ONE - On-device Neural Engine
|
Namespaces | |
namespace | bias_op |
namespace | cpu_backend_threadpool |
namespace | depthwise_conv_op |
namespace | detail |
namespace | eigen_support |
namespace | functor |
namespace | gemm_support |
namespace | multithreaded |
namespace | optimized |
namespace | optimized_integer_ops |
namespace | random |
namespace | reference |
namespace | reference_integer_ops |
namespace | ruy_support |
namespace | train |
namespace | training_ops |
namespace | xent_ops |
Typedefs | |
template<typename Scalar > | |
using | VectorMap = typename std::conditional< std::is_const< Scalar >::value, Eigen::Map< const Eigen::Matrix< typename std::remove_const< Scalar >::type, Eigen::Dynamic, 1 > >, Eigen::Map< Eigen::Matrix< Scalar, Eigen::Dynamic, 1 > > >::type |
template<typename Scalar > | |
using | MatrixMap = typename std::conditional< std::is_const< Scalar >::value, Eigen::Map< const Eigen::Matrix< typename std::remove_const< Scalar >::type, Eigen::Dynamic, Eigen::Dynamic > >, Eigen::Map< Eigen::Matrix< Scalar, Eigen::Dynamic, Eigen::Dynamic > > >::type |
template<typename T > | |
using | ComparisonFn = bool(*)(T, T) |
using | ShapeVec = std::vector< int32_t > |
using | Labels = std::vector< int32_t > |
using | OperandLabels = std::vector< Labels > |
using | LabelCounts = std::vector< int32_t > |
using | OperandLabelCounts = std::vector< LabelCounts > |
using | LabelToDimSizes = std::vector< int32_t > |
typedef Eigen::ThreadPoolDevice | CPUDevice |
typedef TTypes< float, 1 >::Tensor32Bit::Index | Index32 |
Enumerations | |
enum | DimensionType { kBroadcasting = 0 , kBatch = 1 , kFree = 2 , kContract = 3 , kReduce = 4 } |
enum class | FusedActivationFunctionType { kNone = 0 , kRelu6 = 1 , kRelu1 = 2 , kRelu = 3 , kTanh = 4 , kSigmoid = 6 } |
enum class | PaddingType { kNone = 0 , kSame = 1 , kValid = 2 } |
enum class | BinaryArithmeticOpType { ADD = 0 , SUB = 1 , MUL = 2 , DIV = 3 , POW = 4 } |
enum class | ComparisonOpType { Equal , NotEqual , Greater , GreaterEqual , Less , LessEqual } |
enum class | RoPEMode { kGptNeox = 0 , kGptJ = 1 } |
enum class | BroadcastableOpCategory : uint8_t { kNone , kNonBroadcast , kFirstInputBroadcastsFast , kSecondInputBroadcastsFast , kGenericBroadcast } |
enum | LSTMKernelType { kTfLiteLSTMFullKernel = 0 , kTfLiteLSTMBasicKernel } |
enum class | Order { kColMajor , kRowMajor } |
enum class | CachePolicy : std::uint8_t { kNeverCache , kCacheIfLargeSpeedup , kAlwaysCache } |
enum class | QuantizationFlavor { kFloatingPoint , kIntegerWithUniformMultiplier , kIntegerWithPerRowMultiplier } |
Functions | |
template<typename Scalar > | |
VectorMap< Scalar > | MapAsVector (Scalar *data, const Shape &shape) |
template<typename Scalar > | |
MatrixMap< Scalar > | MapAsMatrixWithLastDimAsRows (Scalar *data, const Shape &shape) |
template<typename T > | |
void | AddN (const Shape &input_shape, const size_t num_inputs, const T **input_data, T *output_data) |
template<typename T1 , typename T2 , typename Cmp > | |
void | ArgMinMax (const Shape &input1_shape, const T1 *input1_data, const Shape &output_shape, T2 *output_data, int32_t axis, const Cmp &cmp) |
template<typename T > | |
void | AveragePool (const PoolParams &, const Shape &, const T *, const Shape &, T *) |
template<> | |
void | AveragePool< float > (const PoolParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | AveragePool16 (const PoolParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
void | AveragePool32 (const PoolParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
template<> | |
void | AveragePool< uint8_t > (const PoolParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
template<> | |
void | AveragePool< int8_t > (const PoolParams ¶ms, const Shape &input_shape, const int8_t *input_data, const Shape &output_shape, int8_t *output_data) |
void | GetIndexRange (int spatial_index_dim, int block_shape_dim, int input_dim, int output_dim, int *start_index, int *end_index) |
template<typename T > | |
void | BatchToSpaceND (const Shape &unextended_input1_shape, const T *input1_data, const int32_t *block_shape_data, const int32_t *crops_data, const Shape &unextended_output_shape, T *output_data) |
bool | ProcessBroadcastShapes (const Shape &shape0, const Shape &shape1, BinaryArithmeticOpParam *params) |
template<BinaryArithmeticOpType op_type, typename T > | |
std::enable_if_t<!is_quant8< T >::value &&!std::is_same< T, bool >::value > | BinaryArithmeticOp (const BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<BinaryArithmeticOpType op_type, typename T > | |
