void Mouse::setLuggage(uint32 resId, uint32 rate) {
_currentLuggageId = resId;
_frame = 0;
_activeFrame = -1;
createPointer(_currentPtrId, resId);
}
Domain *
Constructors::createGetElementPtr(const llvm::ConstantExpr &value,
const std::vector<const Domain*> &operands,
const llvm::Value &place) const
{
std::vector<Domain*> offsets;
std::vector<const Domain*>::const_iterator it = operands.begin() + 1,
itend = operands.end();
for (; it != itend; ++it)
{
unsigned bitWidth = Integer::Utils::getBitWidth(**it);
CANAL_ASSERT_MSG(bitWidth <= 64,
"Cannot handle GetElementPtr offset"
" with more than 64 bits.");
if (bitWidth < 64)
{
Domain *offset = createInteger(64);
offset->zext(**it);
offsets.push_back(offset);
}
else
offsets.push_back((*it)->clone());
}
const llvm::PointerType &pointerType =
checkedCast<const llvm::PointerType>(*value.getType());
// GetElementPtr on a Pointer
const Pointer::Pointer *pointer = dynCast<Pointer::Pointer>(*operands.begin());
if (pointer)
{
return pointer->getElementPtr(offsets,
pointerType,
*this);
}
else
{
// GetElementPtr on anything except a pointer. For example, it is
// called on arrays and structures.
Domain *result = createPointer(pointerType);
Pointer::Utils::addTarget(*result,
Pointer::Target::Block,
&place,
*value.op_begin(),
offsets,
NULL);
return result;
}
}
void Mouse::setPointer(uint32 resId, uint32 rate) {
_currentPtrId = resId;
_frame = 0;
createPointer(resId, _currentLuggageId);
if ((resId == 0) || (!(Logic::_scriptVars[MOUSE_STATUS] & 1) && (!_mouseOverride))) {
CursorMan.showMouse(false);
} else {
animate();
CursorMan.showMouse(true);
}
}
Domain *
Constructors::create(const llvm::Type &type) const
{
CANAL_ASSERT_MSG(!type.isVoidTy(), "Cannot create value of type Void.");
if (type.isIntegerTy())
{
llvm::IntegerType &integerType = checkedCast<llvm::IntegerType>(type);
return createInteger(integerType.getBitWidth());
}
if (type.isFloatingPointTy())
{
const llvm::fltSemantics &semantics =
Float::Utils::getSemantics(type);
return createFloat(semantics);
}
if (type.isPointerTy())
{
const llvm::PointerType &pointerType =
checkedCast<llvm::PointerType>(type);
return createPointer(pointerType);
}
if (type.isArrayTy() || type.isVectorTy())
{
const llvm::SequentialType &stype =
checkedCast<llvm::SequentialType>(type);
return createArray(stype);
}
if (type.isStructTy())
{
const llvm::StructType &structType =
checkedCast<llvm::StructType>(type);
std::vector<Domain*> members;
for (unsigned i = 0; i < structType.getNumElements(); i ++)
members.push_back(create(*structType.getElementType(i)));
return createStructure(structType, members);
}
CANAL_DIE_MSG("Unsupported llvm::Type::TypeID: " << type.getTypeID());
}
void Mouse::initialize() {
_numObjs = 0;
Logic::_scriptVars[MOUSE_STATUS] = 0; // mouse off and unlocked
_getOff = 0;
_inTopMenu = false;
_lastState = 0;
_mouseOverride = false;
_currentPtrId = _currentLuggageId = 0;
for (uint8 cnt = 0; cnt < 17; cnt++) // force res manager to keep mouse
_resMan->resOpen(MSE_POINTER + cnt); // cursors in memory all the time
CursorMan.showMouse(false);
createPointer(0, 0);
}
Domain *
Constructors::createBitCast(const llvm::ConstantExpr &value,
const std::vector<const Domain*> &operands,
const llvm::Value &place) const
{
// BitCast from Pointer. It is always a bitcast to some other
// pointer.
const Pointer::Pointer *pointer = dynCast<Pointer::Pointer>(*operands.begin());
const llvm::PointerType *pointerType =
checkedCast<llvm::PointerType>(value.getType());
if (pointer)
{
CANAL_ASSERT(pointerType);
return pointer->bitcast(*pointerType);
}
// BitCast from anything to a pointer.
if (pointerType)
{
Domain *result;
result = createPointer(*pointerType);
Pointer::Utils::addTarget(*result,
Pointer::Target::Block,
&place,
*value.op_begin(),
std::vector<Domain*>(),
NULL);
return result;
}
// BitCast from non-pointer to another non-pointer.
