((CPpmd_State_Ref)REF(ptr))
|
|
|
|
#define CTX(ref) ((CPpmd7_Context *)Ppmd7_GetContext(p, ref))
|
|
#define STATS(ctx) Ppmd7_GetStats(p, ctx)
|
|
#define ONE_STATE(ctx) Ppmd7Context_OneState(ctx)
|
|
#define SUFFIX(ctx) CTX((ctx)->Suffix)
|
|
|
|
typedef CPpmd7_Context * CTX_PTR;
|
|
|
|
struct CPpmd7_Node_;
|
|
|
|
typedef Ppmd_Ref_Type(struct CPpmd7_Node_) CPpmd7_Node_Ref;
|
|
|
|
typedef struct CPpmd7_Node_
|
|
{
|
|
UInt16 Stamp; /* must be at offset 0 as CPpmd7_Context::NumStats. Stamp=0 means free */
|
|
UInt16 NU;
|
|
CPpmd7_Node_Ref Next; /* must be at offset >= 4 */
|
|
CPpmd7_Node_Ref Prev;
|
|
} CPpmd7_Node;
|
|
|
|
#define NODE(r) Ppmd_GetPtr_Type(p, r, CPpmd7_Node)
|
|
|
|
void Ppmd7_Construct(CPpmd7 *p)
|
|
{
|
|
unsigned i, k, m;
|
|
|
|
p->Base = NULL;
|
|
|
|
for (i = 0, k = 0; i < PPMD_NUM_INDEXES; i++)
|
|
{
|
|
unsigned step = (i >= 12 ? 4 : (i >> 2) + 1);
|
|
do { p->Units2Indx[k++] = (Byte)i; } while (--step);
|
|
p->Indx2Units[i] = (Byte)k;
|
|
}
|
|
|
|
p->NS2BSIndx[0] = (0 << 1);
|
|
p->NS2BSIndx[1] = (1 << 1);
|
|
memset(p->NS2BSIndx + 2, (2 << 1), 9);
|
|
memset(p->NS2BSIndx + 11, (3 << 1), 256 - 11);
|
|
|
|
for (i = 0; i < 3; i++)
|
|
p->NS2Indx[i] = (Byte)i;
|
|
|
|
for (m = i, k = 1; i < 256; i++)
|
|
{
|
|
p->NS2Indx[i] = (Byte)m;
|
|
if (--k == 0)
|
|
k = (++m) - 2;
|
|
}
|
|
|
|
memcpy(p->ExpEscape, PPMD7_kExpEscape, 16);
|
|
}
|
|
|
|
|
|
void Ppmd7_Free(CPpmd7 *p, ISzAllocPtr alloc)
|
|
{
|
|
ISzAlloc_Free(alloc, p->Base);
|
|
p->Size = 0;
|
|
p->Base = NULL;
|
|
}
|
|
|
|
|
|
BoolInt Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAllocPtr alloc)
|
|
{
|
|
if (!p->Base || p->Size != size)
|
|
{
|
|
Ppmd7_Free(p, alloc);
|
|
p->AlignOffset = (4 - size) & 3;
|
|
if ((p->Base = (Byte *)ISzAlloc_Alloc(alloc, p->AlignOffset + size)) == NULL)
|
|
return False;
|
|
p->Size = size;
|
|
}
|
|
return True;
|
|
}
|
|
|
|
|
|
|
|
// ---------- Internal Memory Allocator ----------
|
|
|
|
/* We can use CPpmd7_Node in list of free units (as in Ppmd8)
|
|
But we still need one additional list walk pass in GlueFreeBlocks().
|
|
So we use simple CPpmd_Void_Ref instead of CPpmd7_Node in InsertNode() / RemoveNode()
|
|
*/
|
|
|
|
#define EMPTY_NODE 0
|
|
|
|
|
|
static void InsertNode(CPpmd7 *p, void *node, unsigned indx)
|
|
{
|
|
*((CPpmd_Void_Ref *)node) = p->FreeList[indx];
|
|
// ((CPpmd7_Node *)node)->Next = (CPpmd7_Node_Ref)p->FreeList[indx];
|
|
|
|
p->FreeList[indx] = REF(node);
|
|
|
|
}
|
|
|
|
|
|
static void *RemoveNode(CPpmd7 *p, unsigned indx)
|
|
{
|
|
CPpmd_Void_Ref *node = (CPpmd_Void_Ref *)Ppmd7_GetPtr(p, p->FreeList[indx]);
|
|
p->FreeList[indx] = *node;
|
|
// CPpmd7_Node *node = NODE((CPpmd7_Node_Ref)p->FreeList[indx]);
|
|
// p->FreeList[indx] = node->Next;
|
|
return node;
|
|
}
|
|
|
|
|
|
static void SplitBlock(CPpmd7 *p, void *ptr, unsigned oldIndx, unsigned newIndx)
|
|
{
|
|
unsigned i, nu = I2U(oldIndx) - I2U(newIndx);
|
|
ptr = (Byte *)ptr + U2B(I2U(newIndx));
|
|
if (I2U(i = U2I(nu)) != nu)
|
|
{
|
|
unsigned k = I2U(--i);
|
|
InsertNode(p, ((Byte *)ptr) + U2B(k), nu - k - 1);
|
|
}
|
|
InsertNode(p, ptr, i);
|
|
}
|
|
|
|
|
|
/* we use CPpmd7_Node_Union union to solve XLC -O2 strict pointer aliasing problem */
|
|
|
|
typedef union _CPpmd7_Node_Union
|
|
{
|
|
CPpmd7_Node Node;
|
|
CPpmd7_Node_Ref NextRef;
|
|
} CPpmd7_Node_Union;
|
|
|
|
/* Original PPmdH (Ppmd7) code uses doubly linked list in GlueFreeBlocks()
|
|
we use single linked list similar to Ppmd8 code */
|
|
|
|
|
|
static void GlueFreeBlocks(CPpmd7 *p)
|
|
{
|
|
/*
|
|
we use first UInt16 field of 12-bytes UNITs as record type stamp
|
|
CPpmd_State { Byte Symbol; Byte Freq; : Freq != 0
|
|
CPpmd7_Context { UInt16 NumStats; : NumStats != 0
|
|
CPpmd7_Node { UInt16 Stamp : Stamp == 0 for free record
|
|
: Stamp == 1 for head record and guard
|
|
Last 12-bytes UNIT in array is always contains 12-bytes order-0 CPpmd7_Context record.