std::enable_if_t<!is_quant8< T >::value &&std::is_same< T, bool >::value > | BinaryArithmeticOp (const BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<BinaryArithmeticOpType op_type, typename T > | |
std::enable_if_t< is_quant8< T >::value > | BinaryArithmeticOp (const BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<BinaryArithmeticOpType op_type> | |
void | BinaryArithmeticOp (const BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const float *input1_data, const Shape &input2_shape, const float *input2_data, const Shape &output_shape, float *output_data) |
template<BinaryArithmeticOpType op_type, typename T > | |
std::enable_if_t<!is_quant8< T >::value > | BroadcastBinaryArithmeticOp (BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<BinaryArithmeticOpType op_type, typename T > | |
std::enable_if_t< is_quant8< T >::value > | BroadcastBinaryArithmeticOp (BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<BinaryArithmeticOpType op_type> | |
void | BroadcastBinaryArithmeticOp (BinaryArithmeticOpParam ¶ms, const Shape &input1_shape, const float *input1_data, const Shape &input2_shape, const float *input2_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
void | BroadcastTo (const Shape &input_shape, T *input_data, const Shape &output_shape, T *output_data) |
void | BiasAndClamp (float clamp_min, float clamp_max, int bias_size, const float *bias_data, int array_size, float *array_data) |
template<typename T > | |
bool | EqualFn (T lhs, T rhs) |
template<typename T > | |
bool | NotEqualFn (T lhs, T rhs) |
template<typename T > | |
bool | GreaterFn (T lhs, T rhs) |
template<typename T > | |
bool | GreaterEqualFn (T lhs, T rhs) |
template<typename T > | |
bool | LessFn (T lhs, T rhs) |
template<typename T > | |
bool | LessEqualFn (T lhs, T rhs) |
template<typename T , ComparisonFn< T > F> | |
void | ComparisonImpl (const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, bool *output_data) |
template<ComparisonFn< float > F> | |
void | Comparison (const Shape &input1_shape, const float *input1_data, const Shape &input2_shape, const float *input2_data, const Shape &output_shape, bool *output_data) |
template<typename T , ComparisonFn< int32_t > F> | |
void | ComparisonWithScaling (ComparisonParams ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, bool *output_data) |
template<typename T , ComparisonFn< T > F> | |
void | BroadcastComparison4DSlowImpl (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, bool *output_data) |
template<typename T , ComparisonFn< T > F> | |
void | BroadcastComparison4DSlow (const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, bool *output_data) |
template<typename T , ComparisonFn< int32_t > F> | |
void | BroadcastComparison4DSlowWithScaling (ComparisonParams ¶ms, const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, bool *output_data) |
TFLITE_COMPARISON_OP (Equal) | |
TFLITE_COMPARISON_OP (NotEqual) | |
TFLITE_COMPARISON_OP (Greater) | |
TFLITE_COMPARISON_OP (GreaterEqual) | |
TFLITE_COMPARISON_OP (Less) | |
TFLITE_COMPARISON_OP (LessEqual) | |
template<typename Scalar > | |
void | Concatenation (const ConcatenationParams ¶ms, const Shape *const *input_shapes, const Scalar *const *input_data, const Shape &output_shape, Scalar *output_data) |
void | ConcatenationWithScaling (const ConcatenationParams ¶ms, const Shape *const *input_shapes, const uint8_t *const *input_data, const Shape &output_shape, uint8_t *output_data) |
template<typename T > | |
void | DepthToSpace (const Shape &unextended_input_shape, const T *input_data, const Shape &unextended_output_shape, T *output_data, int32_t block_size) |
int | HowManyConvThreads (const Shape &output_shape, const Shape &filter_shape) |
bool | MultithreadAlongBatches (int thread_count, int batches) |
template<typename T , typename TS > | |
void | DepthwiseConv (const DepthwiseConvParams ¶ms, const Shape &input_shape, const T *input_data, const Shape &filter_shape, const T *filter_data, const Shape &bias_shape, const TS *bias_data, const Shape &output_shape, T *output_data, ruy::Context *ruy_context) |
void | DepthwiseConvOp (const DepthwiseConvParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &filter_shape, const float *filter_data, const Shape &bias_shape, const float *bias_data, float *padded_filter_data, bool pad_filter, float *filter_buffers_data, const Shape &output_shape, float *output_data) |
