/external/eigen/unsupported/Eigen/CXX11/src/Tensor/ |
D | TensorCostModel.h | 25 class TensorOpCost { 55 TensorOpCost() : bytes_loaded_(0), bytes_stored_(0), compute_cycles_(0) {} in TensorOpCost() function 57 TensorOpCost(double bytes_loaded, double bytes_stored, double compute_cycles) in TensorOpCost() function 63 TensorOpCost(double bytes_loaded, double bytes_stored, double compute_cycles, in TensorOpCost() function 97 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost cwiseMin( in cwiseMin() 98 const TensorOpCost& rhs) const { in cwiseMin() 102 return TensorOpCost(bytes_loaded, bytes_stored, compute_cycles); in cwiseMin() 106 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost cwiseMax( in cwiseMax() 107 const TensorOpCost& rhs) const { in cwiseMax() 111 return TensorOpCost(bytes_loaded, bytes_stored, compute_cycles); in cwiseMax() [all …]
|
D | TensorBroadcasting.h | 351 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 353 double compute_cost = TensorOpCost::AddCost<Index>(); 356 compute_cost += TensorOpCost::DivCost<Index>(); 359 TensorOpCost::MulCost<Index>() + TensorOpCost::AddCost<Index>(); 362 compute_cost += TensorOpCost::MulCost<Index>() + 363 TensorOpCost::ModCost<Index>() + 364 TensorOpCost::AddCost<Index>(); 368 TensorOpCost::MulCost<Index>() + TensorOpCost::AddCost<Index>(); 372 TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|
D | TensorPadding.h | 189 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 190 TensorOpCost cost = m_impl.costPerCoeff(vectorized); 238 void updateCostPerDimension(TensorOpCost& cost, int i, bool first) const { 246 cost += TensorOpCost(0, 0, 2 * TensorOpCost::AddCost<Index>() + 247 reduction * (1 * TensorOpCost::AddCost<Index>())); 249 cost += TensorOpCost(0, 0, 2 * TensorOpCost::AddCost<Index>() + 250 2 * TensorOpCost::MulCost<Index>() + 251 reduction * (2 * TensorOpCost::MulCost<Index>() + 252 1 * TensorOpCost::DivCost<Index>()));
|
D | TensorInflation.h | 203 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 204 const double compute_cost = NumDims * (3 * TensorOpCost::DivCost<Index>() + 205 3 * TensorOpCost::MulCost<Index>() + 206 2 * TensorOpCost::AddCost<Index>()); 210 return TensorOpCost(); 212 TensorOpCost(sizeof(CoeffReturnType) * input_size / output_size, 0,
|
D | TensorEvaluator.h | 106 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { in costPerCoeff() 107 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, in costPerCoeff() 205 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 206 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, 268 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 270 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, 338 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 341 TensorOpCost(0, 0, functor_cost, vectorized, PacketSize); 416 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 421 TensorOpCost(0, 0, functor_cost, vectorized, PacketSize); [all …]
|
D | TensorReverse.h | 212 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 213 double compute_cost = NumDims * (2 * TensorOpCost::AddCost<Index>() + 214 2 * TensorOpCost::MulCost<Index>() + 215 TensorOpCost::DivCost<Index>()); 218 compute_cost += 2 * TensorOpCost::AddCost<Index>(); 222 TensorOpCost(0, 0, compute_cost, false /* vectorized */, PacketSize);
|
D | TensorArgMax.h | 115 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 117 return m_impl.costPerCoeff(vectorized) + TensorOpCost(0, 0, 1); 257 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 260 … (m_return_dim < 0 ? 0.0 : (TensorOpCost::ModCost<Index>() + TensorOpCost::DivCost<Index>())); 262 m_impl.costPerCoeff(vectorized) + TensorOpCost(0, 0, compute_cost);
|
D | TensorChipping.h | 244 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 251 cost += TensorOpCost::MulCost<Index>() + TensorOpCost::AddCost<Index>(); 256 cost += TensorOpCost::AddCost<Index>(); 258 cost += 3 * TensorOpCost::MulCost<Index>() + TensorOpCost::DivCost<Index>() + 259 3 * TensorOpCost::AddCost<Index>(); 263 TensorOpCost(0, 0, cost, vectorized, PacketSize);
|
D | TensorGenerator.h | 148 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 152 return TensorOpCost(0, 0, TensorOpCost::AddCost<Scalar>() + 153 TensorOpCost::MulCost<Scalar>());
|
D | TensorShuffling.h | 180 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 181 const double compute_cost = NumDims * (2 * TensorOpCost::AddCost<Index>() + 182 2 * TensorOpCost::MulCost<Index>() + 183 TensorOpCost::DivCost<Index>()); 185 TensorOpCost(0, 0, compute_cost, false /* vectorized */, PacketSize);
