CToken::range_type CToken::next_range()
{
skip();
string::size_type start = cur_;
string::size_type tmp = cur_;
if ( cur_ != string::npos ) cur_ = str_.find_first_of(del_,cur_);
if ( cur_ == string::npos ) return range_type(start,str_.length()-start);
if ( ret_ && start == cur_ && del_.find(str_[cur_]) != string::npos ) ++cur_;
return range_type(start,cur_-start);
}
/*
//case other.col is odd
3 ) else if ( other.col() & 1 )
{
//<1>
if ( other.col() & 2 ) {...}
//<2>
else
{
cut the last col from other get two new matrices
[new_other < - other] [C, OC - 1]
[last_col < - other] [C, 1]
do the multications
[new_ans < - this * new_other] [R, OC - 1]
[last_col_ans < - this * last_col] [R, 1]
merge the two matrices
[ans < - new_ans | last_col_ans] [R, 1]
[i.e. last_col_ans as the last col of ans]
}
}
*/
zen_type& oc2( const zen_type& other )
{
zen_type& zen = static_cast<zen_type&>( *this );
const zen_type new_other( other, range_type( 0, other.row() ), range_type( 0, other.col() - 1 ) );
const zen_type last_col( other, range_type( 0, other.row() ), range_type( other.col() - 1, other.col() ) );
const zen_type& new_ans = zen * new_other;
const zen_type& last_col_ans = zen * last_col;
const zen_type& ans = new_ans || last_col_ans;
zen.clone( ans, 0, zen.row(), 0, other.col() );
return zen;
}
/*
1 ) else if ( row() & 1 )
{
//<1>
if ( row() & 2 ) {...}
//<2>
else
{
cut last row from the matrix, get two new matrices
[ new_this < - this ] [R - 1, C]
[ last_row < - this ] [1, C]
do the multiplications
[ new_ans < - new_this * other ] [R - 1, OC]
[ last_row_ans < - last_row * other ] [1, OC]
merge the two matrices to generate the ans
[ ans < - new_ans | last_row_ans ] [R, OC]
[ i.e. last_row_ans appended to new_ans as the last row ]
}
}
*/
zen_type& rr2( const zen_type& other )
{
zen_type& zen = static_cast<zen_type&>( *this );
const zen_type new_this( zen, range_type( 0, zen.row() - 1 ), range_type( 0, zen.col() ) );
const zen_type last_row( zen, range_type( zen.row() - 1, zen.row() ), range_type( 0, zen.col() ) );
const zen_type& new_ans = new_this * other;
const zen_type& last_row_ans = last_row * other;
const zen_type& ans = new_ans && last_row_ans;
zen.clone( ans, 0, zen.row(), 0, other.col() );
return zen;
}
/*for each range in the option ,set up the to_print array*/
int set_each_range(const char range[], int to_print[]) {
int type;
int pos;
int counter;
int i;
type = range_type(range);
if (type == 0) {
fprintf(stderr,"invalid option range %s\n", range);
return 0;
}
sscanf(range, "%d:%d", &pos, &counter);
if (pos >= 1024)
return 1;
if (type == 1) {
to_print[pos - 1] = 1;
return 1;
}
if (type == 2) {
for (i = pos - 1; i < 1024; i++)
to_print[i] = 1;
return 1;
}
for (i = pos - 1; i < pos + counter - 1 && i < 1024; i++)
to_print[i] = 1;
return 1;
}
void
range_run<Char>::clear(range_type const& range)
{
BOOST_ASSERT(is_valid(range));
if (!run.empty())
{
// search the ranges for one that potentially includes 'range'
typename storage_type::iterator iter =
std::upper_bound(
run.begin(), run.end(), range,
range_compare<range_type>()
);
// 'range' starts with or after another range:
if (iter != run.begin())
{
typename storage_type::iterator const left_iter = iter-1;
// 'range' starts after '*left_iter':