void | Dequantize (const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, float *output_data, const float scale, const int32_t zero_point) |
void | Dequantize (const Shape &input_shape, const int8_t *input_data, const Shape &output_shape, float *output_data, const float scale, const int32_t zero_point) |
void | Dequantize (const Shape &input_shape, const int16_t *input_data, const Shape &output_shape, float *output_data, const float scale, const int32_t zero_point) |
void | Sin (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Cos (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Abs (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Rsqrt (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
void | Neg (const Shape &input_shape, const T *input_data, const Shape &output_shape, T *output_data) |
void | Log (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Floor (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Sqrt (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Square (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ELU (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Erf (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | Exp (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
void | Fill (const T *value_data, const Shape &output_shape, T *output_data) |
template<typename T > | |
void | FloorDivBroadcast (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | FloorDivElementwise (const Shape &shape, const T *input1_data, const T *input2_data, T *output_data) |
template<typename T > | |
void | FloorModBroadcast (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | FloorModElementwise (const Shape &shape, const T *input1_data, const T *input2_data, T *output_data) |
void | FullyConnected (const FullyConnectedParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &weights_shape, const float *weights_data, const Shape &, const float *bias_data, const Shape &, float *output_data) |
void | FullyConnected (const FullyConnectedParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &filter_shape, const uint8_t *filter_data, const Shape &bias_shape, const int32_t *bias_data, const Shape &output_shape, uint8_t *output_data) |
void | FullyConnectedHybrid (const FullyConnectedParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &filter_shape, const int8_t *filter_data, const Shape &, const float *bias_data, const Shape &output_shape, float *output_data, FCTempArena &temp_arena, ruy::Context *ruy_context) |
void | FullyConnectedSparseWeightRandom (const FullyConnectedParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &weights_shape, const float *weights_data, const Shape &bias_shape, const float *bias_data, const Shape &output_shape, float *output_data, const uint16_t *w1_segments, const uint16_t *w1_indices) |
void | FullyConnectedSparseWeight16x1 (const FullyConnectedParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &weights_shape, const float *weights_data, const Shape &bias_shape, const float *bias_data, const Shape &output_shape, float *output_data, const uint16_t *w1_segments, const uint16_t *w1_indices) |
template<typename T , typename CoordsT = int32_t> | |
void | Gather (const GatherParams &op_params, const Shape &input_shape, const T *input_data, const Shape &coords_shape, const CoordsT *coords_data, const Shape &, T *output_data) |
void | ComputeBatchIndices (const int32_t output_batch_size, const std::vector< int32_t > &reshape, const std::vector< int32_t > &bcast, std::vector< int32_t > *out_indices) |
template<typename DSizes > | |
Eigen::DSizes< Index32, DSizes::count > | To32BitDims (const DSizes &in) |
template<typename TensorType > | |
TTypes< typenameTensorType::Scalar, TensorType::NumIndices >::Tensor32Bit | To32Bit (TensorType in) |
void | InstanceNorm (const InstanceNormParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &gamma_shape, const float *gamma_data, const Shape &beta_shape, const float *beta_data, const Shape &output_shape, float *output_data) |
void | L2NormalizeFloat32 (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | L2NormalizeQuant8 (L2NormParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
void | LeakyReLU (const LeakyReluParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
void | LogicalAndBroadcast (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | LogicalAndElementwise (const Shape &shape, const T *input1_data, const T *input2_data, T *output_data) |
