void Program::FinalizeSegments () {
header.a_magic = ZMAGIC;
header.a_text = RoundUp(textAlloc, PAGSIZ);
header.a_data = RoundUp(dataAlloc, PAGSIZ);
header.a_bss = bssAlloc - (header.a_data - dataAlloc);
#ifdef sun
header.a_machtype = M_68020;
header.a_entry = PAGSIZ + sizeof(header);
UpdateSegment(textStart, header.a_entry, N_TEXT);
UpdateSegment(dataStart, RoundUp(textAlloc, SEGSIZ), N_DATA);
segPos[N_TEXT] = sizeof(header);
segPos[N_DATA] = header.a_text;
#endif
#ifdef vax
UpdateSegment(textStart, 0, N_TEXT);
UpdateSegment(dataStart, header.a_text, N_DATA);
segPos[N_TEXT] = PAGSIZ;
segPos[N_DATA] = segPos[N_TEXT] + header.a_text;
#endif
UpdateSegment(bssStart, dataStart + dataAlloc, N_BSS);
segPos[N_BSS] = segPos[N_DATA] + dataAlloc;
segPos[ANON_SEG] = segPos[N_BSS] + bssAlloc;
int newAnonStart = AlignTo(WORDSIZE/BYTESIZE, bssStart + bssAlloc);
if (anonStart != newAnonStart) {
anonStart = newAnonStart;
segStart[ANON_SEG] = newAnonStart;
anonList->NeedToReloc();
}
}
SortTerms ComputeSortTerms(int numSortThreads, int valuesPerThread,
bool useTransList, int numBits, int numElements, int numSMs) {
SortTerms terms;
int numValues = numSortThreads * valuesPerThread;
terms.numSortBlocks = DivUp(numElements, numValues);
terms.numCountBlocks = DivUp(terms.numSortBlocks, NumCountWarps);
terms.countValuesPerThread = numValues / WarpSize;
int numBuckets = 1<< numBits;
terms.countSize = 4 * std::max(WarpSize, numBuckets * NumCountWarps) *
terms.numCountBlocks;
int numChannels = numBuckets / 2;
int numSortBlocksPerCountWarp =
std::min(NumCountWarps, WarpSize / numChannels);
terms.numHistRows = DivUp(terms.numSortBlocks, numSortBlocksPerCountWarp);
terms.numHistBlocks = std::min(numSMs,
DivUp(terms.numHistRows, NumHistWarps));
int bucketCodeBlockSize = numBuckets;
if(useTransList) bucketCodeBlockSize += numBuckets + WarpSize;
terms.scatterStructSize = RoundUp(bucketCodeBlockSize, WarpSize);
// hist3 kernel (for 1 - 5 radix bits) may write two blocks of codes at a
// time, even if only one block is required. To support this, round the
// number of sort blocks up to a multiple of 2.
terms.bucketCodesSize = 4 * RoundUp(terms.numSortBlocks, 2) *
terms.scatterStructSize;
terms.numEndKeys = RoundUp(numElements, numValues) - numElements;
return terms;
}
void
GCInit(void)
{
init_int(&minOffRequest, 1 * TILT_PAGESIZE);
init_int(&maxOffRequest, 8 * TILT_PAGESIZE);
init_int(&minOnRequest, 1 * TILT_PAGESIZE);
init_int(©PageSize, TILT_PAGESIZE / 2);
init_int(©CheckSize, TILT_PAGESIZE / 2);
init_int(©ChunkSize, 256);
minOffRequest = RoundUp(minOffRequest, TILT_PAGESIZE);
minOnRequest = RoundUp(minOnRequest, TILT_PAGESIZE);
reset_statistic(&minorSurvivalStatistic);
reset_statistic(&heapSizeStatistic);
reset_statistic(&majorSurvivalStatistic);
switch (collector_type) {
case Semispace:
GCFun = GC_Semi;
GCReleaseFun = GCRelease_Semi;
GCPollFun = NULL;
GCInit_Semi();
break;
case Generational:
GCFun = GC_Gen;
GCReleaseFun = GCRelease_Gen;
GCPollFun = NULL;
GCInit_Gen();
break;
case SemispaceParallel:
GCFun = GC_SemiPara;
GCReleaseFun = GCRelease_SemiPara;
GCPollFun = GCPoll_SemiPara;
GCInit_SemiPara();
break;
case GenerationalParallel:
GCFun = GC_GenPara;
GCReleaseFun = GCRelease_GenPara;
GCPollFun = GCPoll_GenPara;
GCInit_GenPara();
break;
case SemispaceConcurrent:
GCFun = GC_SemiConc;
GCReleaseFun = GCRelease_SemiConc;
GCPollFun = GCPoll_SemiConc;
GCInit_SemiConc();
break;
case GenerationalConcurrent:
GCFun = GC_GenConc;
GCReleaseFun = GCRelease_GenConc;
GCPollFun = GCPoll_GenConc;
GCInit_GenConc();
break;
default:
DIE("bad collector type");
}
if (forceMirrorArray)
mirrorArray = 1;
}
BOOL SaveLibraryToFile(LPVOID lpBase, TCHAR* lpOutFileName)
{
DWORD numberOfBytesWritten;
HANDLE hOutFile = CreateFile( lpOutFileName, GENERIC_WRITE, 0, NULL, CREATE_ALWAYS, 0, NULL);
if(hOutFile == INVALID_HANDLE_VALUE)
return FALSE;
// DOS header