CANAL_NOT_IMPLEMENTED();
}
int sci_umf_lufact(char* fname, void* pvApiCtx)
{
SciErr sciErr;
int stat = 0;
SciSparse AA;
CcsSparse A;
int mA = 0; // rows
int nA = 0; // cols
int iNbItem = 0;
int* piNbItemRow = NULL;
int* piColPos = NULL;
double* pdblSpReal = NULL;
double* pdblSpImg = NULL;
/* umfpack stuff */
double* Control = NULL;
double* Info = NULL;
void* Symbolic = NULL;
void* Numeric = NULL;
int* piAddr1 = NULL;
int iComplex = 0;
int iType1 = 0;
/* Check numbers of input/output arguments */
CheckInputArgument(pvApiCtx, 1, 1);
CheckOutputArgument(pvApiCtx, 1, 1);
/* get A the sparse matrix to factorize */
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr1);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
/* check if the first argument is a sparse matrix */
sciErr = getVarType(pvApiCtx, piAddr1, &iType1);
if (sciErr.iErr || iType1 != sci_sparse)
{
printError(&sciErr, 0);
Scierror(999, _("%s: Wrong type for input argument #%d: A sparse matrix expected.\n"), fname, 1);
return 1;
}
if (isVarComplex(pvApiCtx, piAddr1))
{
iComplex = 1;
sciErr = getComplexSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal, &pdblSpImg);
}
else
{
sciErr = getSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal);
}
if (sciErr.iErr)
{
FREE(piNbItemRow);
FREE(piColPos);
FREE(pdblSpReal);
if (pdblSpImg)
{
FREE(pdblSpImg);
}
printError(&sciErr, 0);
return 1;
}
// fill struct sparse
AA.m = mA;
AA.n = nA;
AA.it = iComplex;
AA.nel = iNbItem;
AA.mnel = piNbItemRow;
AA.icol = piColPos;
AA.R = pdblSpReal;
AA.I = pdblSpImg;
if (nA <= 0 || mA <= 0)
{
FREE(piNbItemRow);
FREE(piColPos);
FREE(pdblSpReal);
if (pdblSpImg)
{
FREE(pdblSpImg);
}
Scierror(999, _("%s: Wrong size for input argument #%d.\n"), fname, 1);
return 1;
}
SciSparseToCcsSparse(&AA, &A);
FREE(piNbItemRow);
FREE(piColPos);
FREE(pdblSpReal);
if (pdblSpImg)
{
FREE(pdblSpImg);
}
/* symbolic factorization */
if (A.it == 1)
{
stat = umfpack_zi_symbolic(nA, mA, A.p, A.irow, A.R, A.I, &Symbolic, Control, Info);
}
else
{
stat = umfpack_di_symbolic(nA, mA, A.p, A.irow, A.R, &Symbolic, Control, Info);
}
if (stat != UMFPACK_OK)
{
freeCcsSparse(A);
Scierror(999, _("%s: An error occurred: %s: %s\n"), fname, _("symbolic factorization"), UmfErrorMes(stat));
return 1;
}
/* numeric factorization */
if (A.it == 1)
{
stat = umfpack_zi_numeric(A.p, A.irow, A.R, A.I, Symbolic, &Numeric, Control, Info);
}
else
{
stat = umfpack_di_numeric(A.p, A.irow, A.R, Symbolic, &Numeric, Control, Info);
}
if (A.it == 1)
{
umfpack_zi_free_symbolic(&Symbolic);
}
else
{
umfpack_di_free_symbolic(&Symbolic);
}
if ( stat != UMFPACK_OK && stat != UMFPACK_WARNING_singular_matrix )
{
freeCcsSparse(A);
Scierror(999, _("%s: An error occurred: %s: %s\n"), fname, _("symbolic factorization"), UmfErrorMes(stat));
return 1;
}
if ( stat == UMFPACK_WARNING_singular_matrix && mA == nA )
{
if (getWarningMode())
{
Sciwarning("\n%s:%s\n", _("Warning"), _("The (square) matrix appears to be singular."));
}
}
/* add the pointer in the list ListNumeric */
if (! AddAdrToList(Numeric, A.it, &ListNumeric))
{
/* AddAdrToList return 0 if malloc have failed : as it is just
for storing 2 pointers this is unlikely to occurs but ... */
if (A.it == 1)
{
umfpack_zi_free_numeric(&Numeric);
}
else
{
umfpack_di_free_numeric(&Numeric);
}
freeCcsSparse(A);
Scierror(999, _("%s: An error occurred: %s\n"), fname, _("no place to store the LU pointer in ListNumeric."));
return 1;
}
freeCcsSparse(A);
/* create the scilab object to store the pointer onto the LU factors */
sciErr = createPointer(pvApiCtx, 2, Numeric);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
/* return the pointer */
AssignOutputVariable(pvApiCtx, 1) = 2;
ReturnArguments(pvApiCtx);
return 0;
}
int sci_gpuDotMult(char *fname)
{
CheckRhs(2, 2);
CheckLhs(1, 1);
SciErr sciErr;
int* piAddr_A = NULL;
int* piAddr_B = NULL;
GpuPointer* gpuPtrA = NULL;
GpuPointer* gpuPtrB = NULL;
GpuPointer* gpuPtrC = NULL;
double* h = NULL;
double* hi = NULL;
int rows = 0;
int cols = 0;
void* pvPtrA = NULL;
void* pvPtrB = NULL;
int inputType_A;
int inputType_B;
try
{
if (!isGpuInit())