void | LogicalNot (const Shape &input_shape, const bool *input_data, const Shape &output_shape, bool *output_data) |
template<typename T > | |
void | LogicalOrBroadcast (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | LogicalOrElementwise (const Shape &shape, const T *input1_data, const T *input2_data, T *output_data) |
void | Logistic (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | LogSoftmax (const SoftmaxParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | LogSoftmax (const SoftmaxParams ¶ms, float input_scale, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
void | CalculateLstmGateFloat (const float *input, const float *input_to_gate_weights, const float *aux_input, const float *aux_input_to_gate_weights, const float *output_state, const float *recurrent_to_gate_weights, const float *cell_state, const float *cell_to_gate_weights, const float *layer_norm_coefficients, const float *gate_bias, const int n_batch, const int n_input, const int n_aux_input, const int n_output, const int n_cell, const FusedActivationFunctionType activation, float *gate, const bool is_input_all_zeros, const bool is_aux_input_all_zeros) |
void | UpdateLstmCellFloat (int n_batch, int n_cell, float *cell_state, const float *input_gate, float *forget_gate, const float *cell_gate, bool use_cifg, float clip) |
void | CalculateLstmOutputFloat (int n_batch, int n_cell, int n_output, const float *cell_state, const float *output_gate, FusedActivationFunctionType activation, const float *projection_weights, const float *projection_bias, const float proj_clip, float *output_state, float *scratch) |
void | LstmStepFloat (const float *input_ptr, const float *input_to_input_weights_ptr, const float *input_to_forget_weights_ptr, const float *input_to_cell_weights_ptr, const float *input_to_output_weights_ptr, const float *aux_input_ptr, const float *aux_input_to_input_weights_ptr, const float *aux_input_to_forget_weights_ptr, const float *aux_input_to_cell_weights_ptr, const float *aux_input_to_output_weights_ptr, const float *recurrent_to_input_weights_ptr, const float *recurrent_to_forget_weights_ptr, const float *recurrent_to_cell_weights_ptr, const float *recurrent_to_output_weights_ptr, const float *cell_to_input_weights_ptr, const float *cell_to_forget_weights_ptr, const float *cell_to_output_weights_ptr, const float *input_layer_norm_coefficients_ptr, const float *forget_layer_norm_coefficients_ptr, const float *cell_layer_norm_coefficients_ptr, const float *output_layer_norm_coefficients_ptr, const float *input_gate_bias_ptr, const float *forget_gate_bias_ptr, const float *cell_gate_bias_ptr, const float *output_gate_bias_ptr, const float *projection_weights_ptr, const float *projection_bias_ptr, const LSTMParams *params, int n_batch, int n_cell, int n_input, int n_aux_input, int n_output, int output_batch_leading_dim, float *output_state_ptr, float *cell_state_ptr, float *scratch0, float *scratch1, float *scratch2, float *scratch3, float *output_ptr) |
template<typename T > | |
void | MatrixBandPart (const T num_lower_diags, const T num_upper_diags, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T , typename Op > | |
void | MaximumMinimumBroadcast4DSlow (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data, Op op) |
template<typename T > | |
void | Max (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | Min (const Shape &unextended_input1_shape, const T *input1_data, const Shape &unextended_input2_shape, const T *input2_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | MaxPool (const PoolParams &, const Shape &, const T *, const Shape &, T *) |
template<> | |
void | MaxPool< float > (const PoolParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<> | |
void | MaxPool< uint8_t > (const PoolParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
template<typename T , typename TI > | |
void | OneHot (const int32_t depth, const T on_value, const T off_value, int32_t axis, const Shape &indices_shape, const TI *indices_data, const Shape &, T *output_data) |
template<typename Scalar > | |
void | Pack (const PackParams ¶ms, const Scalar *const *input_data, const Shape &output_shape, Scalar *output_data) |
template<typename T > | |
void | Pad (const int32_t *padding_data, int32_t pad_rank, const Shape &input_shape, const T *input_data, const Shape &output_shape, T *output_data, const T *constant_value_data) |
template<typename T > | |
void | powImpl (const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