|
|
*/
|
|
CPpmd7_Node_Ref head, n = 0;
|
|
|
|
p->GlueCount = 255;
|
|
|
|
|
|
/* we set guard NODE at LoUnit */
|
|
if (p->LoUnit != p->HiUnit)
|
|
((CPpmd7_Node *)(void *)p->LoUnit)->Stamp = 1;
|
|
|
|
{
|
|
/* Create list of free blocks.
|
|
We still need one additional list walk pass before Glue. */
|
|
unsigned i;
|
|
for (i = 0; i < PPMD_NUM_INDEXES; i++)
|
|
{
|
|
const UInt16 nu = I2U_UInt16(i);
|
|
CPpmd7_Node_Ref next = (CPpmd7_Node_Ref)p->FreeList[i];
|
|
p->FreeList[i] = 0;
|
|
while (next != 0)
|
|
{
|
|
/* Don't change the order of the following commands: */
|
|
CPpmd7_Node_Union *un = (CPpmd7_Node_Union *)NODE(next);
|
|
const CPpmd7_Node_Ref tmp = next;
|
|
next = un->NextRef;
|
|
un->Node.Stamp = EMPTY_NODE;
|
|
un->Node.NU = nu;
|
|
un->Node.Next = n;
|
|
n = tmp;
|
|
}
|
|
}
|
|
}
|
|
|
|
head = n;
|
|
/* Glue and Fill must walk the list in same direction */
|
|
{
|
|
/* Glue free blocks */
|
|
CPpmd7_Node_Ref *prev = &head;
|
|
while (n)
|
|
{
|
|
CPpmd7_Node *node = NODE(n);
|
|
UInt32 nu = node->NU;
|
|
n = node->Next;
|
|
if (nu == 0)
|
|
{
|
|
*prev = n;
|
|
continue;
|
|
}
|
|
prev = &node->Next;
|
|
for (;;)
|
|
{
|
|
CPpmd7_Node *node2 = node + nu;
|
|
nu += node2->NU;
|
|
if (node2->Stamp != EMPTY_NODE || nu >= 0x10000)
|
|
break;
|
|
node->NU = (UInt16)nu;
|
|
node2->NU = 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* Fill lists of free blocks */
|
|
for (n = head; n != 0;)
|
|
{
|
|
CPpmd7_Node *node = NODE(n);
|
|
UInt32 nu = node->NU;
|
|
unsigned i;
|
|
n = node->Next;
|
|
if (nu == 0)
|
|
continue;
|
|
for (; nu > 128; nu -= 128, node += 128)
|
|
InsertNode(p, node, PPMD_NUM_INDEXES - 1);
|
|
if (I2U(i = U2I(nu)) != nu)
|
|
{
|
|
unsigned k = I2U(--i);
|
|
InsertNode(p, node + k, (unsigned)nu - k - 1);
|
|
}
|
|
InsertNode(p, node, i);
|
|
}
|
|
}
|
|
|
|
|
|
MY_NO_INLINE
|
|
static void *AllocUnitsRare(CPpmd7 *p, unsigned indx)
|
|
{
|
|
unsigned i;
|
|
|
|
if (p->GlueCount == 0)
|
|
{
|
|
GlueFreeBlocks(p);
|
|
if (p->FreeList[indx] != 0)
|
|
return RemoveNode(p, indx);
|
|
}
|
|
|
|
i = indx;
|
|
|
|
do
|
|
{
|
|
if (++i == PPMD_NUM_INDEXES)
|
|
{
|
|
UInt32 numBytes = U2B(I2U(indx));
|
|
Byte *us = p->UnitsStart;
|
|
p->GlueCount--;
|
|
return ((UInt32)(us - p->Text) > numBytes) ? (p->UnitsStart = us - numBytes) : NULL;
|
|
}
|
|
}
|
|
while (p->FreeList[i] == 0);
|
|
|
|
{
|
|
void *block = RemoveNode(p, i);
|
|
SplitBlock(p, block, i, indx);
|
|
return block;
|
|
}
|
|
}
|
|
|
|
|
|
static void *AllocUnits(CPpmd7 *p, unsigned indx)
|
|
{
|
|
if (p->FreeList[indx] != 0)
|
|
return RemoveNode(p, indx);
|
|
{
|
|
UInt32 numBytes = U2B(I2U(indx));
|
|
Byte *lo = p->LoUnit;
|
|
if ((UInt32)(p->HiUnit - lo) >= numBytes)
|
|
{
|
|
p->LoUnit = lo + numBytes;
|
|
return lo;
|
|
}
|
|
}
|
|
return AllocUnitsRare(p, indx);
|
|
}
|
|
|
|
|
|
#define MyMem12Cpy(dest, src, num) \
|
|
{ UInt32 *d = (UInt32 *)dest; const UInt32 *z = (const UInt32 *)src; UInt32 n = num; \