{
throw "gpu is not initialised. Please launch gpuInit() before use this function.";
}
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr_A);
if (sciErr.iErr)
{
throw sciErr;
}
sciErr = getVarAddressFromPosition(pvApiCtx, 2, &piAddr_B);
if (sciErr.iErr)
{
throw sciErr;
}
/* ---- Check type of arguments and get data ---- */
/* */
/* Pointer to host / Pointer to device */
/* Matrix real / Matrix complex */
/* */
/* ---------------------------------------------- */
sciErr = getVarType(pvApiCtx, piAddr_A, &inputType_A);
if (sciErr.iErr)
{
throw sciErr;
}
sciErr = getVarType(pvApiCtx, piAddr_B, &inputType_B);
if (sciErr.iErr)
{
throw sciErr;
}
if (inputType_A == sci_pointer)
{
sciErr = getPointer(pvApiCtx, piAddr_A, (void**)&pvPtrA);
if (sciErr.iErr)
{
throw sciErr;
}
gpuPtrA = (GpuPointer*)pvPtrA;
if (!PointerManager::getInstance()->findGpuPointerInManager(gpuPtrA))
{
throw "gpuDotMult : Bad type for input argument #1: Variables created with GPU functions expected.";
}
if (useCuda() && gpuPtrA->getGpuType() != GpuPointer::CudaType)
{
throw "gpuDotMult : Bad type for input argument #1: A Cuda pointer expected.";
}
if (useCuda() == false && gpuPtrA->getGpuType() != GpuPointer::OpenCLType)
{
throw "gpuDotMult : Bad type for input argument #1: A OpenCL pointer expected.";
}
}
else if (inputType_A == sci_matrix)
{
if (isVarComplex(pvApiCtx, piAddr_A))
{
sciErr = getComplexMatrixOfDouble(pvApiCtx, piAddr_A, &rows, &cols, &h, &hi);
if (sciErr.iErr)
{
throw sciErr;
}
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrA = new PointerCuda(h, hi, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "gpuDotMult: not implemented with OpenCL.";
}
#endif
}
else
{
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_A, &rows, &cols, &h);
if (sciErr.iErr)
{
throw sciErr;
}
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrA = new PointerCuda(h, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "gpuDotMult: not implemented with OpenCL.";
}
#endif
}
}
else
{
throw "gpuDotMult : Bad type for input argument #1: A GPU or CPU matrix expected.";
}
if (inputType_B == sci_pointer)
{
sciErr = getPointer(pvApiCtx, piAddr_B, (void**)&pvPtrB);
if (sciErr.iErr)
{
throw sciErr;
}
gpuPtrB = (GpuPointer*)pvPtrB;
if (!PointerManager::getInstance()->findGpuPointerInManager(gpuPtrB))
{
throw "gpuDotMult : Bad type for input argument #2: Variables created with GPU functions expected.";
}
if (useCuda() && gpuPtrB->getGpuType() != GpuPointer::CudaType)
{
throw "gpuDotMult : Bad type for input argument #2: A Cuda pointer expected.";
}
if (useCuda() == false && gpuPtrB->getGpuType() != GpuPointer::OpenCLType)
{
throw "gpuDotMult : Bad type for input argument #2: A OpenCL pointer expected.";
}
}
else if (inputType_B == sci_matrix)
{
if (isVarComplex(pvApiCtx, piAddr_B))
{
sciErr = getComplexMatrixOfDouble(pvApiCtx, piAddr_B, &rows, &cols, &h, &hi);
if (sciErr.iErr)
{
throw sciErr;
}
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrB = new PointerCuda(h, hi, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "gpuDotMult: not implemented with OpenCL.";
}
#endif
}
else
{
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_B, &rows, &cols, &h);
if (sciErr.iErr)
{
throw sciErr;
}
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrB = new PointerCuda(h, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "gpuDotMult: not implemented with OpenCL.";
}
#endif
}
}
else
{
throw "gpuDotMult : Bad type for input argument #2: A GPU or CPU matrix expected.";
}
//performe operation.
if (gpuPtrA->getSize() == 1 || gpuPtrB->getSize() == 1)
{
gpuPtrC = *gpuPtrA * *gpuPtrB;
}
else if (gpuPtrA->getRows() == gpuPtrB->getRows() && gpuPtrA->getCols() == gpuPtrB->getCols())
{
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrC = cudaDotMult(dynamic_cast<PointerCuda*>(gpuPtrA), dynamic_cast<PointerCuda*>(gpuPtrB));
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "gpuDotMult: not implemented with OpenCL.";
}
#endif
}
else
{
throw "gpuDotMult : Bad size for inputs arguments: Same sizes expected.";
}
// Keep the result on the Device.