PIMAGE_DOS_HEADER dos_header;
dos_header = (PIMAGE_DOS_HEADER) lpBase;
if (dos_header->e_magic != IMAGE_DOS_SIGNATURE || dos_header->e_lfanew == 0)
{
CloseHandle(hOutFile);
return FALSE;
}
if( !WriteFile( hOutFile, lpBase, dos_header->e_lfanew, &numberOfBytesWritten, NULL))
return FALSE;
// PE header
PIMAGE_NT_HEADERS32 pe_header;
pe_header = (PIMAGE_NT_HEADERS32)((DWORD) lpBase + (DWORD)dos_header->e_lfanew );
if (pe_header->Signature != IMAGE_NT_SIGNATURE)
{
CloseHandle(hOutFile);
return FALSE;
}
DWORD dwSizeHeader = pe_header->FileHeader.SizeOfOptionalHeader +
sizeof(pe_header->FileHeader) + 4;// +
//sizeof(IMAGE_SECTION_HEADER)*pe_header->FileHeader.NumberOfSections;
PIMAGE_SECTION_HEADER sections = (PIMAGE_SECTION_HEADER)((DWORD)(lpBase) + dwSizeHeader + dos_header->e_lfanew);
if (sections[0].PointerToRawData == 0)
WriteFile( hOutFile, pe_header, 0x400 - dos_header->e_lfanew, &numberOfBytesWritten,NULL);
else
WriteFile( hOutFile, pe_header, sections[0].PointerToRawData - dos_header->e_lfanew, &numberOfBytesWritten,NULL);
// write sections...
for(unsigned short i = 0; i < pe_header->FileHeader.NumberOfSections; i++)
{
LPVOID addr = (LPVOID)((DWORD)lpBase + sections[i].VirtualAddress);
// experiment for ".text"
if (strcmp((char *) sections[i].Name, ".text") == 0)
{ //
WriteFile( hOutFile, addr, RoundUp(sections[i].Misc.VirtualSize, pe_header->OptionalHeader.FileAlignment), &numberOfBytesWritten, NULL);
}
else
WriteFile( hOutFile, addr, RoundUp(sections[i].SizeOfRawData, pe_header->OptionalHeader.FileAlignment), &numberOfBytesWritten, NULL);
}
CloseHandle(hOutFile);
return TRUE;
}
void IntelH264Decoder::CreateWorkSurface(mfxFrameSurface1& surf)
{
// Setup the "work surface". I don't understand why doesn't do this itself (malloc control is all I can assume).
memset(&surf, 0, sizeof(surf));
surf.Info = VideoParam.mfx.FrameInfo;
surf.Info.FourCC = MFX_FOURCC_NV12; // IMSDK docs say NV12 is "Native Format" for IMSDK
surf.Data.PitchLow = RoundUp(VideoParam.mfx.FrameInfo.Width, 64);
surf.Data.Y = (mfxU8 *) AlignedAlloc(surf.Data.PitchLow * RoundUp(VideoParam.mfx.FrameInfo.Height, 64) * 3 / 2, 16);
surf.Data.UV = surf.Data.Y + (surf.Data.PitchLow * RoundUp(VideoParam.mfx.FrameInfo.Height, 64));
}
sortStatus_t SORTAPI sortAllocData(sortEngine_t engine, sortData_t data) {
if(data->valueCount > 6) return SORT_STATUS_INVALID_VALUE;
DeviceMemPtr mem[7 * 2];
uint maxElements = RoundUp(data->maxElements, 2048);
int count = (-1 == data->valueCount) ? 1 : data->valueCount;
CUresult result = CUDA_SUCCESS;
CUdeviceptr* memPtr = data->keys;
for(int i = 0; (i < 2 + 2 * count) && (CUDA_SUCCESS == result); ++i) {
if(!memPtr[i]) {
result = engine->context->MemAlloc<uint>(maxElements, &mem[i]);
if(CUDA_SUCCESS != result) return SORT_STATUS_DEVICE_ALLOC_FAILED;
}
}
if(CUDA_SUCCESS != result) return SORT_STATUS_DEVICE_ALLOC_FAILED;
for(int i(0); i < 7 * 2; ++i)
if(mem[i]) {
memPtr[i] = mem[i]->Handle();
mem[i].release();
}
data->parity = 0;
return SORT_STATUS_SUCCESS;
}
sortStatus_t SORTAPI sortCreateData(sortEngine_t engine, int maxElements,
int valueCount, sortData_t* data) {
if(valueCount > 6) return SORT_STATUS_INVALID_VALUE;
std::auto_ptr<sortData_d> d(new sortData_d);
// sortData_d
d->maxElements = RoundUp(maxElements, 2048);
d->numElements = maxElements;
d->valueCount = valueCount;
d->firstBit = 0;
d->endBit = 32;
d->preserveEndKeys = false;
d->earlyExit = false;
d->keys[0] = d->keys[1] = 0;
d->values1[0] = d->values1[1] = 0;
d->values2[0] = d->values2[1] = 0;
d->values3[0] = d->values3[1] = 0;
d->values4[0] = d->values4[1] = 0;
d->values5[0] = d->values5[1] = 0;
d->values6[0] = d->values6[1] = 0;
sortAllocData(engine, d.get());
*data = d.release();
return SORT_STATUS_SUCCESS;
}
char* AllocateBufferSpace(const DWORD bufSize, const DWORD bufCount, DWORD& totalBufferSize, DWORD& totalBufferCount)
{