|
|
do { d[0] = z[0]; d[1] = z[1]; d[2] = z[2]; z += 3; d += 3; } while (--n); }
|
|
|
|
|
|
/*
|
|
static void *ShrinkUnits(CPpmd7 *p, void *oldPtr, unsigned oldNU, unsigned newNU)
|
|
{
|
|
unsigned i0 = U2I(oldNU);
|
|
unsigned i1 = U2I(newNU);
|
|
if (i0 == i1)
|
|
return oldPtr;
|
|
if (p->FreeList[i1] != 0)
|
|
{
|
|
void *ptr = RemoveNode(p, i1);
|
|
MyMem12Cpy(ptr, oldPtr, newNU);
|
|
InsertNode(p, oldPtr, i0);
|
|
return ptr;
|
|
}
|
|
SplitBlock(p, oldPtr, i0, i1);
|
|
return oldPtr;
|
|
}
|
|
*/
|
|
|
|
|
|
#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
|
|
static void SetSuccessor(CPpmd_State *p, CPpmd_Void_Ref v)
|
|
{
|
|
Ppmd_SET_SUCCESSOR(p, v);
|
|
}
|
|
|
|
|
|
|
|
MY_NO_INLINE
|
|
static
|
|
void RestartModel(CPpmd7 *p)
|
|
{
|
|
unsigned i, k;
|
|
|
|
memset(p->FreeList, 0, sizeof(p->FreeList));
|
|
|
|
p->Text = p->Base + p->AlignOffset;
|
|
p->HiUnit = p->Text + p->Size;
|
|
p->LoUnit = p->UnitsStart = p->HiUnit - p->Size / 8 / UNIT_SIZE * 7 * UNIT_SIZE;
|
|
p->GlueCount = 0;
|
|
|
|
p->OrderFall = p->MaxOrder;
|
|
p->RunLength = p->InitRL = -(Int32)((p->MaxOrder < 12) ? p->MaxOrder : 12) - 1;
|
|
p->PrevSuccess = 0;
|
|
|
|
{
|
|
CPpmd7_Context *mc = (CTX_PTR)(void *)(p->HiUnit -= UNIT_SIZE); /* AllocContext(p); */
|
|
CPpmd_State *s = (CPpmd_State *)p->LoUnit; /* AllocUnits(p, PPMD_NUM_INDEXES - 1); */
|
|
|
|
p->LoUnit += U2B(256 / 2);
|
|
p->MaxContext = p->MinContext = mc;
|
|
p->FoundState = s;
|
|
|
|
mc->NumStats = 256;
|
|
mc->Union2.SummFreq = 256 + 1;
|
|
mc->Union4.Stats = REF(s);
|
|
mc->Suffix = 0;
|
|
|
|
for (i = 0; i < 256; i++, s++)
|
|
{
|
|
s->Symbol = (Byte)i;
|
|
s->Freq = 1;
|
|
SetSuccessor(s, 0);
|
|
}
|
|
|
|
#ifdef PPMD7_ORDER_0_SUPPPORT
|
|
if (p->MaxOrder == 0)
|
|
{
|
|
CPpmd_Void_Ref r = REF(mc);
|
|
s = p->FoundState;
|
|
for (i = 0; i < 256; i++, s++)
|
|
SetSuccessor(s, r);
|
|
return;
|
|
}
|
|
#endif
|
|
}
|
|
|
|
for (i = 0; i < 128; i++)
|
|
|
|
|
|
|
|
for (k = 0; k < 8; k++)
|
|
{
|
|
unsigned m;
|
|
UInt16 *dest = p->BinSumm[i] + k;
|
|
UInt16 val = (UInt16)(PPMD_BIN_SCALE - kInitBinEsc[k] / (i + 2));
|
|
for (m = 0; m < 64; m += 8)
|
|
dest[m] = val;
|
|
}
|
|
|
|
|
|
for (i = 0; i < 25; i++)
|
|
{
|
|
|
|
CPpmd_See *s = p->See[i];
|
|
|
|
|
|
|
|
unsigned summ = ((5 * i + 10) << (PPMD_PERIOD_BITS - 4));
|
|
for (k = 0; k < 16; k++, s++)
|
|
{
|
|
s->Summ = (UInt16)summ;
|
|
s->Shift = (PPMD_PERIOD_BITS - 4);
|
|
s->Count = 4;
|
|
}
|
|
}
|
|
|
|
p->DummySee.Summ = 0; /* unused */
|
|
p->DummySee.Shift = PPMD_PERIOD_BITS;
|
|
p->DummySee.Count = 64; /* unused */
|
|
}
|
|
|
|
|
|
void Ppmd7_Init(CPpmd7 *p, unsigned maxOrder)
|
|
{
|
|
p->MaxOrder = maxOrder;
|
|
|
|
RestartModel(p);
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
CreateSuccessors()
|
|
It's called when (FoundState->Successor) is RAW-Successor,
|
|
that is the link to position in Raw text.
