bool FeatureIndex::buildBigramFeature(LearnerPath *path,
const char *rfeature,
const char *lfeature) {
char rbuf[BUFSIZE];
char lbuf[BUFSIZE];
char *R[POSSIZE];
char *L[POSSIZE];
feature_.clear();
std::strncpy(lbuf, rfeature, BUFSIZE);
std::strncpy(rbuf, lfeature, BUFSIZE);
size_t lsize = tokenizeCSV(lbuf, L, POSSIZE);
size_t rsize = tokenizeCSV(rbuf, R, POSSIZE);
for (std::vector<const char*>::const_iterator it = bigram_templs_.begin();
it != bigram_templs_.end(); ++it) {
const char *p = *it;
os_.clear();
for (; *p; p++) {
switch (*p) {
default: os_ << *p; break;
case '\\': os_ << getEscapedChar(*++p); break;
case '%': {
switch (*++p) {
case 'L': {
const char *r = getIndex(const_cast<char **>(&p), L, lsize);
if (!r) goto NEXT;
os_ << r;
} break;
case 'R': {
const char *r = getIndex(const_cast<char **>(&p), R, rsize);
if (!r) goto NEXT;
os_ << r;
} break;
case 'l': os_ << lfeature; break; // use lfeature as it is
case 'r': os_ << rfeature; break;
default:
CHECK_FALSE(false) << "unkonwn meta char: " << *p;
}
}
}
}
os_ << '\0';
ADDB(os_.str());
NEXT: continue;
}
COPY_FEATURE(path->fvector);
return true;
}
/* 写成二进制格式 */
int CInstQuoteDay::DumpToBytes(BYTE **pByte)
{
int i;
st_quoteitem *pLast;
BYTE *p,*pBack;
long lRec;
long lBackWritePos;
int iSize;
pLast = nullptr;
iSize = 0;
/* 写文件头 */
p = *pByte;
iSize += DumpHeaderToBytes(&p, &lBackWritePos);
ADDB(m_iQuoteArrayCount);
for(i=0;i<m_iQuoteArrayCount;i++)
{
ADDB(m_pQuoteArray[i]);
}
lRec = 0;
for(i=0;i<m_iQuoteArrayCount;i++)
{
if (m_pQuoteArray[i].m_pQuote == nullptr)
continue;
iSize += m_pQuoteArray[i].m_pQuote->DumpToBytes(&p);
lRec++;
}
pBack = *pByte + lBackWritePos;
memcpy(pBack,&m_iQuoteArrayCount,sizeof(long));
pBack+=sizeof(long);
memcpy(pBack,&lRec,sizeof(long));
*pByte = p;
return iSize;
}
bool FeatureIndex::buildUnigramFeature(LearnerPath *path,
const char *ufeature) {
char ubuf[BUFSIZE];
char *F[POSSIZE];
feature_.clear();
std::strncpy(ubuf, ufeature, BUFSIZE);
size_t usize = tokenizeCSV(ubuf, F, POSSIZE);
for (std::vector<const char*>::const_iterator it = unigram_templs_.begin();
it != unigram_templs_.end(); ++it) {
const char *p = *it;
os_.clear();
for (; *p; p++) {
switch (*p) {
default: os_ << *p; break;
case '\\': os_ << getEscapedChar(*++p); break;
case '%': {
switch (*++p) {
case 'F': {
const char *r = getIndex(const_cast<char **>(&p), F, usize);
if (!r) goto NEXT;
os_ << r;
} break;
case 't': os_ << (size_t)path->rnode->char_type; break;
case 'u': os_ << ufeature; break;
default:
CHECK_FALSE(false) << "unkonwn meta char: " << *p;
}
}
}
}
os_ << '\0';
ADDB(os_.str());
NEXT: continue;
}
COPY_FEATURE(path->rnode->fvector);
return true;
}
int CInstQuoteDay::DumpHeaderToBytes(BYTE **pByte, long *plBackWritePos)