template<typename InputT , typename OutputT > | |
void | Quantize (const Shape &input_shape, const InputT *input_data, const Shape &output_shape, OutputT *output_data, const float output_scale, const int32_t output_offset) |
template<> | |
void | Quantize (const Shape &input_shape, const float *input_data, const Shape &output_shape, int8_t *output_data, const float scale, const int32_t zero_point) |
template<> | |
void | Quantize (const Shape &input_shape, const float *input_data, const Shape &output_shape, uint8_t *output_data, const float scale, const int32_t zero_point) |
template<> | |
void | Quantize (const Shape &input_shape, const float *input_data, const Shape &output_shape, int16_t *output_data, const float scale, const int32_t zero_point) |
void | Quantize (const int32_t *multiplier, const int32_t *shift, int32_t channel_size, int32_t total_size, int32_t output_zp, int32_t output_min, int32_t output_max, int32_t *scratch, int8_t *output) |
template<typename input_type , typename output_type > | |
void | Requantize (const input_type *input_data, int32_t size, int32_t effective_scale_multiplier, int32_t effective_scale_shift, int32_t input_zeropoint, int32_t output_zeropoint, output_type *output_data) |
template<> | |
void | Requantize< uint8_t, int8_t > (const uint8_t *input_data, int32_t size, int32_t effective_scale_multiplier, int32_t effective_scale_shift, int32_t input_zeropoint, int32_t output_zeropoint, int8_t *output_data) |
template<> | |
void | Requantize< int8_t, uint8_t > (const int8_t *input_data, int32_t size, int32_t effective_scale_multiplier, int32_t effective_scale_shift, int32_t input_zeropoint, int32_t output_zeropoint, uint8_t *output_data) |
template<typename T > | |
int | GetSize (T start, T limit, T delta) |
template<typename T > | |
void | Range (const T *start_data, const T *limit_data, const T *delta_data, T *output_data) |
template<typename In , typename Out > | |
bool | ReduceImpl (const In *input_data, const Shape &input_shape, const Shape &, const int *axis, const int num_axis, int *input_iter, Out reducer(const Out current, const In in), Out *output_data) |
bool | ResolveAxis (const int num_dims, const std::vector< int > &axes, int *out_axis, int *out_num_axis) |
template<typename T > | |
bool | InitTensorDataForReduce (const Shape &shape, const T init_value, T *data) |
float | round_nearest (float value) |
template<typename Out , typename In > | |
Out | mean_reducer (const Out data1, const In data2, int normalizer) |
template<typename In > | |
int | sum_reducer (const int data1, const In data2) |
template<typename In , typename Out > | |
bool | ReduceMeanImpl (const In *input_data, const Shape &input_shape, const int *axis, const int num_axis, int *input_iter, Out reducer(const Out current, const In in, int normalizer), Out *output_data) |
template<typename In > | |
size_t | ReduceSumQuantImpl (const In *input_data, const Shape &input_shape, const int *axis, const int num_axis, int *input_iter, int reducer(const int current, const In in), int *temp_sum) |
template<typename In , typename Out > | |
void | Mean (const Shape &input_shape, const In *input_data, const Shape &output_shape, Out *output_data, const std::vector< int > &axes) |
template<typename In , typename Out > | |
void | MeanQ8Asymm (const Shape &input_shape, const In *input_data, float input_scale, int32_t input_offset, const Shape &output_shape, Out *output_data, float output_scale, int32_t output_offset, const std::vector< int > &axes) |
template<typename In , typename Out > | |
void | MeanAxis1And2 (const Shape &input_shape, const In *input_data, const Shape &output_shape, Out *output_data) |
void | ReLU (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ReLU6 (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ResizeBilinearKernel2x2 (int32_t x0, int32_t x1, int32_t y0, int32_t y1, int32_t x, int32_t y, int32_t depth, int32_t batch, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ResizeBilinear2x2 (int32_t batches, int32_t input_height, int32_t input_width, int32_t depth, int32_t output_height, int32_t output_width, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ResizeBilinearKernel (const float *input_ptr, int32_t depth, float scale, float *output_ptr) |
void | ComputeInterpolationValues (const float value, const float scale, const bool half_pixel_centers, int32_t input_size, float *scaled_value, int32_t *lower_bound, int32_t *upper_bound) |