|
|
So we create Context records and write the links to
|
|
FoundState->Successor and to identical RAW-Successors in suffix
|
|
contexts of MinContex.
|
|
|
|
The function returns:
|
|
if (OrderFall == 0) then MinContext is already at MAX order,
|
|
{ return pointer to new or existing context of same MAX order }
|
|
else
|
|
{ return pointer to new real context that will be (Order+1) in comparison with MinContext
|
|
|
|
also it can return pointer to real context of same order,
|
|
*/
|
|
|
|
MY_NO_INLINE
|
|
static CTX_PTR CreateSuccessors(CPpmd7 *p)
|
|
{
|
|
CTX_PTR c = p->MinContext;
|
|
CPpmd_Byte_Ref upBranch = (CPpmd_Byte_Ref)SUCCESSOR(p->FoundState);
|
|
Byte newSym, newFreq;
|
|
unsigned numPs = 0;
|
|
CPpmd_State *ps[PPMD7_MAX_ORDER];
|
|
|
|
if (p->OrderFall != 0)
|
|
ps[numPs++] = p->FoundState;
|
|
|
|
while (c->Suffix)
|
|
{
|
|
CPpmd_Void_Ref successor;
|
|
CPpmd_State *s;
|
|
c = SUFFIX(c);
|
|
|
|
|
|
if (c->NumStats != 1)
|
|
{
|
|
Byte sym = p->FoundState->Symbol;
|
|
for (s = STATS(c); s->Symbol != sym; s++);
|
|
|
|
}
|
|
else
|
|
{
|
|
s = ONE_STATE(c);
|
|
|
|
}
|
|
successor = SUCCESSOR(s);
|
|
if (successor != upBranch)
|
|
{
|
|
// (c) is real record Context here,
|
|
c = CTX(successor);
|
|
if (numPs == 0)
|
|
{
|
|
// (c) is real record MAX Order Context here,
|
|
// So we don't need to create any new contexts.
|
|
return c;
|
|
}
|
|
break;
|
|
}
|
|
ps[numPs++] = s;
|
|
}
|
|
|
|
// All created contexts will have single-symbol with new RAW-Successor
|
|
// All new RAW-Successors will point to next position in RAW text
|
|
// after FoundState->Successor
|
|
|
|
newSym = *(const Byte *)Ppmd7_GetPtr(p, upBranch);
|
|
upBranch++;
|
|
|
|
|
|
if (c->NumStats == 1)
|
|
newFreq = ONE_STATE(c)->Freq;
|
|
else
|
|
{
|
|
UInt32 cf, s0;
|
|
CPpmd_State *s;
|
|
for (s = STATS(c); s->Symbol != newSym; s++);
|
|
cf = (UInt32)s->Freq - 1;
|
|
s0 = (UInt32)c->Union2.SummFreq - c->NumStats - cf;
|
|
/*
|
|
cf - is frequency of symbol that will be Successor in new context records.
|
|
s0 - is commulative frequency sum of another symbols from parent context.
|
|
max(newFreq)= (s->Freq + 1), when (s0 == 1)
|
|
we have requirement (Ppmd7Context_OneState()->Freq <= 128) in BinSumm[]
|
|
so (s->Freq < 128) - is requirement for multi-symbol contexts
|
|
*/
|
|
newFreq = (Byte)(1 + ((2 * cf <= s0) ? (5 * cf > s0) : (2 * cf + s0 - 1) / (2 * s0) + 1));
|
|
}
|
|
|
|
// Create new single-symbol contexts from low order to high order in loop
|
|
|
|
do
|
|
{
|
|
CTX_PTR c1;
|
|
/* = AllocContext(p); */
|
|
if (p->HiUnit != p->LoUnit)
|
|
c1 = (CTX_PTR)(void *)(p->HiUnit -= UNIT_SIZE);
|
|
else if (p->FreeList[0] != 0)
|
|
c1 = (CTX_PTR)RemoveNode(p, 0);
|
|
else
|
|
{
|
|
c1 = (CTX_PTR)AllocUnitsRare(p, 0);
|
|
if (!c1)
|
|
return NULL;
|
|
}
|
|
|
|
c1->NumStats = 1;
|
|
ONE_STATE(c1)->Symbol = newSym;
|
|
ONE_STATE(c1)->Freq = newFreq;
|
|
SetSuccessor(ONE_STATE(c1), upBranch);
|
|
c1->Suffix = REF(c);
|
|
SetSuccessor(ps[--numPs], REF(c1));
|
|
c = c1;
|
|
}
|
|
while (numPs != 0);
|
|
|
|
return c;
|
|
}
|
|
|
|
|
|
|
|
#define SwapStates(s) \
|
|
{ CPpmd_State tmp = s[0]; s[0] = s[-1]; s[-1] = tmp; }
|
|
|
|
|
|
void Ppmd7_UpdateModel(CPpmd7 *p);
|
|
MY_NO_INLINE
|
|
void Ppmd7_UpdateModel(CPpmd7 *p)
|
|
{
|
|
CPpmd_Void_Ref maxSuccessor, minSuccessor;
|
|
CTX_PTR c, mc;
|
|
unsigned s0, ns;
|
|
|
|
|
|
|
|
if (p->FoundState->Freq < MAX_FREQ / 4 && p->MinContext->Suffix != 0)
|
|
{
|
|
/* Update Freqs in Suffix Context */
|
|
|
|
c = SUFFIX(p->MinContext);
|
|
|
|
if (c->NumStats == 1)
|
|
{
|
|