SYSTEM_INFO systemInfo;
::GetSystemInfo(&systemInfo);
const unsigned __int64 granularity = systemInfo.dwAllocationGranularity;
const unsigned __int64 desiredSize = bufSize * bufCount;
unsigned __int64 actualSize = RoundUp(desiredSize, granularity);
if (actualSize > UINT_MAX )
{
actualSize = (UINT_MAX / granularity) * granularity;
}
totalBufferCount = min(bufCount, static_cast<DWORD>(actualSize / bufSize));
totalBufferSize = static_cast<DWORD>(actualSize);
char* pBuffer = reinterpret_cast<char*>(VirtualAllocEx(GetCurrentProcess(), 0, totalBufferSize, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE));
if (pBuffer == 0)
{
printf_s("VirtualAllocEx Error: %d\n", GetLastError());
exit(0);
}
return pBuffer;
}
size_t DSSMReader<ElemType>::RecordsToRead(size_t mbStartSample, bool tail)
{
assert(mbStartSample >= m_epochStartSample);
// determine how far ahead we need to read
bool randomize = Randomize();
// need to read to the end of the next minibatch
size_t epochSample = mbStartSample;
epochSample %= m_epochSize;
// determine number left to read for this epoch
size_t numberToEpoch = m_epochSize - epochSample;
// we will take either a minibatch or the number left in the epoch
size_t numberToRead = min(numberToEpoch, m_mbSize);
if (numberToRead == 0 && !tail)
numberToRead = m_mbSize;
if (randomize)
{
size_t randomizeSweep = RandomizeSweep(mbStartSample);
// if first read or read takes us to another randomization range
// we need to read at least randomization range records
if (randomizeSweep != m_randomordering.CurrentSeed()) // the range has changed since last time
{
numberToRead = RoundUp(epochSample, m_randomizeRange) - epochSample;
if (numberToRead == 0 && !tail)
numberToRead = m_randomizeRange;
}
}
return numberToRead;
}
void
Heap_Resize(Heap_t* h, long newSize, int reset)
{
long usedSize;
long maxSize = (h->mappedTop - h->bottom) * sizeof(val_t);
long oldWriteableSize = (h->writeableTop - h->bottom) * sizeof(val_t);
long newSizeRound = RoundUp(newSize, TILT_PAGESIZE);
if (newSize > maxSize) DIE("resized heap too big");
if (reset) {
h->cursor = h->bottom;
}
usedSize = (h->cursor - h->bottom) * sizeof(val_t);
assert(usedSize <= newSize);
h->top = h->bottom + (newSize / sizeof(val_t));
if (newSizeRound > oldWriteableSize) {
my_mprotect(6,(caddr_t) h->writeableTop,
newSizeRound - oldWriteableSize,
PROT_READ | PROT_WRITE);
h->writeableTop = h->bottom + newSizeRound / sizeof(val_t);
}
else if (paranoid && newSizeRound < oldWriteableSize) {
my_mprotect(7,(caddr_t) (h->bottom +
newSizeRound / sizeof(val_t)),
oldWriteableSize - newSizeRound, PROT_NONE);
h->writeableTop = h->bottom + newSizeRound / sizeof(val_t);
}
assert(h->bottom <= h->cursor);
assert(h->cursor <= h->top);
assert(h->top <= h->writeableTop);
assert(h->writeableTop <= h->mappedTop);
}
Heap_t*
Heap_Alloc(int MinSize, int MaxSize)
{
static int heap_count = 0;
Heap_t *res = &(Heaps[heap_count++]);
int maxsize_pageround = RoundUp(MaxSize,TILT_PAGESIZE);
res->size = maxsize_pageround;
res->bottom = (mem_t) my_mmap(maxsize_pageround,
PROT_READ | PROT_WRITE);
res->cursor = res->bottom;
res->top = res->bottom + MinSize / (sizeof (val_t));
res->writeableTop = res->bottom +
maxsize_pageround / (sizeof (val_t));
res->mappedTop = res->writeableTop;
SetRange(&(res->range), res->bottom, res->mappedTop);
res->valid = 1;
res->bitmap = paranoid ? CreateBitmap(maxsize_pageround / 4) : NULL;
res->freshPages = (int *)emalloc(
DivideUp(maxsize_pageround / TILT_PAGESIZE, 32) * sizeof(int));
memset((int *)res->freshPages, 0,
DivideUp(maxsize_pageround / TILT_PAGESIZE, 32) * sizeof(int));
assert(res->bottom != (mem_t) -1);
assert(heap_count < NumHeap);
assert(MaxSize >= MinSize);
/*
Try to lock down pages and force page-table to be initialized;
otherwise, PadHeapArea can often take 0.1 - 0.2 ms. Even with
this, there are occasional (but far fewer) page table misses.