void | ResizeBilinearGeneric (int32_t batches, int32_t input_height, int32_t input_width, int32_t depth, int32_t output_height, int32_t output_width, float height_scale, float width_scale, const Shape &input_shape, const float *input_data, float *output_data, const bool half_pixel_centers) |
template<typename T > | |
void | ResizeBilinearGenericSmallChannel (int32_t batches, int32_t input_height, int32_t input_width, int32_t depth, int32_t output_height, int32_t output_width, float height_scale, float width_scale, const Shape &input_shape, const T *input_data, T *output_data, const bool half_pixel_centers) |
void | ResizeBilinear (ResizeBilinearParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
void | ResizeBilinear (ResizeBilinearParams ¶ms, const Shape &input_shape, const uint8_t *input_data, const Shape &output_shape, uint8_t *output_data) |
void | ComputeInterpolationValues (const int32_t value, const int32_t scale_10, const bool half_pixel_centers, int32_t input_size, int32_t *scaled_value, int32_t *lower_bound, int32_t *upper_bound) |
void | ResizeBilinear (const ResizeBilinearParams &op_params, const Shape &unextended_input_shape, const int8_t *input_data, const Shape &unextended_output_shape, int8_t *output_data) |
template<typename Scalar > | |
void | Reverse (int axis, const Shape &input_shape, const Scalar *input_data, const Shape &, Scalar *output_data) |
void | RmsNorm (const RmsNormParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &gamma_shape, const float *gamma_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
void | RoPE (const RoPEMode mode, const Shape &input_shape, const T *input_data, const Shape &sin_table_shape, const T *sin_table_data, const Shape &cos_table_shape, const T *cos_table_data, const Shape &output_shape, T *output_data) |
float | RoundToNearest (float value) |
void | Round (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename D , typename T > | |
void | Select (const Shape &input_condition_shape, const D *input_condition_data, const Shape &input_x_shape, const T *input_x_data, const Shape &input_y_shape, const T *input_y_data, const Shape &output_shape, T *output_data) |
template<typename D , typename T > | |
void | RankOneSelect (const Shape &input_condition_shape, const D *input_condition_data, const Shape &input_x_shape, const T *input_x_data, const Shape &input_y_shape, const T *input_y_data, const Shape &output_shape, T *output_data) |
template<typename D , typename T > | |
void | BroadcastSelect4DSlow (const Shape &input_condition_shape, const D *input_condition_data, const Shape &input_x_shape, const T *input_x_data, const Shape &input_y_shape, const T *input_y_data, const Shape &output_shape, T *output_data) |
template<typename T > | |
void | Slice (const SliceParams &op_params, const Shape &input_shape, SequentialTensorWriter< T > *writer) |
template<typename T > | |
void | Slice (const SliceParams &op_params, const Shape &input_shape, const T *input_data, T *output_data) |
void | Softmax (const float *in, const int input_size, const int batch_size, const float beta, float *out) |
void | Softmax (const SoftmaxParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T > | |
int32_t | QuantizeSoftmaxOutput (float prob_rescaled, int32_t zero_point) |
template<> | |
int32_t | QuantizeSoftmaxOutput< uint8_t > (float prob_rescaled, int32_t) |
void | PopulateSoftmaxLookupTable (float *table, float input_scale, float beta) |
template<typename In , typename Out > | |
void | Softmax (const SoftmaxParams ¶ms, const Shape &input_shape, const In *input_data, const Shape &output_shape, Out *output_data) |
template<typename T > | |
void | SpaceToBatchND (const SpaceToBatchParams ¶ms, const Shape &unextended_input_shape, const T *input_data, const Shape &unextended_block_shape_shape, const int32_t *block_shape_data, const Shape &unextended_padding_shape, const int32_t *paddings_data, const Shape &unextended_output_shape, T *output_data) |
template<typename T > | |
void | SpaceToDepth (const SpaceToDepthParams ¶ms, const Shape &unextended_input_shape, const T *input_data, const Shape &unextended_output_shape, T *output_data) |
template<typename Scalar > | |
void | Split (const SplitParams ¶ms, const Shape &input_shape, const Scalar *input_data, const Shape &output_shape, Scalar *const *output_data) |
template<typename Scalar > | |
void | SplitV (const SplitVParams ¶ms, const Shape &input_shape, const Scalar *input_data, std::vector< nnfw::cker::Shape > &output_shapes, Scalar *const *output_data) |