CPpmd_State *s = ONE_STATE(c);
|
|
if (s->Freq < 32)
|
|
s->Freq++;
|
|
}
|
|
else
|
|
{
|
|
CPpmd_State *s = STATS(c);
|
|
Byte sym = p->FoundState->Symbol;
|
|
|
|
if (s->Symbol != sym)
|
|
{
|
|
do
|
|
{
|
|
// s++; if (s->Symbol == sym) break;
|
|
s++;
|
|
}
|
|
while (s->Symbol != sym);
|
|
|
|
if (s[0].Freq >= s[-1].Freq)
|
|
{
|
|
SwapStates(s);
|
|
s--;
|
|
}
|
|
}
|
|
|
|
if (s->Freq < MAX_FREQ - 9)
|
|
{
|
|
s->Freq = (Byte)(s->Freq + 2);
|
|
c->Union2.SummFreq = (UInt16)(c->Union2.SummFreq + 2);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
if (p->OrderFall == 0)
|
|
{
|
|
/* MAX ORDER context */
|
|
/* (FoundState->Successor) is RAW-Successor. */
|
|
p->MaxContext = p->MinContext = CreateSuccessors(p);
|
|
if (!p->MinContext)
|
|
{
|
|
RestartModel(p);
|
|
return;
|
|
}
|
|
SetSuccessor(p->FoundState, REF(p->MinContext));
|
|
return;
|
|
}
|
|
|
|
|
|
/* NON-MAX ORDER context */
|
|
|
|
{
|
|
Byte *text = p->Text;
|
|
*text++ = p->FoundState->Symbol;
|
|
p->Text = text;
|
|
if (text >= p->UnitsStart)
|
|
{
|
|
RestartModel(p);
|
|
return;
|
|
}
|
|
maxSuccessor = REF(text);
|
|
}
|
|
|
|
minSuccessor = SUCCESSOR(p->FoundState);
|
|
|
|
if (minSuccessor)
|
|
{
|
|
// there is Successor for FoundState in MinContext.
|
|
// So the next context will be one order higher than MinContext.
|
|
|
|
if (minSuccessor <= maxSuccessor)
|
|
{
|
|
// minSuccessor is RAW-Successor. So we will create real contexts records:
|
|
CTX_PTR cs = CreateSuccessors(p);
|
|
if (!cs)
|
|
{
|
|
RestartModel(p);
|
|
return;
|
|
}
|
|
minSuccessor = REF(cs);
|
|
}
|
|
|
|
// minSuccessor now is real Context pointer that points to existing (Order+1) context
|
|
|
|
if (--p->OrderFall == 0)
|
|
{
|
|
/*
|
|
if we move to MaxOrder context, then minSuccessor will be common Succesor for both:
|
|
MinContext that is (MaxOrder - 1)
|
|
MaxContext that is (MaxOrder)
|
|
so we don't need new RAW-Successor, and we can use real minSuccessor
|
|
as succssors for both MinContext and MaxContext.
|
|
*/
|
|
maxSuccessor = minSuccessor;
|
|
|
|
/*
|
|
if (MaxContext != MinContext)
|
|
{
|
|
there was order fall from MaxOrder and we don't need current symbol
|
|
to transfer some RAW-Succesors to real contexts.
|
|
So we roll back pointer in raw data for one position.
|
|
}
|
|
*/
|
|
p->Text -= (p->MaxContext != p->MinContext);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
/*
|
|
FoundState has NULL-Successor here.
|
|
And only root 0-order context can contain NULL-Successors.
|
|
We change Successor in FoundState to RAW-Successor,
|
|
And next context will be same 0-order root Context.
|
|
*/
|
|
SetSuccessor(p->FoundState, maxSuccessor);
|
|
minSuccessor = REF(p->MinContext);
|
|
}
|
|
|
|
mc = p->MinContext;
|
|
c = p->MaxContext;
|
|
|
|
p->MaxContext = p->MinContext = CTX(minSuccessor);
|
|
|
|
if (c == mc)
|
|
return;
|
|
|
|
// s0 : is pure Escape Freq
|
|
s0 = mc->Union2.SummFreq - (ns = mc->NumStats) - ((unsigned)p->FoundState->Freq - 1);
|
|
|
|
do
|
|
{
|
|
unsigned ns1;
|
|
UInt32 sum;
|
|
|
|
if ((ns1 = c->NumStats) != 1)
|
|
{
|
|
if ((ns1 & 1) == 0)
|
|
{
|
|
/* Expand for one UNIT */
|
|
unsigned oldNU = ns1 >> 1;
|
|
unsigned i = U2I(oldNU);
|
|
if (i != U2I((size_t)oldNU + 1))
|
|
{
|
|
void *ptr = AllocUnits(p, i + 1);
|
|
void *oldPtr;
|
|
if (!ptr)
|
|
{
|
|
RestartModel(p);
|
|
return;
|
|
}
|
|
oldPtr = STATS(c);
|
|
MyMem12Cpy(ptr, oldPtr, oldNU);
|
|
InsertNode(p, oldPtr, i);
|
|
c->Union4.Stats = STATS_REF(ptr);
|
|
}
|
|
}
|
|
sum = c->Union2.SummFreq;
|
|
/* max increase of Escape_Freq is 3 here.