{
int i;
char m_sTmp[20];
long l;
BYTE *p;
int iSize;
memset(m_sTmp,0,sizeof(m_sTmp));
strcpy(m_sTmp,VERSION_STR);
p = *pByte;
iSize = 0;
#define ADDA(a,b) memcpy(p,a,b); \
p+=b; \
iSize += b;
#define ADDB(a) memcpy(p,&a,sizeof(a)); \
p+=sizeof(a); \
iSize += sizeof(a);
ADDA(m_sTmp,20);
ADDA(m_sExchangeID, 10);
ADDA(m_sInstrumentID, 40);
ADDA(m_sTradingDay, 10);
ADDA(m_sInstrumentName, 40);
ADDB(m_ctrl.m_wnSeconds);
ADDB(m_ctrl.m_wmQuotes);
ADDB(m_ctrl.m_iTimeSegmentCount);
for(i=0;i<m_ctrl.m_iTimeSegmentCount;i++)
{
ADDB(m_ctrl.m_dwSegmentBeginTime[i]);
ADDB(m_ctrl.m_dwSegmentEndTime[i]);
}
/* 然后是总记录数 / 实际记录数*/
*plBackWritePos = (long)iSize;
l = 0;
ADDB(l);
ADDB(l);
*pByte = p;
return iSize;
}
static UINT32 opINCB(v60_state *cpustate) /* TRUSTED */
{
UINT8 appb;
cpustate->modadd = cpustate->PC + 1;
cpustate->moddim = 0;
cpustate->amlength1 = ReadAMAddress(cpustate);
if (cpustate->amflag)
appb = (UINT8)cpustate->reg[cpustate->amout];
else
appb = cpustate->program->read_byte(cpustate->amout);
ADDB(appb, 1);
if (cpustate->amflag)
SETREG8(cpustate->reg[cpustate->amout], appb);
else
cpustate->program->write_byte(cpustate->amout, appb);
return cpustate->amlength1 + 1;
}
UINT32 opINCB(void) /* TRUSTED */
{
UINT8 appb;
modAdd=PC+1;
modDim=0;
amLength1=ReadAMAddress();
if (amFlag)
appb=(UINT8)v60.reg[amOut];
else
appb=MemRead8(amOut);
ADDB(appb, 1);
if (amFlag)
SETREG8(v60.reg[amOut], appb);
else
MemWrite8(amOut, appb);
return amLength1+1;
}
int main ( void )
{
int* x3 = malloc(3);
float f, *f3 = malloc(3);
int* x4 = malloc(4);
double d, *d7 = malloc(7);
int* x5 = malloc(5);
long long int lli, *lli7 = malloc(7);
int* x6 = malloc(6);
char c, *c0 = malloc(0);
int* x7 = malloc(7);
short int s, *s1 = malloc(1);
int x;
int* y4 = malloc(4);
int* y5 = malloc(5);
int* y6 = malloc(6);
int* y7 = malloc(7);
#define ADDB(ptr, n) ((int*)(((unsigned long)(ptr)) + (n)))
// All these overrun by a single byte; the reads are happening at
// different alignments.
x = * ADDB(x3,0); // ok if --partial-loads-ok=yes
x = * ADDB(x4,1);
x = * ADDB(x5,2);
x = * ADDB(x6,3);
x = * ADDB(x7,4); // ok if --partial-loads-ok=yes
// These are fine
x = * ADDB(y4,0);
x = * ADDB(y5,1);
x = * ADDB(y6,2);
x = * ADDB(y7,3);
// These are all bad, at different points along
x = * ADDB(x3,-1); // before
x = * ADDB(x3, 0); // inside // ok if --partial-loads-ok=yes ...
x = * ADDB(x3, 1); // inside // ... but only on 32 bit platforms
x = * ADDB(x3, 2); // inside // ... ditto
x = * ADDB(x3, 3); // after
// These are all bad
f = * f3; // ok if --partial-loads-ok=yes // ... ditto
d = * d7;
lli = * lli7; // ok if --partial-loads-ok=yes see XXX below
c = * c0;
s = * s1;
return 0;
}