if (left_iter->first < range.first)
{
// if 'range' is completely included inside '*left_iter':
// need to break it apart into two ranges (punch a hole),
if (left_iter->last > range.last)
{
Char save_last = left_iter->last;
left_iter->last = range.first-1;
run.insert(iter, range_type(range.last+1, save_last));
return;
}
// if 'range' contains 'left_iter->last':
// truncate '*left_iter' (clip its right)
else if (left_iter->last >= range.first)
{
left_iter->last = range.first-1;
}
}
// 'range' has the same left bound as '*left_iter': it
// must be removed or truncated by the code below
else
{
iter = left_iter;
}
}
// remove or truncate subsequent ranges that overlap with 'range':
typename storage_type::iterator i = iter;
// 1. skip subsequent ranges completely included in 'range'
while (i != run.end() && i->last <= range.last)
++i;
// 2. clip left of next range if overlapping with 'range'
if (i != run.end() && i->first <= range.last)
i->first = range.last+1;
// erase all ranges that 'range' contained
run.erase(iter, i);
}
}
/*
//case col is odd
2 ) else if ( col() & 1 )
{
//<1>
if ( zen.col() & 2 ) {...}
//<2>
else
{
cut last col of this from right side
[new_this < - this] [R, C - 1]
[last_col < - this] [R, 1]
cut last row of other from downside
[new_other < - other] [C - 1, OC]
[last_row < - other] [1, OC]
do the multiplicaitons
[new_ans < - new_this * new_other] [R, OC]
[res_col_row < - last_col * last_row] [R, OC]
do the addition to generate ans
[ans < - new_ans + res_col_row] [R, OC]
}
*/
zen_type& cc2( const zen_type & other )
{
zen_type& zen = static_cast<zen_type&>( *this );
//[new_this <- this] [R,C-1]
const zen_type new_this( zen, range_type( 0, zen.row() ), range_type( 0, zen.col() - 1 ) );
//[last_col <- this] [R,1]
const zen_type last_col( zen, range_type( 0, zen.row() ), range_type( zen.col() - 1, zen.col() ) );
//[new_other <- other] [C-1,OC]
const zen_type new_other( other, range_type( 0, other.row() - 1 ), range_type( 0, other.col() ) );
//[last_row <- other] [1, OC]
const zen_type last_row( other, range_type( other.row() - 1, other.row() ), range_type( 0, other.col() ) );
const zen_type& new_ans = new_this * new_other;
const zen_type& res_col_row = last_col * last_row;
const zen_type& ans = new_ans + res_col_row;
zen.clone( ans, 0, zen.row(), 0, other.col() );
return zen;
}
typename sge::sprite::intrusive::ordered::collection<
Choices,
Order
>::range_type
sge::sprite::intrusive::ordered::collection<
Choices,
Order
>::range()
{
return
range_type(
fcppt::make_ref(
collections_
)
);
}
static char *
parse_class(Cpattern p, char *iptr)
{
int endchar, firsttime = 1;
char *optr, *nptr;
if (*iptr++ == '[') {
endchar = ']';
/* TODO: surely [^]] is valid? */
if ((*iptr == '!' || *iptr == '^') && iptr[1] != ']') {
p->tp = CPAT_NCLASS;
iptr++;
} else
p->tp = CPAT_CCLASS;
} else {
endchar = '}';
p->tp = CPAT_EQUIV;
}
/* find end of class. End character can appear literally first. */
for (optr = iptr; optr == iptr || *optr != endchar; optr++)
if (!*optr)
return optr;
/*
* We can always fit the parsed class within the same length
* because of the tokenization (including a null byte).
*
* As the input string is metafied, but shouldn't contain shell
* tokens, we can just add our own tokens willy nilly.