|
|
total increase of Union2.SummFreq for all symbols is less than 256 here */
|
|
sum += (UInt32)(2 * ns1 < ns) + 2 * ((unsigned)(4 * ns1 <= ns) & (sum <= 8 * ns1));
|
|
/* original PPMdH uses 16-bit variable for (sum) here.
|
|
But (sum < 0x9000). So we don't truncate (sum) to 16-bit */
|
|
// sum = (UInt16)sum;
|
|
}
|
|
else
|
|
{
|
|
// instead of One-symbol context we create 2-symbol context
|
|
CPpmd_State *s = (CPpmd_State*)AllocUnits(p, 0);
|
|
if (!s)
|
|
{
|
|
RestartModel(p);
|
|
return;
|
|
}
|
|
{
|
|
unsigned freq = c->Union2.State2.Freq;
|
|
// s = *ONE_STATE(c);
|
|
s->Symbol = c->Union2.State2.Symbol;
|
|
s->Successor_0 = c->Union4.State4.Successor_0;
|
|
s->Successor_1 = c->Union4.State4.Successor_1;
|
|
// SetSuccessor(s, c->Union4.Stats); // call it only for debug purposes to check the order of
|
|
// (Successor_0 and Successor_1) in LE/BE.
|
|
c->Union4.Stats = REF(s);
|
|
if (freq < MAX_FREQ / 4 - 1)
|
|
freq <<= 1;
|
|
else
|
|
freq = MAX_FREQ - 4;
|
|
// (max(s->freq) == 120), when we convert from 1-symbol into 2-symbol context
|
|
s->Freq = (Byte)freq;
|
|
// max(InitEsc = PPMD7_kExpEscape[*]) is 25. So the max(escapeFreq) is 26 here
|
|
sum = freq + p->InitEsc + (ns > 3);
|
|
}
|
|
}
|
|
|
|
{
|
|
CPpmd_State *s = STATS(c) + ns1;
|
|
UInt32 cf = 2 * (sum + 6) * (UInt32)p->FoundState->Freq;
|
|
UInt32 sf = (UInt32)s0 + sum;
|
|
s->Symbol = p->FoundState->Symbol;
|
|
c->NumStats = (UInt16)(ns1 + 1);
|
|
SetSuccessor(s, maxSuccessor);
|
|
|
|
if (cf < 6 * sf)
|
|
{
|
|
cf = (UInt32)1 + (cf > sf) + (cf >= 4 * sf);
|
|
sum += 3;
|
|
/* It can add (0, 1, 2) to Escape_Freq */
|
|
}
|
|
else
|
|
{
|
|
cf = (UInt32)4 + (cf >= 9 * sf) + (cf >= 12 * sf) + (cf >= 15 * sf);
|
|
sum += cf;
|
|
}
|
|
|
|
c->Union2.SummFreq = (UInt16)sum;
|
|
s->Freq = (Byte)cf;
|
|
}
|
|
c = SUFFIX(c);
|
|
}
|
|
while (c != mc);
|
|
}
|
|
|
|
|
|
|
|
MY_NO_INLINE
|
|
static void Rescale(CPpmd7 *p)
|
|
{
|
|
unsigned i, adder, sumFreq, escFreq;
|
|
CPpmd_State *stats = STATS(p->MinContext);
|
|
CPpmd_State *s = p->FoundState;
|
|
|
|
/* Sort the list by Freq */
|
|
if (s != stats)
|
|
{
|
|
CPpmd_State tmp = *s;
|
|
do
|
|
s[0] = s[-1];
|
|
while (--s != stats);
|
|
*s = tmp;
|
|
}
|
|
|
|
sumFreq = s->Freq;
|
|
escFreq = p->MinContext->Union2.SummFreq - sumFreq;
|
|
|
|
/*
|
|
if (p->OrderFall == 0), adder = 0 : it's allowed to remove symbol from MAX Order context
|
|
if (p->OrderFall != 0), adder = 1 : it's NOT allowed to remove symbol from NON-MAX Order context
|
|
*/
|
|
|
|
adder = (p->OrderFall != 0);
|
|
|
|
#ifdef PPMD7_ORDER_0_SUPPPORT
|
|
adder |= (p->MaxOrder == 0); // we don't remove symbols from order-0 context
|
|
#endif
|
|
|
|
sumFreq = (sumFreq + 4 + adder) >> 1;
|
|
i = (unsigned)p->MinContext->NumStats - 1;
|
|
s->Freq = (Byte)sumFreq;
|
|
|
|
do
|
|
{
|
|
unsigned freq = (++s)->Freq;
|
|
escFreq -= freq;
|
|
freq = (freq + adder) >> 1;
|
|
sumFreq += freq;
|
|
s->Freq = (Byte)freq;
|
|
if (freq > s[-1].Freq)
|
|
{
|
|
CPpmd_State tmp = *s;
|
|
CPpmd_State *s1 = s;
|
|
do
|
|
{
|
|
s1[0] = s1[-1];
|
|
}
|
|
while (--s1 != stats && freq > s1[-1].Freq);
|
|
*s1 = tmp;
|
|
}
|
|
}
|
|
while (--i);
|
|
|
|
if (s->Freq == 0)
|
|
{
|
|
/* Remove all items with Freq == 0 */
|
|
CPpmd7_Context *mc;
|
|
unsigned numStats, numStatsNew, n0, n1;
|
|
|
|
i = 0; do { i++; } while ((--s)->Freq == 0);
|
|
|
|
/* We increase (escFreq) for the number of removed symbols.