*/
if (geteuid() == 0)
(void)mlock((caddr_t) res->bottom, maxsize_pageround);
return res;
}
static void _UpdateRange(_Inout_ OVS_PI_RANGE* pPiRange, SIZE_T offset, SIZE_T size)
{
SIZE_T startPos = 0;
SIZE_T endPos = 0;
OVS_CHECK(pPiRange);
startPos = RoundDown(offset, sizeof(UINT64));
endPos = RoundUp(offset + size, sizeof(UINT64));
if (!pPiRange)
{
return;
}
//i.e. in the beginning
if (pPiRange->startRange == pPiRange->endRange)
{
pPiRange->startRange = startPos;
pPiRange->endRange = endPos;
}
else
{
//i.e. if it was set before
if (pPiRange->startRange > startPos)
{
pPiRange->startRange = startPos;
}
if (pPiRange->endRange < endPos)
{
pPiRange->endRange = endPos;
}
}
}
/* internal routines */
static void
init_dispinfo(struct dispinfo *d)
{
int total_squares;
ulong h;
d->win_width = WIN_WIDTH;
d->win_height = WIN_HEIGHT;
d->pixel_xmult = PIX_SIZE;
d->pixel_ymult = PIX_SIZE;
total_squares = WinSize(d) / PixelSize(d);
d->bytes_vp = d->alloc_unit;
/* compute how many bytes of memory one square on the display should
* represent so you have the smallest granularity without the window size
* exceeding the initial height.
*/
while (total_squares < DivideRoundedUp(d->arena_size, d->bytes_vp))
d->bytes_vp *= 2;
h = DivideRoundedUp(d->arena_size, d->bytes_vp) * PixelSize(d);
d->win_height = DivideRoundedUp(h, d->win_width);
d->win_height = RoundUp(d->win_height, d->pixel_ymult);
sprintf(d->title + strlen(d->title), " (1 square = %d bytes)",
d->bytes_vp);
}
bool iupPlotTickIterLog::AdjustRange (double &ioMin, double &ioMax) const
{
double theBase = mAxis->mLogBase;
if (mAxis->mMaxDecades > 0)
ioMin = ioMax/iupPlotExp (mAxis->mMaxDecades, theBase);
if (ioMin == 0 && ioMax == 0)
{
ioMin = kLogMinClipValue;
ioMax = 1.0;
}
if (ioMin <= 0 || ioMax<=0)
return false;
ioMin = RoundDown(ioMin*kLittleIncrease);
ioMax = RoundUp(ioMax*kLittleDecrease);
if (ioMin<kLogMinClipValue)
ioMin = kLogMinClipValue;
if (mAxis->mMaxDecades > 0)
ioMin = ioMax/iupPlotExp (mAxis->mMaxDecades, theBase);
return true;
}
int Round(double x, Rounding rounding) {
switch (rounding) {
case ROUND_UP : return RoundUp (x);
case ROUND_DOWN: return RoundDown (x);
default : return RoundNearest(x);
}
}
/* compute (xUL,yUL) and nrRows, nrCols from some coordinates
* RcomputeExtend computes parameters to create a raster maps
* from minimum and maximum x and y coordinates, projection information,
* cellsize and units. The resulting parameters are computed that the
* smallest raster map can be created that will include the two
* coordinates given, assuming a default angle of 0.
* Which coordinates are the maximum or minimum are
* determined by the function itself.