PointerManager::getInstance()->addGpuPointerInManager(gpuPtrC);
sciErr = createPointer(pvApiCtx, Rhs + 1, (void*)gpuPtrC);
if (sciErr.iErr)
{
throw sciErr;
}
LhsVar(1) = Rhs + 1;
if (inputType_A == sci_matrix && gpuPtrA != NULL)
{
delete gpuPtrA;
}
if (inputType_B == sci_matrix && gpuPtrB != NULL)
{
delete gpuPtrB;
}
PutLhsVar();
return 0;
}
catch (const char* str)
{
Scierror(999, "%s\n", str);
}
catch (SciErr E)
{
printError(&E, 0);
}
if (inputType_A == sci_matrix && gpuPtrA != NULL)
{
delete gpuPtrA;
}
if (inputType_B == sci_matrix && gpuPtrB != NULL)
{
delete gpuPtrB;
}
if (gpuPtrC != NULL)
{
delete gpuPtrC;
}
return EXIT_FAILURE;
}
int sci_taucs_chfact(char* fname, void* pvApiCtx)
{
SciErr sciErr;
int stat = 0;
int* perm = NULL;
int* invperm = NULL;
taucs_ccs_matrix *PAPT;
taucs_ccs_matrix B;
void *C = NULL;
taucs_handle_factors *pC;
SciSparse A;
int mA = 0; // rows
int nA = 0; // cols
int iNbItem = 0;
int* piNbItemRow = NULL;
int* piColPos = NULL;
double* pdblSpReal = NULL;
double* pdblSpImg = NULL;
int iComplex = 0;
int* piAddr1 = NULL;
/* Check numbers of input/output arguments */
CheckInputArgument(pvApiCtx, 1, 1);
CheckOutputArgument(pvApiCtx, 1, 1);
/* get A the sparse matrix to factorize */
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr1);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
if (isVarComplex(pvApiCtx, piAddr1))
{
iComplex = 1;
sciErr = getComplexSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal, &pdblSpImg);
}
else
{
sciErr = getSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal);
}
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
// fill struct sparse
A.m = mA;
A.n = nA;
A.it = iComplex;
A.nel = iNbItem;
A.mnel = piNbItemRow;
A.icol = piColPos;
A.R = pdblSpReal;
A.I = pdblSpImg;
stat = spd_sci_sparse_to_taucs_sparse(&A, &B);
if ( stat != A_PRIORI_OK )
{
if ( stat == MAT_IS_NOT_SPD )
{
freeTaucsSparse(B);
Scierror(999, _("%s: Wrong value for input argument #%d: Must be symmetric positive definite matrix."), fname, 1);
}
/* the message for the other problem (not enough memory in stk) is treated automaticaly */
return 1;
}
/* find the permutation */
taucs_ccs_genmmd(&B, &perm, &invperm);
if ( !perm )
{
freeTaucsSparse(B);
Scierror(999, _("%s: No more memory.\n") , fname);
return 1;
}
/* apply permutation */
PAPT = taucs_ccs_permute_symmetrically(&B, perm, invperm);
FREE(invperm);
freeTaucsSparse(B);
/* factor */
C = taucs_ccs_factor_llt_mf(PAPT);
taucs_ccs_free(PAPT);
if (C == NULL)
{
/* Note : an error indicator is given in the main scilab window
* (out of memory, no positive definite matrix , etc ...)
*/
Scierror(999, _("%s: An error occurred: %s\n"), fname, _("factorization"));
return 1;
}
/* put in an handle (Chol fact + perm + size) */
pC = (taucs_handle_factors*)MALLOC( sizeof(taucs_handle_factors) );
pC->p = perm;
pC->C = C;
pC->n = A.n;
/* add in the list of Chol Factors */
AddAdrToList((Adr) pC, 0, &ListCholFactors); /* FIXME add a test here .. */
/* create the scilab object to store the pointer onto the Chol handle */
sciErr = createPointer(pvApiCtx, 2, (void *)pC);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
/* return the pointer */
AssignOutputVariable(pvApiCtx, 1) = 2;
ReturnArguments(pvApiCtx);
return 0;
}
Domain *
Constructors::create(const llvm::Constant &value,
const llvm::Value &place,
const State *state) const
{
if (llvm::isa<llvm::UndefValue>(value))
return create(*value.getType());
if (llvm::isa<llvm::ConstantInt>(value))
{
const llvm::ConstantInt &intValue =
checkedCast<llvm::ConstantInt>(value);
const llvm::APInt &i = intValue.getValue();
return createInteger(i);
}
if (llvm::isa<llvm::ConstantPointerNull>(value))
{
const llvm::ConstantPointerNull &nullValue =
checkedCast<llvm::ConstantPointerNull>(value);
const llvm::PointerType &pointerType = *nullValue.getType();
Domain *constPointer = createPointer(pointerType);
constPointer->setZero(&place);
return constPointer;
}
if (llvm::isa<llvm::ConstantExpr>(value))
{
const llvm::ConstantExpr &exprValue =
checkedCast<llvm::ConstantExpr>(value);
return createConstantExpr(exprValue, place, state);
}
if (llvm::isa<llvm::ConstantFP>(value))
{
const llvm::ConstantFP &fp = checkedCast<llvm::ConstantFP>(value);
const llvm::APFloat &number = fp.getValueAPF();
return createFloat(number);
}
if (llvm::isa<llvm::ConstantStruct>(value))
{
const llvm::ConstantStruct &structValue =
checkedCast<llvm::ConstantStruct>(value);
uint64_t elementCount = structValue.getType()->getNumElements();
std::vector<Domain*> members;
for (uint64_t i = 0; i < elementCount; ++i)
{
members.push_back(create(*structValue.getOperand(i),
*structValue.getOperand(i),
state));
}
return createStructure(*structValue.getType(), members);
}
if (llvm::isa<llvm::ConstantVector>(value))
{
const llvm::ConstantVector &vectorValue =
checkedCast<llvm::ConstantVector>(value);
// VectorType::getNumElements returns unsigned int.