|
|
So we will have (0.5) increase for Escape_Freq in avarage per
|
|
removed symbol after Escape_Freq halving */
|
|
escFreq += i;
|
|
mc = p->MinContext;
|
|
numStats = mc->NumStats;
|
|
numStatsNew = numStats - i;
|
|
mc->NumStats = (UInt16)(numStatsNew);
|
|
n0 = (numStats + 1) >> 1;
|
|
|
|
if (numStatsNew == 1)
|
|
{
|
|
/* Create Single-Symbol context */
|
|
unsigned freq = stats->Freq;
|
|
|
|
do
|
|
{
|
|
escFreq >>= 1;
|
|
freq = (freq + 1) >> 1;
|
|
}
|
|
while (escFreq > 1);
|
|
|
|
s = ONE_STATE(mc);
|
|
*s = *stats;
|
|
s->Freq = (Byte)freq; // (freq <= 260 / 4)
|
|
p->FoundState = s;
|
|
InsertNode(p, stats, U2I(n0));
|
|
return;
|
|
}
|
|
|
|
n1 = (numStatsNew + 1) >> 1;
|
|
if (n0 != n1)
|
|
{
|
|
// p->MinContext->Union4.Stats = STATS_REF(ShrinkUnits(p, stats, n0, n1));
|
|
unsigned i0 = U2I(n0);
|
|
unsigned i1 = U2I(n1);
|
|
if (i0 != i1)
|
|
{
|
|
if (p->FreeList[i1] != 0)
|
|
{
|
|
void *ptr = RemoveNode(p, i1);
|
|
p->MinContext->Union4.Stats = STATS_REF(ptr);
|
|
MyMem12Cpy(ptr, (const void *)stats, n1);
|
|
InsertNode(p, stats, i0);
|
|
}
|
|
else
|
|
SplitBlock(p, stats, i0, i1);
|
|
}
|
|
}
|
|
}
|
|
{
|
|
CPpmd7_Context *mc = p->MinContext;
|
|
mc->Union2.SummFreq = (UInt16)(sumFreq + escFreq - (escFreq >> 1));
|
|
// Escape_Freq halving here
|
|
p->FoundState = STATS(mc);
|
|
}
|
|
}
|
|
|
|
|
|
CPpmd_See *Ppmd7_MakeEscFreq(CPpmd7 *p, unsigned numMasked, UInt32 *escFreq)
|
|
{
|
|
CPpmd_See *see;
|
|
const CPpmd7_Context *mc = p->MinContext;
|
|
unsigned numStats = mc->NumStats;
|
|
if (numStats != 256)
|
|
{
|
|
unsigned nonMasked = numStats - numMasked;
|
|
see = p->See[(unsigned)p->NS2Indx[(size_t)nonMasked - 1]]
|
|
+ (nonMasked < (unsigned)SUFFIX(mc)->NumStats - numStats)
|
|
+ 2 * (unsigned)(mc->Union2.SummFreq < 11 * numStats)
|
|
+ 4 * (unsigned)(numMasked > nonMasked) +
|
|
p->HiBitsFlag;
|
|
{
|
|
// if (see->Summ) field is larger than 16-bit, we need only low 16 bits of Summ
|
|
unsigned summ = (UInt16)see->Summ; // & 0xFFFF
|
|
unsigned r = (summ >> see->Shift);
|
|
see->Summ = (UInt16)(summ - r);
|
|
*escFreq = r + (r == 0);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
see = &p->DummySee;
|
|
*escFreq = 1;
|
|
}
|
|
return see;
|
|
}
|
|
|
|
|
|
static void NextContext(CPpmd7 *p)
|
|
{
|
|
CTX_PTR c = CTX(SUCCESSOR(p->FoundState));
|
|
if (p->OrderFall == 0 && (const Byte *)c > p->Text)
|
|
p->MaxContext = p->MinContext = c;
|
|
else
|
|
Ppmd7_UpdateModel(p);
|
|
}
|
|
|
|
|
|
void Ppmd7_Update1(CPpmd7 *p)
|
|
{
|
|
CPpmd_State *s = p->FoundState;
|
|
unsigned freq = s->Freq;
|
|
freq += 4;
|
|
p->MinContext->Union2.SummFreq = (UInt16)(p->MinContext->Union2.SummFreq + 4);
|
|
s->Freq = (Byte)freq;
|
|
if (freq > s[-1].Freq)
|
|
{
|
|
SwapStates(s);
|
|
p->FoundState = --s;
|
|
if (freq > MAX_FREQ)
|
|
Rescale(p);
|
|
}
|
|
NextContext(p);
|
|
}
|
|
|
|
|
|
void Ppmd7_Update1_0(CPpmd7 *p)
|
|
{
|
|
CPpmd_State *s = p->FoundState;
|
|
CPpmd7_Context *mc = p->MinContext;
|
|
unsigned freq = s->Freq;
|
|
unsigned summFreq = mc->Union2.SummFreq;
|
|
p->PrevSuccess = (2 * freq > summFreq);
|
|
p->RunLength += (int)p->PrevSuccess;
|
|
mc->Union2.SummFreq = (UInt16)(summFreq + 4);
|
|
freq += 4;
|
|
s->Freq = (Byte)freq;
|
|
if (freq > MAX_FREQ)
|
|
Rescale(p);
|
|
NextContext(p);
|
|
}
|
|
|
|
|
|
/*
|
|
void Ppmd7_UpdateBin(CPpmd7 *p)
|
|
{
|
|
unsigned freq = p->FoundState->Freq;
|
|
p->FoundState->Freq = (Byte)(freq + (freq < 128));
|
|
p->PrevSuccess = 1;
|
|
p->RunLength++;
|
|
NextContext(p);
|
|
}
|
|
*/
|
|
|
|
void Ppmd7_Update2(CPpmd7 *p)
|
|
{
|
|
CPpmd_State *s = p->FoundState;
|
|
unsigned freq = s->Freq;
|
|
freq += 4;
|
|
p->RunLength = p->InitRL;
|
|
p->MinContext->Union2.SummFreq = (UInt16)(p->MinContext->Union2.SummFreq + 4);