template<typename T , int N> | |
void | SqDiffImpl (const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data, NdArrayDesc< N > *desc1_in, NdArrayDesc< N > *desc2_in, NdArrayDesc< N > *desc_out) |
template<typename T > | |
void | SqDiff (const Shape &input1_shape, const T *input1_data, const Shape &input2_shape, const T *input2_data, const Shape &output_shape, T *output_data) |
void | GenerateKey (Tensor seed, random::PhiloxRandom::Key *out_key, random::PhiloxRandom::ResultType *out_counter) |
template<typename Device , class Distribution > | |
void | Fill (random::PhiloxRandom random, Tensor *output) |
void | StatelessRandomUniform (const Shape &shape_shape, const int32_t *shape_data, const Shape &seed_shape, const int32_t *seed_data, const Shape &output_shape, float *output_data) |
int | Clamp (const int v, const int lo, const int hi) |
void | StridedSlicePadIndices (StridedSliceParams *p, int dim_count) |
int | StartForAxis (const StridedSliceParams ¶ms, const Shape &input_shape, int axis) |
int | StopForAxis (const StridedSliceParams ¶ms, const Shape &input_shape, int axis, int start_for_axis) |
bool | LoopCondition (int index, int stop, int stride) |
template<typename T > | |
StridedSliceParams | buildStridedSliceParams (const T *begin, const T *end, const T *strides, const uint32_t begin_mask, const uint32_t end_mask, const uint32_t shrink_axis_mask, const uint8_t rank) |
void | checkOutputSize (const StridedSliceParams &op_params, const Shape &input_shape, const Shape &output_shape, uint32_t rank) |
template<typename T > | |
void | StridedSlice (const StridedSliceParams &op_params, const Shape &unextended_input_shape, const T *input_data, const Shape &unextended_output_shape, T *output_data) |
void | Tanh (const Shape &input_shape, const float *input_data, const Shape &output_shape, float *output_data) |
template<typename T , typename M > | |
void | CopyMultipleTimes (const T *in_data, int32_t in_size, M multiplier, T *out_data) |
template<typename T , typename M > | |
std::pair< int, int > | TileOneDimension (const Shape &in_dimensions, const T *in_data, const M *multipliers, T *out_data, int dimension) |
template<typename T > | |
void | Transpose2D (const Shape &input_shape, const T *input_data, const Shape &output_shape, T *output_data) |
template<typename T > | |
void | Transpose3D (const TransposeParams ¶ms, const Shape &input_shape, const T *input_data, const Shape &, T *output_data) |
template<typename T > | |
void | TransposeImpl (const TransposeParams ¶ms, const Shape &input_shape, const T *input_data, const Shape &output_shape, T *output_data) |
template<typename T > | |
void | Transpose (const TransposeParams &unshrunk_params, const Shape &unshrunk_input_shape, const T *input_data, const Shape &unshrunk_output_shape, T *output_data) |
void | TransposeConv (const TransposeConvParams ¶ms, const Shape &input_shape, const float *input_data, const Shape &filter_shape, const float *filter_data, const Shape &output_shape, float *output_data) |
template<typename Scalar > | |
void | Unpack (const UnpackParams ¶ms, const Shape &input_shape, const Scalar *input_data, const Shape &output_shape, Scalar *const *output_datas) |
template<typename T > | |
void | PortableCwiseClipping (T *vector, const int v_size, const T clipping_value) |
void | PortableVectorBatchVectorAssign (const float *vector, int v_size, int n_batch, float *batch_vector) |
void | PortableVectorBatchVectorAdd (const float *vector, int v_size, int n_batch, float *batch_vector) |
bool | PortableIsZeroVector (const float *vector, int v_size) |
void | PortableApplyActivationToVector (const float *vector, int v_size, FusedActivationFunctionType activation, float *result) |
void | PortableSub1Vector (const float *vector, int v_size, float *result) |
void | PortableSymmetricQuantizeFloats (const float *values, const int size, int8_t *quantized_values, float *min_value, float *max_value, float *scaling_factor) |
void | PortableAsymmetricQuantizeFloats (const float *values, const int size, int8_t *quantized_values, float *scaling_factor, int32_t *offset) |
void | PortableMatrixBatchVectorMultiplyAccumulate (const int8_t *__restrict__ matrix, const int m_rows, const int m_cols, const int8_t *__restrict__ vectors, const float *scaling_factors, int n_batch, float *__restrict__ result, int result_stride) |