unsigned elementCount = vectorValue.getType()->getNumElements();
std::vector<Domain*> values;
for (unsigned i = 0; i < elementCount; ++i)
{
values.push_back(create(*vectorValue.getOperand(i),
*vectorValue.getOperand(i),
state));
}
return createArray(*vectorValue.getType(), values);
}
if (llvm::isa<llvm::ConstantArray>(value))
{
const llvm::ConstantArray &arrayValue =
checkedCast<llvm::ConstantArray>(value);
// ArrayType::getNumElements returns uint64_t.
uint64_t elementCount = arrayValue.getType()->getNumElements();
std::vector<Domain*> values;
for (uint64_t i = 0; i < elementCount; ++i)
{
values.push_back(create(*arrayValue.getOperand(i),
*arrayValue.getOperand(i),
state));
}
return createArray(*arrayValue.getType(), values);
}
#if (LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 1) || LLVM_VERSION_MAJOR > 3
// llvm::isa<llvm::ConstantDataSequential> returns false for an
// llvm::ConstantDataArray/Vector instance at least on on LLVM
// 3.1.
if (llvm::isa<llvm::ConstantDataVector>(value) ||
llvm::isa<llvm::ConstantDataArray>(value))
{
const llvm::ConstantDataSequential &sequentialValue =
checkedCast<llvm::ConstantDataSequential>(value);
unsigned elementCount = sequentialValue.getNumElements();
std::vector<Domain*> values;
for (unsigned i = 0; i < elementCount; ++i)
{
values.push_back(create(*sequentialValue.getElementAsConstant(i),
place,
state));
}
return createArray(*sequentialValue.getType(), values);
}
#endif
if (llvm::isa<llvm::ConstantAggregateZero>(value))
{
const llvm::Type *type = value.getType();
Domain *result = Constructors::create(*type);
result->setZero(&place);
return result;
}
if (llvm::isa<llvm::Function>(value))
{
const llvm::Function &functionValue =
checkedCast<llvm::Function>(value);
Domain *constPointer;
constPointer = createPointer(*llvm::PointerType::getUnqual(
functionValue.getFunctionType()));
Pointer::Utils::addTarget(*constPointer,
Pointer::Target::Function,
&place,
&value,
std::vector<Domain*>(),
NULL);
return constPointer;
}
CANAL_NOT_IMPLEMENTED();
}
/* ========================================================================== */
int sci_gpuLU(char *fname)
{
CheckRhs(1,2);
CheckLhs(2,2);
#ifdef WITH_CUDA
cublasStatus status;
#endif
SciErr sciErr;
int* piAddr_A = NULL;
double* h_A = NULL;
double* hi_A = NULL;
int rows_A;
int cols_A;
int* piAddr_Opt = NULL;
double* option = NULL;
int rows_Opt;
int cols_Opt;
void* d_A = NULL;
int na;
void* pvPtr = NULL;
int size_A = sizeof(double);
bool bComplex_A = FALSE;
int inputType_A;
int inputType_Opt;
double res;
int posOutput = 1;
try
{
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr_A);
if(sciErr.iErr) throw sciErr;
if(Rhs == 2)
{
sciErr = getVarAddressFromPosition(pvApiCtx, 2, &piAddr_Opt);
if(sciErr.iErr) throw sciErr;
sciErr = getVarType(pvApiCtx, piAddr_Opt, &inputType_Opt);
if(sciErr.iErr) throw sciErr;
if(inputType_Opt == sci_matrix)
{
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_Opt, &rows_Opt, &cols_Opt, &option);
if(sciErr.iErr) throw sciErr;
}
else
throw "Option syntax is [number,number].";
}
else
{
rows_Opt=1;
cols_Opt=2;
option = (double*)malloc(2*sizeof(double));
option[0]=0;
option[1]=0;
}
if(rows_Opt != 1 || cols_Opt != 2)
throw "Option syntax is [number,number].";
if((int)option[1] == 1 && !isGpuInit())
throw "gpu is not initialised. Please launch gpuInit() before use this function.";
sciErr = getVarType(pvApiCtx, piAddr_A, &inputType_A);
if(sciErr.iErr) throw sciErr;
#ifdef WITH_CUDA
if (useCuda())
{
if(inputType_A == sci_pointer)
{
sciErr = getPointer(pvApiCtx, piAddr_A, (void**)&pvPtr);
if(sciErr.iErr) throw sciErr;
gpuMat_CUDA* gmat;
gmat = static_cast<gpuMat_CUDA*>(pvPtr);
if(!gmat->useCuda)
throw "Please switch to OpenCL mode before use this data.";
rows_A=gmat->rows;
cols_A=gmat->columns;
if(gmat->complex)
{
bComplex_A = TRUE;
size_A = sizeof(cuDoubleComplex);
d_A=(cuDoubleComplex*)gmat->ptr->get_ptr();
}
else