*/
void RcomputeExtend(
REAL8 *xUL, /* write-only, resulting xUL */
REAL8 *yUL, /* write-only, resulting yUL */
size_t *nrRows, /* write-only, resulting nrRows */
size_t *nrCols, /* write-only, resulting nrCols */
double x_1, /* first x-coordinate */
double y_1, /* first y-coordinate */
double x_2, /* second x-coordinate */
double y_2, /* second y-coordinate */
CSF_PT projection, /* required projection */
REAL8 cellSize, /* required cellsize, > 0 */
double rounding) /* assure that (xUL/rounding), (yUL/rouding)
* (xLL/rounding) and (yLL/rounding) will
* will all be an integers values > 0
*/
{
/*
* xUL ______
| |
| |
| |
------
*/
double yLL,xUR = x_1 > x_2 ? x_1 : x_2;
*xUL = x_1 < x_2 ? x_1 : x_2;
*xUL = RoundDown(*xUL, rounding); /* Round down */
xUR = RoundUp( xUR, rounding); /* Round up */
POSTCOND(*xUL <= xUR);
*nrCols = (size_t)ceil((xUR - *xUL)/cellSize);
if (projection == PT_YINCT2B)
{
yLL = y_1 > y_2 ? y_1 : y_2; /* highest value at bottom */
*yUL = y_1 < y_2 ? y_1 : y_2; /* lowest value at top */
*yUL = RoundDown(*yUL, rounding);
yLL = RoundUp( yLL, rounding);
}
else
{
yLL = y_1 < y_2 ? y_1 : y_2; /* lowest value at bottom */
*yUL = y_1 > y_2 ? y_1 : y_2; /* highest value at top */
*yUL = RoundUp( *yUL, rounding);
yLL = RoundDown( yLL, rounding);
}
*nrRows = (size_t)ceil(fabs(yLL - *yUL)/cellSize);
}
// [A]
void* Thread::Alloc(size_t size) {
ASSERT(size > 0);
auto const alloc_size = RoundUp(size, Arch::Align_Object);
for (;;) {
if (auto const pv = object_area_->Alloc(alloc_size)) {
return pv;
}
object_area_ = Mm::GetDataArea(Mm::Area::ScanType_Record, alloc_size);
}
}
/*
Interface Routines
*/
ptr_t
alloc_bigintarray(int elemLen, int initVal, int ptag)
{
ArraySpec_t spec;
spec.type = IntField;
spec.elemLen = elemLen;
spec.byteLen = RoundUp(elemLen, 4); /* excluding tag */
spec.intVal = initVal;
return alloc_bigdispatcharray(&spec);
}
static void
resize_display(struct dispinfo *d, int new_width, int new_height)
{
int total_squares;
uint h;
char *cp;
/* round down to even multiple of square size */
new_width = RoundDown(new_width, d->pixel_xmult);
new_height = RoundDown(new_height, d->pixel_ymult);
if (new_width <= 0 || new_height <= 0)
return;
total_squares = (new_width * new_height) / PixelSize(d);
d->bytes_vp = d->alloc_unit;
/* compute how many bytes of memory one square on the display should
* represent, so you have the smallest granularity without the window size
* exceeding the initial height.
*/
while (total_squares < DivideRoundedUp(d->arena_size, d->bytes_vp))
d->bytes_vp *= 2;
h = DivideRoundedUp(d->arena_size, d->bytes_vp) * PixelSize(d);
new_height = DivideRoundedUp(h, new_width);
new_height = RoundUp(new_height, d->pixel_ymult);
cp = strchr(d->title, '=');
if (cp)
sprintf(cp, "= %d bytes)", d->bytes_vp);
if (!(d->memory_pixel = DXReAllocate((Pointer)d->memory_pixel,
new_width * new_height))) {
xerror = 1;
return;
}
d->win_width = new_width;
d->win_height = new_height;
memset (d->memory_pixel, color_free, d->win_width * d->win_height);
/* don't let X free our pixel buffer while destroying old image */
d->memory_image->data = NULL;
XDestroyImage(d->memory_image);
d->memory_image = XCreateImage(d->disp,
XDefaultVisual(d->disp, XDefaultScreen(d->disp)),
8, ZPixmap, 0, (char *) d->memory_pixel,
d->win_width, d->win_height, 8, 0);
XResizeWindow(d->disp, d->wind, d->win_width, d->win_height);
XStoreName(d->disp, d->wind, d->title);
}
int
reducedToExpanded(int size, double rate, int phases)
{
if (phases == 0)
return size;
size = RoundUp(size, minOnRequest);
size = size + (2 * NumProc) * minOnRequest;
size = (int) size / (1.0 - computeReserve(rate, phases));
size = RoundDown(size, minOnRequest);
return size;
}
/*
Compute the new (reduced) heap size given the liveness ratio and
amount of live data.