*/
optr = p->u.str = zhalloc((optr-iptr) + 1);
while (firsttime || *iptr != endchar) {
int ch;
if (*iptr == '[' && iptr[1] == ':' &&
(nptr = strchr((char *)iptr + 2, ':')) && nptr[1] == ']') {
/* Range type */
iptr += 2;
ch = range_type((char *)iptr, nptr-iptr);
iptr = nptr + 2;
if (ch != PP_UNKWN)
*optr++ = STOUC(Meta) + ch;
} else {
/* characters stay metafied */
char *ptr1 = iptr;
if (*iptr == Meta)
iptr++;
iptr++;
if (*iptr == '-' && iptr[1] && iptr[1] != endchar) {
/* a run of characters */
iptr++;
/* range token */
*optr++ = Meta + PP_RANGE;
/* start of range character */
if (*ptr1 == Meta) {
*optr++ = Meta;
*optr++ = ptr1[1] ^ 32;
} else
*optr++ = *ptr1;
if (*iptr == Meta) {
*optr++ = *iptr++;
*optr++ = *iptr++;
} else
*optr++ = *iptr++;
} else {
if (*ptr1 == Meta) {
*optr++ = Meta;
*optr++ = ptr1[1] ^ 32;
} else
*optr++ = *ptr1;
}
}
firsttime = 0;
}
*optr = '\0';
return iptr;
}
/*
4 ) else
{
//strassen algorithm
}
*/
zen_type& strassen_multiply( const zen_type& other )
{
zen_type& zen = static_cast<zen_type&>( *this );
const size_type R_2 = zen.row() >> 1;
const size_type C_2 = zen.col() >> 1;
const size_type OR_2 = C_2;
const size_type OC_2 = other.col() >> 1;
const zen_type a_00( zen, range_type( 0, R_2 ), range_type( 0, C_2 ) );
const zen_type a_01( zen, range_type( 0, R_2 ), range_type( C_2, zen.col() ) );
const zen_type a_10( zen, range_type( R_2, zen.row() ), range_type( 0, C_2 ) );
const zen_type a_11( zen, range_type( R_2, zen.row() ), range_type( C_2, zen.col() ) );
const zen_type b_00( other, range_type( 0, OR_2 ), range_type( 0, OC_2 ) );
const zen_type b_01( other, range_type( 0, OR_2 ), range_type( OC_2, other.col() ) );
const zen_type b_10( other, range_type( OR_2, other.row() ), range_type( 0, OC_2 ) );
const zen_type b_11( other, range_type( OR_2, other.row() ), range_type( OC_2, other.col() ) );
const zen_type& Q_0 = ( a_00 + a_11 ) * ( b_00 + b_11 );
const zen_type& Q_1 = ( a_10 + a_11 ) * b_00;
const zen_type& Q_2 = a_00 * ( b_01 - b_11 );
const zen_type& Q_3 = a_11 * ( -b_00 + b_10 );
const zen_type& Q_4 = ( a_00 + a_01 ) * b_11;
const zen_type& Q_5 = ( -a_00 + a_10 ) * ( b_00 + b_01 );
const zen_type& Q_6 = ( a_01 - a_11 ) * ( b_10 + b_11 );
const zen_type& c_00 = Q_0 + Q_3 - Q_4 + Q_6;
const zen_type& c_10 = Q_1 + Q_3;
const zen_type& c_01 = Q_2 + Q_4;
const zen_type& c_11 = Q_0 - Q_1 + Q_2 + Q_5;
const zen_type& ans = ( c_00 || c_01 ) &&
( c_10 || c_11 );
zen.clone( ans, 0, zen.row(), 0, other.col() );
return zen;
}
int ComputeBasis_HGRAD_Vector(const ordinal_type nworkset,
const ordinal_type C,
const ordinal_type order,
const bool verbose) {
typedef Vector<VectorTagType> VectorType;
typedef typename VectorTagType::value_type ValueType;
constexpr int VectorLength = VectorTagType::length;
Teuchos::RCP<std::ostream> verboseStream;
Teuchos::oblackholestream bhs; // outputs nothing
if (verbose)
verboseStream = Teuchos::rcp(&std::cout, false);
else
verboseStream = Teuchos::rcp(&bhs, false);
Teuchos::oblackholestream oldFormatState;
oldFormatState.copyfmt(std::cout);
typedef typename
Kokkos::Impl::is_space<DeviceSpaceType>::host_mirror_space::execution_space HostSpaceType ;
*verboseStream << "DeviceSpace:: "; DeviceSpaceType::print_configuration(*verboseStream, false);
*verboseStream << "HostSpace:: "; HostSpaceType::print_configuration(*verboseStream, false);
*verboseStream << "VectorLength:: " << (VectorLength) << "\n";
using BasisTypeHost = Basis_HGRAD_HEX_C1_FEM<HostSpaceType,ValueType,ValueType>;
using ImplBasisType = Impl::Basis_HGRAD_HEX_C1_FEM;
using range_type = Kokkos::pair<ordinal_type,ordinal_type>;
constexpr size_t LLC_CAPACITY = 32*1024*1024;
Intrepid2::Test::Flush<LLC_CAPACITY,DeviceSpaceType> flush;
Kokkos::Impl::Timer timer;
double t_vectorize = 0;
int errorFlag = 0;
BasisTypeHost hostBasis;
const auto cellTopo = hostBasis.getBaseCellTopology();