d_A=(double*)gmat->ptr->get_ptr();
// Initialize CUBLAS
status = cublasInit();
if (status != CUBLAS_STATUS_SUCCESS) throw status;
na = rows_A * cols_A;
}
else if(inputType_A == 1)
{
// Get size and data
if(isVarComplex(pvApiCtx, piAddr_A))
{
sciErr = getComplexMatrixOfDouble(pvApiCtx, piAddr_A, &rows_A, &cols_A, &h_A, &hi_A);
if(sciErr.iErr) throw sciErr;
size_A = sizeof(cuDoubleComplex);
bComplex_A = TRUE;
}
else
{
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_A, &rows_A, &cols_A, &h_A);
if(sciErr.iErr) throw sciErr;
}
na = rows_A * cols_A;
// Initialize CUBLAS
status = cublasInit();
if (status != CUBLAS_STATUS_SUCCESS) throw status;
// Allocate device memory
status = cublasAlloc(na, size_A, (void**)&d_A);
if (status != CUBLAS_STATUS_SUCCESS) throw status;
// Initialize the device matrices with the host matrices
if(!bComplex_A)
{
status = cublasSetMatrix(rows_A,cols_A, sizeof(double), h_A, rows_A, (double*)d_A, rows_A);
if (status != CUBLAS_STATUS_SUCCESS) throw status;
}
else
writecucomplex(h_A, hi_A, rows_A, cols_A, (cuDoubleComplex *)d_A);
}
else
throw "Bad argument type.";
cuDoubleComplex resComplex;
// Performs operation
if(!bComplex_A)
status = decomposeBlockedLU(rows_A, cols_A, rows_A, (double*)d_A, 1);
// else
// resComplex = cublasZtrsm(na,(cuDoubleComplex*)d_A);
if (status != CUBLAS_STATUS_SUCCESS) throw status;
// Put the result in scilab
switch((int)option[0])
{
case 2 :
case 1 : sciprint("The first option must be 0 for this function. Considered as 0.\n");
case 0 : // Keep the result on the Host.
{ // Put the result in scilab
if(!bComplex_A)
{
double* h_res = NULL;
sciErr=allocMatrixOfDouble(pvApiCtx, Rhs + posOutput, rows_A, cols_A, &h_res);
if(sciErr.iErr) throw sciErr;
status = cublasGetMatrix(rows_A,cols_A, sizeof(double), (double*)d_A, rows_A, h_res, rows_A);
if (status != CUBLAS_STATUS_SUCCESS) throw status;
}
else
{
sciErr = createComplexMatrixOfDouble(pvApiCtx, Rhs + posOutput, 1, 1, &resComplex.x,&resComplex.y);
if(sciErr.iErr) throw sciErr;
}
LhsVar(posOutput)=Rhs+posOutput;
posOutput++;
break;
}
default : throw "First option argument must be 0 or 1 or 2.";
}
switch((int)option[1])
{
case 0 : // Don't keep the data input on Device.
{
if(inputType_A == sci_matrix)
{
status = cublasFree(d_A);
if (status != CUBLAS_STATUS_SUCCESS) throw status;
d_A = NULL;
}
break;
}
case 1 : // Keep data of the fisrt argument on Device and return the Device pointer.
{
if(inputType_A == sci_matrix)
{
gpuMat_CUDA* dptr;
gpuMat_CUDA tmp={getCudaContext()->genMatrix<double>(getCudaQueue(),rows_A*cols_A),rows_A,cols_A};
dptr=new gpuMat_CUDA(tmp);
dptr->useCuda = true;
dptr->ptr->set_ptr((double*)d_A);
if(bComplex_A)
dptr->complex=TRUE;
else
dptr->complex=FALSE;
sciErr = createPointer(pvApiCtx,Rhs+posOutput, (void*)dptr);
if(sciErr.iErr) throw sciErr;
LhsVar(posOutput)=Rhs+posOutput;
}
else
throw "The first input argument is already a GPU variable.";
posOutput++;
break;
}
default : throw "Second option argument must be 0 or 1.";
}
// Shutdown
status = cublasShutdown();
if (status != CUBLAS_STATUS_SUCCESS) throw status;
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
throw "not implemented with OpenCL.";
}
#endif
if(Rhs == 1)
{
free(option);
option = NULL;
}
if(posOutput < Lhs+1)
throw "Too many output arguments.";
if(posOutput > Lhs+1)
throw "Too few output arguments.";
PutLhsVar();
return 0;
}
catch(const char* str)
{
Scierror(999,"%s\n",str);
}
catch(SciErr E)
{
printError(&E, 0);
}
#ifdef WITH_CUDA
catch(cudaError_t cudaE)
{
GpuError::treat_error<CUDAmode>((CUDAmode::Status)cudaE);
}
catch(cublasStatus CublasE)
{
GpuError::treat_error<CUDAmode>((CUDAmode::Status)CublasE,1);
}
if (useCuda())
{
if(inputType_A == 1 && d_A != NULL) cudaFree(d_A);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
Scierror(999,"not implemented with OpenCL.\n");
}
#endif
if(Rhs == 1 && option != NULL) free(option);
return EXIT_FAILURE;
}