*/
long
ComputeHeapSize(long live, double curRatio, double rate, int phases)
{
double rawWhere = (live - (1024 * MinRatioSize)) /
(1024.0 * (MaxRatioSize - MinRatioSize));
double where = (rawWhere > 1.0)
? 1.0
: ((rawWhere < 0.0) ? 0.0 : rawWhere);
double newratio = MinRatio + where * (MaxRatio - MinRatio);
long newReducedSize = RoundUp(live / newratio, 1024);
long newExpandedSize = reducedToExpanded(newReducedSize, rate, phases);
/*
maxReducedSize and minReducedSize are not reduced if relaxed
is true or if phases is already zero.
*/
long maxReducedSize = expandedToReduced(MaxHeapByte, rate,
relaxed ? 0 : phases);
long minReducedSize = expandedToReduced(MinHeapByte, rate,
relaxed ? 0 : phases);
if (live > maxReducedSize) {
fprintf(stderr,"GC error: Amount of live data (%ld) exceeds maxiumum heap size (%ld)\n",
live, maxReducedSize);
DIE("out of memory");
}
if (newExpandedSize > MaxHeapByte) {
double constrainedRatio = ((double)live) / maxReducedSize;
if (collectDiag >= 1 || constrainedRatio > 0.95)
fprintf(stderr,"GC warning: There is %ld kb of live data. The desired new heap size is %ld kb but is downwardly constrained to %ld kb.\n",
live / 1024, newReducedSize / 1024,
maxReducedSize / 1024);
if (constrainedRatio >= 1.00)
fprintf(stderr,"GC warning: New liveness ratio is too high %lf.\n",
constrainedRatio);
else if (constrainedRatio > 0.90)
fprintf(stderr,"GC warning: New liveness ratio is dangerously high %lf.\n",
constrainedRatio);
newReducedSize = maxReducedSize;
newExpandedSize = MaxHeapByte;
}
if (newExpandedSize < MinHeapByte) {
if (collectDiag >= 1)
fprintf(stderr,"GC warning: There is %ld kb of live data. The desired new heap size is %ld kb but is upwardly constrained to %ld kb.\n",
live / 1024, newReducedSize / 1024,
minReducedSize / 1024);
newReducedSize = minReducedSize;
newExpandedSize = MinHeapByte;
}
assert(newExpandedSize >= MinHeapByte);
assert(newExpandedSize <= MaxHeapByte);
assert(newReducedSize >= live);
return newReducedSize;
}
double initGrid(double low, double step, int logFlag)
{
double ratio, x;
double RoundUp(), stepGrid();
gridNJuke = gridCurJuke = 0;
gridJuke[gridNJuke++] = 0.0;
if (logFlag) {
ratio = pow(10.0, step);
gridBase = floor(low);
gridStep = ceil(step);
if (ratio <= 3.0) {
if (ratio > 2.0) {
ADD_GRID(3.0);
} else if (ratio > 1.333) {
ADD_GRID(2.0); ADD_GRID(5.0);
} else if (ratio > 1.25) {
ADD_GRID(1.5); ADD_GRID(2.0); ADD_GRID(3.0);
ADD_GRID(5.0); ADD_GRID(7.0);
} else {
for (x = 1.0; x < 10.0 && (x+.5)/(x+.4) >= ratio; x += .5) {
ADD_GRID(x + .1); ADD_GRID(x + .2);
ADD_GRID(x + .3); ADD_GRID(x + .4);
ADD_GRID(x + .5);
}
if (floor(x) != x) ADD_GRID(x += .5);
for ( ; x < 10.0 && (x+1.0)/(x+.5) >= ratio; x += 1.0) {
ADD_GRID(x + .5); ADD_GRID(x + 1.0);
}
for ( ; x < 10.0 && (x+1.0)/x >= ratio; x += 1.0) {
ADD_GRID(x + 1.0);
}
if (x == 7.0) {
gridNJuke--;
x = 6.0;
}
if (x < 7.0) {
ADD_GRID(x + 2.0);
}
if (x == 10.0) gridNJuke--;
}
x = low - gridBase;
for (gridCurJuke = -1; x >= gridJuke[gridCurJuke+1]; gridCurJuke++){
}
}
} else {
gridStep = RoundUp(step);
gridBase = floor(low / gridStep) * gridStep;
}
return(stepGrid());
}
/* bytesToAlloc does not include alignment */
mem_t
AllocFromHeap(Heap_t* heap, Thread_t* thread, int bytesToAlloc, Align_t align)
{
mem_t start, cursor, limit;
int padBytes = bytesToAlloc + ((align == NoWordAlign) ? 0 : 4);
int pagePadBytes = RoundUp(padBytes, TILT_PAGESIZE);
GetHeapArea(fromSpace, pagePadBytes, &start, &cursor, &limit);
if (start == NULL)
return NULL;
cursor = AlignMemoryPointer(cursor, align);
PadHeapArea(cursor + bytesToAlloc / sizeof(val_t), limit);
return cursor;
}
void NPC::CreatePath() {
mNPCData.mPathIndex = 0;
int x = (RoundUp((int) mNPCData.mPosition.x, 32) / 32);
int y = (RoundUp((int) mNPCData.mPosition.y, 32) / 32);
int startIndex = x + (y * mMapWidth);
WalkableFunctor walkF = WalkableFunctor();
ICostFunctor costF = ICostFunctor();
IHeuristicFunctor heuristicF = IHeuristicFunctor();
mGraph->SearchAStar(startIndex, mNPCData.mPathEnd, walkF, costF,
heuristicF);
NodeList nodeList = mGraph->GetPath();
mPath.clear();
for (NodeList::iterator iter = nodeList.begin(); iter != nodeList.end();
++iter) {
mPath.push_back((*iter)->GetPosition());
}
mReachedDestination = false;
}
/*!