|
D | TensorAssign.h | 153 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 158 TensorOpCost left = m_leftImpl.costPerCoeff(vectorized); 160 TensorOpCost( 163 TensorOpCost(0, sizeof(CoeffReturnType), 0, vectorized, PacketSize);
|
D | TensorPatch.h | 248 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 249 const double compute_cost = NumDims * (TensorOpCost::DivCost<Index>() + 250 TensorOpCost::MulCost<Index>() + 251 2 * TensorOpCost::AddCost<Index>()); 253 TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|
D | TensorStriding.h | 211 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 212 double compute_cost = (NumDims - 1) * (TensorOpCost::AddCost<Index>() + 213 TensorOpCost::MulCost<Index>() + 214 TensorOpCost::DivCost<Index>()) + 215 TensorOpCost::MulCost<Index>(); 222 TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|
D | TensorConvolution.h | 451 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 456 TensorOpCost::AddCost<Scalar>() + TensorOpCost::MulCost<Scalar>(); 459 (2 * TensorOpCost::AddCost<Index>() + 2 * TensorOpCost::MulCost<Index>() + 460 TensorOpCost::DivCost<Index>()); 461 return TensorOpCost(0, 0, firstIndex_compute_cost, vectorized, PacketSize) + 464 TensorOpCost(0, 0, convolve_compute_cost, vectorized, 1065 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 1072 TensorOpCost::AddCost<Scalar>() + TensorOpCost::MulCost<Scalar>(); 1075 (2 * TensorOpCost::AddCost<Index>() + 2 * TensorOpCost::MulCost<Index>() + 1076 TensorOpCost::DivCost<Index>()); [all …]
|
D | TensorConcatenation.h | 263 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 265 const double compute_cost = NumDims * (2 * TensorOpCost::AddCost<Index>() + 266 2 * TensorOpCost::MulCost<Index>() + 267 TensorOpCost::DivCost<Index>() + 268 TensorOpCost::ModCost<Index>()); 275 TensorOpCost(0, 0, compute_cost);
|
D | TensorConversion.h | 232 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 234 const double cast_cost = TensorOpCost::CastCost<SrcType, TargetType>(); 241 TensorOpCost(0, 0, TgtCoeffRatio * (cast_cost / PacketSize)); 243 return m_impl.costPerCoeff(vectorized) + TensorOpCost(0, 0, cast_cost);
|
D | TensorCustomOp.h | 136 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 138 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize); 291 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 293 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize);
|
D | TensorImagePatch.h | 436 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 441 const double compute_cost = 3 * TensorOpCost::DivCost<Index>() + 442 6 * TensorOpCost::MulCost<Index>() + 443 8 * TensorOpCost::MulCost<Index>(); 445 TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|
D | TensorForcedEval.h | 147 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 148 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize);
|
D | TensorContractionThreadPool.h | 203 const TensorOpCost cost = 703 const TensorOpCost cost = 1008 TensorOpCost contractionCost(Index m, Index n, Index bm, Index bn, Index bk, 1027 TensorOpCost cost = TensorOpCost(0, 0, kd * computeBandwidth, true, packed_size); 1029 cost += TensorOpCost(0, sizeof(CoeffReturnType), 0, true, output_packet_size); 1037 TensorOpCost lhsCost = this->m_leftImpl.costPerCoeff(true) * (kd / n); 1038 TensorOpCost rhsCost = this->m_rightImpl.costPerCoeff(true) * (kd / m);
|
D | TensorEvalTo.h | 155 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 159 TensorOpCost(0, sizeof(CoeffReturnType), 0, vectorized, PacketSize);
|
D | TensorVolumePatch.h | 497 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost 500 10 * TensorOpCost::DivCost<Index>() + 21 * TensorOpCost::MulCost<Index>() + 501 8 * TensorOpCost::AddCost<Index>(); 502 return TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|
D | TensorDeviceThreadPool.h | 182 void parallelFor(Index n, const TensorOpCost& cost, in parallelFor() 266 void parallelFor(Index n, const TensorOpCost& cost, in parallelFor()
|
D | TensorScan.h | 188 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool) const { 189 return TensorOpCost(sizeof(CoeffReturnType), 0, 0);
|
D | TensorReduction.h | 258 const TensorOpCost cost = 260 TensorOpCost(0, 0, internal::functor_traits<Op>::Cost, Vectorizable, 654 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const { 656 return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize); 661 TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
|