|
|
s->Freq = (Byte)freq;
|
|
if (freq > MAX_FREQ)
|
|
Rescale(p);
|
|
Ppmd7_UpdateModel(p);
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
PPMd Memory Map:
|
|
{
|
|
[ 0 ] contains subset of original raw text, that is required to create context
|
|
records, Some symbols are not written, when max order context was reached
|
|
[ Text ] free area
|
|
[ UnitsStart ] CPpmd_State vectors and CPpmd7_Context records
|
|
[ LoUnit ] free area for CPpmd_State and CPpmd7_Context items
|
|
[ HiUnit ] CPpmd7_Context records
|
|
[ Size ] end of array
|
|
}
|
|
|
|
These addresses don't cross at any time.
|
|
And the following condtions is true for addresses:
|
|
(0 <= Text < UnitsStart <= LoUnit <= HiUnit <= Size)
|
|
|
|
Raw text is BYTE--aligned.
|
|
the data in block [ UnitsStart ... Size ] contains 12-bytes aligned UNITs.
|
|
|
|
Last UNIT of array at offset (Size - 12) is root order-0 CPpmd7_Context record.
|
|
The code can free UNITs memory blocks that were allocated to store CPpmd_State vectors.
|
|
The code doesn't free UNITs allocated for CPpmd7_Context records.
|
|
|
|
The code calls RestartModel(), when there is no free memory for allocation.
|
|
And RestartModel() changes the state to orignal start state, with full free block.
|
|
|
|
|
|
The code allocates UNITs with the following order:
|
|
|
|
Allocation of 1 UNIT for Context record
|
|
- from free space (HiUnit) down to (LoUnit)
|
|
- from FreeList[0]
|
|
- AllocUnitsRare()
|
|
|
|
AllocUnits() for CPpmd_State vectors:
|
|
- from FreeList[i]
|
|
- from free space (LoUnit) up to (HiUnit)
|
|
- AllocUnitsRare()
|
|
|
|
AllocUnitsRare()
|
|
- if (GlueCount == 0)
|
|
{ Glue lists, GlueCount = 255, allocate from FreeList[i]] }
|
|
- loop for all higher sized FreeList[...] lists
|
|
- from (UnitsStart - Text), GlueCount--
|
|
- ERROR
|
|
|
|
|
|
Each Record with Context contains the CPpmd_State vector, where each
|
|
CPpmd_State contains the link to Successor.
|
|
There are 3 types of Successor:
|
|
1) NULL-Successor - NULL pointer. NULL-Successor links can be stored
|
|
only in 0-order Root Context Record.
|
|
We use 0 value as NULL-Successor
|
|
2) RAW-Successor - the link to position in raw text,
|
|
that "RAW-Successor" is being created after first
|
|
occurrence of new symbol for some existing context record.
|
|
(RAW-Successor > 0).
|
|
3) RECORD-Successor - the link to CPpmd7_Context record of (Order+1),
|
|
that record is being created when we go via RAW-Successor again.
|
|
|
|
For any successors at any time: the following condtions are true for Successor links:
|
|
(NULL-Successor < RAW-Successor < UnitsStart <= RECORD-Successor)
|
|
|
|
|
|
---------- Symbol Frequency, SummFreq and Range in Range_Coder ----------
|
|
|
|
CPpmd7_Context::SummFreq = Sum(Stats[].Freq) + Escape_Freq
|
|
|
|
The PPMd code tries to fulfill the condition:
|
|
(SummFreq <= (256 * 128 = RC::kBot))
|
|
|
|
We have (Sum(Stats[].Freq) <= 256 * 124), because of (MAX_FREQ = 124)
|
|
So (4 = 128 - 124) is average reserve for Escape_Freq for each symbol.
|
|
If (CPpmd_State::Freq) is not aligned for 4, the reserve can be 5, 6 or 7.
|
|
SummFreq and Escape_Freq can be changed in Rescale() and *Update*() functions.
|
|
Rescale() can remove symbols only from max-order contexts. So Escape_Freq can increase after multiple calls of Rescale() for
|
|
max-order context.
|
|
|
|
When the PPMd code still break (Total <= RC::Range) condition in range coder,
|
|
we have two ways to resolve that problem:
|
|
1) we can report error, if we want to keep compatibility with original PPMd code that has no fix for such cases.
|
|
2) we can reduce (Total) value to (RC::Range) by reducing (Escape_Freq) part of (Total) value.
|
|
*/
|