* Create an instance of class <T>.
*/
static T * Create()
{
// We use placement new() for two reasons:
// - To create a never-destructed instance of <T>. This allows using this instance
// at any time, even during the module's destruction.
// - We could use the regular "new" operator here instead, but placement new is advantageous
// because some clients limit the amount of memory that can be dynamically allocated at
// static initialization time (e.g. clients that replace the "malloc" implementation).
// Allocating the data statically like this for a singleton class has no real disadvantage.
static FUND::UINT8 storage[sizeof(T) + FUND_ALIGNMENT_OF(T)];
return new((void *)RoundUp(&(storage[0]), FUND_ALIGNMENT_OF(T))) T();
}
/**
Write data to the media through the directory cache.
@param aPos linear media position to start writing with
@param aDes data to write
*/
void CMediaWTCache::WriteL(TInt64 aPos,const TDesC8& aDes)
{
#ifdef _DEBUG
if(iCacheDisabled)
{//-- cache is disabled for debug purposes
User::LeaveIfError(iDrive.WriteCritical(aPos,aDes));
return;
}
#endif //_DEBUG
TUint32 dataLen = aDes.Size();
const TUint8* pData = aDes.Ptr();
const TUint32 PageSz = PageSize(); //-- cache page size
//-- find out if aPos is in cache. If not, find a spare page and read data there
TInt nPage = FindOrGrabReadPageL(aPos);
CWTCachePage* pPage = iPages[nPage];
const TUint32 bytesToPageEnd = (TUint32)(pPage->iStartPos+PageSize() - aPos); //-- number of bytes from aPos to the end of the page
// __PRINT5(_L("CMediaWTCache::WriteL: aPos=%lx, aLength=%x, page:%lx, pageSz:%x, bytesToPageEnd=%x"), aPos, dataLen, pPage->iStartPos, PageSz, bytesToPageEnd);
if(dataLen <= bytesToPageEnd)
{//-- data section completely fits to the cache page
Mem::Copy(pPage->PtrInCachePage(aPos), pData, dataLen); //-- update cache
//-- make small write a multiple of a write granularity size (if it is used at all)
//-- this is not the best way to use write granularity, but we would need to refactor cache pages code to make it normal
TPtrC8 desBlock(aDes);
if(iWrGranularityLog2)
{//-- write granularity is used
const TInt64 newPos = (aPos >> iWrGranularityLog2) << iWrGranularityLog2; //-- round position down to the write granularity size
TUint32 newLen = (TUint32)(aPos - newPos)+dataLen; //-- round block size up to the write granularity size
newLen = RoundUp(newLen, iWrGranularityLog2);
const TUint8* pd = pPage->PtrInCachePage(newPos);
desBlock.Set(pd, newLen);
aPos = newPos;
}
//-- write data to the media
const TInt nErr = iDrive.WriteCritical(aPos, desBlock);
if(nErr != KErrNone)
{//-- some serious problem occured during writing, invalidate cache.