int sci_gpuOnes(char *fname)
{
CheckLhs(1, 1);
void* pvPtr = NULL;
int* piAddr = NULL;
SciErr sciErr;
int inputType;
int iRows = 0;
int iCols = 0;
GpuPointer* gpOut = NULL;
try
{
if (!isGpuInit())
{
throw "gpu is not initialised. Please launch gpuInit() before use this function.";
}
if (Rhs == 1)
{
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr);
if (sciErr.iErr)
{
throw sciErr;
}
sciErr = getVarType(pvApiCtx, piAddr, &inputType);
if (inputType == sci_pointer)
{
sciErr = getPointer(pvApiCtx, piAddr, (void**)&pvPtr);
if (sciErr.iErr)
{
throw sciErr;
}
GpuPointer* gmat = (GpuPointer*)(pvPtr);
if (!PointerManager::getInstance()->findGpuPointerInManager(gmat))
{
throw "gpuOnes : Bad type for input argument #1. Only variables created with GPU functions allowed.";
}
if (useCuda() && gmat->getGpuType() != GpuPointer::CudaType)
{
throw "gpuOnes : Bad type for input argument #1: A Cuda pointer expected.";
}
if (useCuda() == false && gmat->getGpuType() != GpuPointer::OpenCLType)
{
throw "gpuOnes : Bad type for input argument #1: A OpenCL pointer expected.";
}
if (gmat->getDims() > 2)
{
throw "gpuOnes : Hypermatrix not yet implemented.";
}
iRows = gmat->getRows();
iCols = gmat->getCols();
}
else if (inputType == sci_matrix)
{
// Get size and data
double* h;
sciErr = getMatrixOfDouble(pvApiCtx, piAddr, &iRows, &iCols, &h);
}
else
{
throw "gpuOnes : Bad type for input argument #1 : A Matrix or GPU pointer expected.";
}
}
else
{
if (Rhs > 2)
{
throw "gpuOnes : Hypermatrix not yet implemented.";
}
int* piDimsArray = new int[Rhs];
for (int i = 0; i < Rhs; i++)
{
sciErr = getVarAddressFromPosition(pvApiCtx, i + 1, &piAddr);
if (sciErr.iErr)
{
throw sciErr;
}
sciErr = getVarType(pvApiCtx, piAddr, &inputType);
if (inputType != sci_matrix)
{
throw "gpuOnes : Bad type for input argument #%d : A Matrix expected.";
}
double* h;
sciErr = getMatrixOfDouble(pvApiCtx, piAddr, &iRows, &iCols, &h);
if (iRows * iCols != 1)
{
char str[100];
sprintf(str, "gpuOnes : Wrong size for input argument #%d : A scalar expected.", i + 1);
throw str;
}
piDimsArray[i] = (int)h[0];
}
iRows = piDimsArray[0];
iCols = piDimsArray[1];
delete piDimsArray;
}
#ifdef WITH_CUDA
if (useCuda())
{
gpOut = new PointerCuda(iRows, iCols, false);
gpOut->initMatrix(1);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
Scierror(999, "gpuOnes: not implemented with OpenCL.\n");
}
#endif
PointerManager::getInstance()->addGpuPointerInManager(gpOut);
sciErr = createPointer(pvApiCtx, Rhs + 1, (void*)gpOut);
if (sciErr.iErr)
{
throw sciErr;
}
LhsVar(1) = Rhs + 1;
PutLhsVar();
return 0;
}
#ifdef WITH_CUDA
catch (cudaError_t cudaE)
{
GpuError::treat_error<CUDAmode>((CUDAmode::Status)cudaE);
}
#endif
catch (const char* str)
{
Scierror(999, "%s\n", str);
}
catch (SciErr E)
{
printError(&E, 0);
}
return EXIT_FAILURE;
}
int sci_gpuMatrix(char *fname)
{
CheckRhs(2, 3);
CheckLhs(1, 1);
SciErr sciErr;
int* piAddr_A = NULL;
int inputType_A = 0;
int* piAddr_R = NULL;
int inputType_R = 0;
int* piAddr_C = NULL;
int inputType_C = 0;
int rows = 0;
int cols = 0;
int newRows = 0;
int newCols = 0;
void* pvPtr = NULL;
GpuPointer* gpuPtrA = NULL;
try
{
if (!isGpuInit())
{
throw "gpu is not initialised. Please launch gpuInit() before use this function.";
}
//--- Get input matrix ---
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr_A);
if (sciErr.iErr)
{
throw sciErr;
}
// Get size and data
sciErr = getVarType(pvApiCtx, piAddr_A, &inputType_A);
if (sciErr.iErr)
{
throw sciErr;
}
//--- Get new Rows size or vector of sizes---
sciErr = getVarAddressFromPosition(pvApiCtx, 2, &piAddr_R);
if (sciErr.iErr)
{
throw sciErr;
}
// Get size and data
sciErr = getVarType(pvApiCtx, piAddr_R, &inputType_R);
if (sciErr.iErr)
{
throw sciErr;
}
if (inputType_R != sci_matrix)
{
throw "gpuMatrix : Bad type for input argument #2: A real scalar or row vector expected.";
}
if (isVarComplex(pvApiCtx, piAddr_A))
{