auto cubature = DefaultCubatureFactory::create<DeviceSpaceType,ValueType,ValueType>(cellTopo, order);
const ordinal_type
numCells = C,
numCellsAdjusted = C/VectorLength + (C%VectorLength > 0),
numVerts = cellTopo.getVertexCount(),
numDofs = hostBasis.getCardinality(),
numPoints = cubature->getNumPoints(),
spaceDim = cubature->getDimension();
Kokkos::DynRankView<ValueType,HostSpaceType> dofCoordsHost("dofCoordsHost", numDofs, spaceDim);
hostBasis.getDofCoords(dofCoordsHost);
const auto refNodesHost = Kokkos::subview(dofCoordsHost, range_type(0, numVerts), Kokkos::ALL());
// pertub nodes
Kokkos::DynRankView<VectorType,HostSpaceType> worksetCellsHost("worksetCellsHost", numCellsAdjusted, numVerts, spaceDim);
for (ordinal_type cell=0;cell<numCells;++cell) {
for (ordinal_type i=0;i<numVerts;++i)
for (ordinal_type j=0;j<spaceDim;++j) {
ValueType val = (rand()/(RAND_MAX + 1.0))*0.2 -0.1;
worksetCellsHost(cell/VectorLength, i, j)[cell%VectorLength] = refNodesHost(i, j) + val;
}
}
auto worksetCells = Kokkos::create_mirror_view(typename DeviceSpaceType::memory_space(), worksetCellsHost);
Kokkos::deep_copy(worksetCells, worksetCellsHost);
Kokkos::DynRankView<ValueType,DeviceSpaceType> refPoints("refPoints", numPoints, spaceDim), refWeights("refWeights", numPoints);
cubature->getCubature(refPoints, refWeights);
std::cout
<< "===============================================================================\n"
<< " Performance Test evaluating ComputeBasis \n"
<< " # of workset = " << nworkset << "\n"
<< " Test Array Structure (C,F,P,D) = " << numCells << ", " << numDofs << ", " << numPoints << ", " << spaceDim << "\n"
<< "===============================================================================\n";
*verboseStream
<< "\n"
<< "===============================================================================\n"
<< "TEST 1: evaluateFields vector version\n"
<< "===============================================================================\n";
try {
Kokkos::DynRankView<ValueType,DeviceSpaceType>
refBasisValues("refBasisValues", numDofs, numPoints),
refBasisGrads ("refBasisGrads", numDofs, numPoints, spaceDim);
ImplBasisType::getValues<DeviceSpaceType>(refBasisValues, refPoints, OPERATOR_VALUE);
ImplBasisType::getValues<DeviceSpaceType>(refBasisGrads, refPoints, OPERATOR_GRAD);
const ordinal_type ibegin = -3;
// testing vertical approach
{
Kokkos::DynRankView<VectorType,DeviceSpaceType>
weightedBasisValues("weightedBasisValues", numCellsAdjusted, numDofs, numPoints),
weightedBasisGrads ("weightedBasisGrads", numCellsAdjusted, numDofs, numPoints, spaceDim);
typedef F_hgrad_eval<VectorType,ValueType,DeviceSpaceType> FunctorType;
using range_policy_type = Kokkos::Experimental::MDRangePolicy
< DeviceSpaceType, Kokkos::Experimental::Rank<2>, Kokkos::IndexType<ordinal_type> >;
range_policy_type policy( { 0, 0 },
{ numCellsAdjusted, numPoints } );
FunctorType functor(weightedBasisValues,
weightedBasisGrads,
refBasisGrads,
worksetCells,
refWeights,
refBasisValues,
refBasisGrads);
for (ordinal_type iwork=ibegin;iwork<nworkset;++iwork) {
flush.run();
DeviceSpaceType::fence();
timer.reset();
Kokkos::parallel_for(policy, functor);
DeviceSpaceType::fence();
t_vectorize += (iwork >= 0)*timer.seconds();
}
}
} catch (std::exception err) {
*verboseStream << "UNEXPECTED ERROR !!! ----------------------------------------------------------\n";
*verboseStream << err.what() << '\n';
*verboseStream << "-------------------------------------------------------------------------------" << "\n\n";
errorFlag = -1000;
}
std::cout
<< "TEST HGRAD "
<< " t_vectorize = " << (t_vectorize/nworkset)
<< std::endl;
if (errorFlag != 0)
std::cout << "End Result: TEST FAILED\n";
else
std::cout << "End Result: TEST PASSED\n";
// reset format state of std::cout
std::cout.copyfmt(oldFormatState);
return errorFlag;
}