InvalidateCache();
User::Leave(nErr);
}
}
void* OS::Allocate(const size_t requested,
size_t* allocated,
bool executable) {
const size_t msize = RoundUp(requested, getpagesize());
int prot = PROT_READ | PROT_WRITE | (executable ? PROT_EXEC : 0);
void* mbase = mmap(NULL, msize, prot, MAP_PRIVATE | MAP_ANON, -1, 0);
if (mbase == MAP_FAILED) {
LOG(StringEvent("OS::Allocate", "mmap failed"));
return NULL;
}
*allocated = msize;
UpdateAllocatedSpaceLimits(mbase, msize);
return mbase;
}
void GLIX::AssignOffsets()
{
int Offset = 0;
for (int h = 0; h < m_HashTableSize; ++h)
{
for (SEQDATA *i = m_HashTable[h]; i; i = i->Next)
{
i->Offset = Offset;
Offset += i->Length;
const int NewOffset = RoundUp(Offset, m_Pad, SEQMAP_BLOCKSIZE);
i->RoundedLength = i->Length + NewOffset - Offset;
Offset = NewOffset;
}
}
m_GlobalLength = Offset;
}
void LDesktopPluginSpace::addDesktopPlugin(QString plugID){
//This is used for generic plugins (QWidget-based)
if(DEBUG){ qDebug() << "Adding Desktop Plugin:" << plugID; }
LDPlugin *plug = NewDP::createPlugin(plugID, this);
if(plug==0){ return; } //invalid plugin
//plug->setAttribute(Qt::WA_TranslucentBackground);
plug->setWhatsThis(plugID);
//Now get the saved geometry for the plugin
QRect geom = plug->gridGeometry(); //grid coordinates
if(geom.isNull()){
//Try the old format (might be slight drift between sessions if the grid size changes)
geom = plug->loadPluginGeometry(); //in pixel coords
if(!geom.isNull()){ geom = geomToGrid(geom); } //convert to grid coordinates
}
if(DEBUG){ qDebug() << "Saved plugin geom:" << geom << plugID; }
//Now determine the position to put it
if(geom.isNull()){
//No previous location - need to calculate initial geom
QSize sz = plug->defaultPluginSize(); //in grid coordinates
geom.setSize(sz);
//if an applauncher - add from top-left, otherwise add in from bottom-right
if(plugID.startsWith("applauncher")){ geom = findOpenSpot(geom.width(), geom.height() ); }
else{ geom = findOpenSpot(geom.width(), geom.height(), RoundUp(this->height()/GRIDSIZE), RoundUp(this->width()/GRIDSIZE), true); }
}else if(!ValidGeometry(plugID, gridToGeom(geom)) ){
//Find a new location for the plugin (saved location is invalid)
geom = findOpenSpot(geom.width(), geom.height(), geom.y(), geom.x(), false); //try to get it within the same general area first
}
if(geom.x() < 0 || geom.y() < 0){
qDebug() << "No available space for desktop plugin:" << plugID << " - IGNORING";
delete plug;
}else{
if(DEBUG){ qDebug() << " - New Plugin Geometry (grid):" << geom; }
//Now place the item in the proper spot/size
plug->setGridGeometry(geom); //save for later
MovePlugin(plug, gridToGeom(geom));
//plug->setGeometry( gridToGeom(geom) );
plug->show();
if(DEBUG){ qDebug() << " - New Plugin Geometry (px):" << plug->geometry(); }
ITEMS << plug;
connect(plug, SIGNAL(StartMoving(QString)), this, SLOT(StartItemMove(QString)) );
connect(plug, SIGNAL(StartResizing(QString)), this, SLOT(StartItemResize(QString)) );
connect(plug, SIGNAL(RemovePlugin(QString)), this, SLOT(RemoveItem(QString)) );
connect(plug, SIGNAL(IncreaseIconSize()), this, SIGNAL(IncreaseIcons()) );
connect(plug, SIGNAL(DecreaseIconSize()), this, SIGNAL(DecreaseIcons()) );
connect(plug, SIGNAL(CloseDesktopMenu()), this, SIGNAL(HideDesktopMenu()) );
}
}
double MgpuBenchmark(searchEngine_t engine, int count, CuDeviceMem* values,
searchType_t type, CuDeviceMem* btree, int numIterations, int numQueries,
CuDeviceMem* keys, CuDeviceMem* indices, const T* valuesHost,
const T* keysHost) {
CuEventTimer timer;
timer.Start();
int size = (SEARCH_TYPE_INT32 == type) ? 4 : 8;
int offset = 0;
for(int it(0); it < numIterations; ++it) {
offset += RoundUp(numQueries, 32);
if(offset + numQueries > MaxQuerySize) offset = 0;
searchStatus_t status = searchKeys(engine, count, type,
values->Handle(), SEARCH_ALGO_LOWER_BOUND,
keys->Handle() + offset * size, numQueries, btree->Handle(),
indices->Handle());
if(SEARCH_STATUS_SUCCESS != status) {
printf("FAIL!\n");
exit(0);
}
}
double elapsed = timer.Stop();
double throughput = (double)numQueries * numIterations / elapsed;
// Verify the results for the last set of queries run.
std::vector<uint> results(numQueries);
indices->ToHost(results);
for(int i(0); i < numQueries; ++i) {
const T* lower = std::lower_bound(valuesHost, valuesHost + count,
keysHost[offset + i]);
if((lower - valuesHost) != results[i]) {
printf("Failure in MGPU Search.\n");
exit(0);
}
}
return throughput;
}