throw "gpuMatrix : Bad type for input argument #2: A real scalar or row vector expected.";
}
else
{
double* dRows = NULL;
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_R, &rows, &cols, &dRows);
if (sciErr.iErr)
{
throw sciErr;
}
if (nbInputArgument(pvApiCtx) == 2)
{
if (rows != 1 || cols != 2)
{
throw "gpuMatrix : Bad size for input argument #2: A row vector of size two expected.";
}
newRows = (int)dRows[0];
newCols = (int)dRows[1];
if (newCols < -1 || newCols == 0)
{
throw "gpuMatrix : Wrong value for input argument #3: -1 or positive value expected.";
}
}
else
{
newRows = (int)(*dRows);
}
if (newRows < -1 || newRows == 0)
{
throw "gpuMatrix : Wrong value for input argument #2: -1 or positive value expected.";
}
}
if (nbInputArgument(pvApiCtx) == 3)
{
//--- Get new Cols size---
sciErr = getVarAddressFromPosition(pvApiCtx, 3, &piAddr_C);
if (sciErr.iErr)
{
throw sciErr;
}
// Get size and data
sciErr = getVarType(pvApiCtx, piAddr_C, &inputType_C);
if (sciErr.iErr)
{
throw sciErr;
}
if (inputType_C != sci_matrix)
{
throw "gpuMatrix : Bad type for input argument #3: A real scalar expected.";
}
if (isVarComplex(pvApiCtx, piAddr_A))
{
throw "gpuMatrix : Bad type for input argument #3: A real scalar expected.";
}
else
{
double* dCols = NULL;
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_C, &rows, &cols, &dCols);
if (sciErr.iErr)
{
throw sciErr;
}
newCols = (int)(*dCols);
if (newCols < -1 || newCols == 0)
{
throw "gpuMatrix : Wrong value for input argument #3: -1 or positive value expected.";
}
}
}
if (inputType_A == sci_pointer)
{
sciErr = getPointer(pvApiCtx, piAddr_A, (void**)&pvPtr);
if (sciErr.iErr)
{
throw sciErr;
}
gpuPtrA = (GpuPointer*)pvPtr;
if (!PointerManager::getInstance()->findGpuPointerInManager(gpuPtrA))
{
throw "gpuMatrix : Bad type for input argument #1: Variables created with GPU functions expected.";
}
if (useCuda() && gpuPtrA->getGpuType() != GpuPointer::CudaType)
{
throw "gpuMatrix : Bad type for input argument #1: A Cuda pointer expected.";
}
if (useCuda() == false && gpuPtrA->getGpuType() != GpuPointer::OpenCLType)
{
throw "gpuMatrix : Bad type for input argument #1: A OpenCL pointer expected.";
}
rows = gpuPtrA->getRows();
cols = gpuPtrA->getCols();
}
else if (inputType_A == sci_matrix)
{
double* h = NULL;
if (isVarComplex(pvApiCtx, piAddr_A))
{
double* hi = NULL;
sciErr = getComplexMatrixOfDouble(pvApiCtx, piAddr_A, &rows, &cols, &h, &hi);
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrA = new PointerCuda(h, hi, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
Scierror(999, "gpuMatrix: not implemented with OpenCL.\n");
}
#endif
}
else
{
sciErr = getMatrixOfDouble(pvApiCtx, piAddr_A, &rows, &cols, &h);
#ifdef WITH_CUDA
if (useCuda())
{
gpuPtrA = new PointerCuda(h, rows, cols);
}
#endif
#ifdef WITH_OPENCL
if (!useCuda())
{
Scierror(999, "gpuMatrix: not implemented with OpenCL.\n");
}
#endif
}
if (sciErr.iErr)
{
throw sciErr;
}
}
else
{
throw "gpuMatrix : Bad type for input argument #1: A GPU or CPU matrix expected.";
}
if (newRows == -1 && newCols != -1)
{
newRows = rows * cols / newCols;
}
else if (newRows != -1 && newCols == -1)
{
newCols = rows * cols / newRows;
}
if (rows * cols != newRows * newCols)
{
throw "gpuMatrix : Wrong value for input arguments #2 and 3: Correct size expected.";
}
#ifdef WITH_OPENCL
if (!useCuda())
{
Scierror(999, "gpuMatrix: not implemented with OpenCL.\n");
}
#endif
GpuPointer* gpuOut = gpuPtrA->clone();
gpuOut->setRows(newRows);
gpuOut->setCols(newCols);
// Put the result in scilab
PointerManager::getInstance()->addGpuPointerInManager(gpuOut);
sciErr = createPointer(pvApiCtx, Rhs + 1, (void*)gpuOut);
LhsVar(1) = Rhs + 1;
if (inputType_A == 1 && gpuPtrA != NULL)
{
delete gpuPtrA;
}
PutLhsVar();
return 0;
}
catch (const char* str)
{
Scierror(999, "%s\n", str);
}
catch (SciErr E)
{
printError(&E, 0);
}
if (inputType_A == 1 && gpuPtrA != NULL)
{
delete gpuPtrA;
}
return EXIT_FAILURE;
}