void | PortableMatrixBatchVectorMultiplyAccumulate (const int8_t *__restrict__ matrix, const int m_rows, const int m_cols, const int8_t *__restrict__ vector, const float *scaling_factors, int n_batch, int32_t *, float *__restrict__ result, int result_stride, ruy::Context *) |
void | PortableMatrixBatchVectorMultiplyAccumulate (const float *matrix, int m_rows, int m_cols, const float *vector, int n_batch, float *result, int result_stride) |
void | PortableMeanStddevNormalization (const float *input_vector, float *output_vector, int v_size, int n_batch) |
void | PortableZeroVector (float *vector, int v_size) |
int | MatchingDim (const Shape &shape1, int index1, const Shape &shape2, int index2) |
template<typename... Args> | |
int | MatchingDim (const Shape &shape1, int index1, const Shape &shape2, int index2, Args... args) |
Shape | GetShape (const std::vector< int32_t > &data) |
int | Offset (const Shape &shape, int i0, int i1, int i2, int i3) |
int | Offset (const Shape &shape, int *index) |
int | FlatSizeSkipDim (const Shape &shape, int skip_dim) |
template<typename... Ts> | |
bool | checkMatching (const Shape &shape, Ts... check_shapes) |
template<typename... Ts> | |
int | MatchingFlatSize (const Shape &shape, Ts... check_shapes) |
int | MatchingFlatSizeSkipDim (const Shape &shape, int skip_dim, const Shape &check_shape_0) |
int | MatchingFlatSizeSkipDim (const Shape &shape, int skip_dim, const Shape &check_shape_0, const Shape &check_shape_1) |
int | MatchingElementsSize (const Shape &shape, const Shape &check_shape_0, const Shape &check_shape_1) |
ShapeIterator | begin (const Shape &s) |
ShapeIterator | end (const Shape &s) |
void | CwiseClipping (float *vector, const int v_size, const float clipping_value) |
void | VectorBatchVectorAdd (const float *vector, int v_size, int n_batch, float *batch_vector) |
void | VectorBatchVectorAssign (const float *vector, int v_size, int n_batch, float *batch_vector) |
template<typename T > | |
void | VectorVectorCwiseProduct (const T *__restrict__ vector1, const T *__restrict__ vector2, int v_size, T *__restrict__ result) |
template<typename T > | |
void | VectorVectorCwiseProductAccumulate (const T *__restrict__ vector1, const T *__restrict__ vector2, int v_size, T *__restrict__ result) |
template<typename T > | |
void | VectorBatchVectorCwiseProduct (const T *vector, int v_size, const T *batch_vector, int n_batch, T *result) |
template<typename T > | |
void | VectorBatchVectorCwiseProductAccumulate (const T *vector, int v_size, const T *batch_vector, int n_batch, T *result) |
bool | IsZeroVector (const float *vector, int v_size) |
void | ApplyActivationToVector (const float *vector, int v_size, FusedActivationFunctionType activation, float *result) |
void | Sub1Vector (const float *vector, int v_size, float *result) |
void | SymmetricQuantizeFloats (const float *values, const int size, int8_t *quantized_values, float *min, float *max, float *scaling_factor) |
void | MatrixBatchVectorMultiplyAccumulate (const int8_t *matrix, const int m_rows, const int m_cols, const int8_t *vector, const float *scaling_factors, int n_batch, float *result, int result_stride) |
void | MatrixBatchVectorMultiplyAccumulate (const float *matrix, int m_rows, int m_cols, const float *vector, int n_batch, float *result, int result_stride) |
void | MatrixBatchVectorMultiplyAccumulate (const int8_t *matrix, const int m_rows, const int m_cols, const int8_t *vectors, const float *scaling_factors, int n_batch, int32_t *scratch, float *result, int result_stride, ruy::Context *ruy_context) |
void | MeanStddevNormalization (const float *input_vector, float *output_vector, int v_size, int n_batch) |
void | ZeroVector (float *vector, int v_size) |
template<typename AccumScalar , typename DstScalar , QuantizationFlavor quantization_flavor> | |
void | ValidateGemmParams (const GemmParams< AccumScalar, DstScalar, quantization_flavor > ¶ms) |
template<typename T > | |
T | ActivationFunctionWithMinMax (T x, T output_activation_min, T output_activation_max) |
void | QuantizeMultiplier (double double_multiplier, int32_t *quantized_multiplier, int *shift) |
void | QuantizeMultiplierSmallerThanOneExp (double double_multiplier, int32_t *quantized_multiplier, int *left_shift) |
int32_t | MultiplyByQuantizedMultiplier (int32_t x, int32_t quantized_multiplier, int shift) |
int32_t | MultiplyByQuantizedMultiplierGreaterThanOne (int32_t x, int32_t quantized_multiplier, int left_shift) |
int32_t | MultiplyByQuantizedMultiplierSmallerThanOneExp (int32_t x, int32_t quantized_multiplier, int left_shift) |
int | NodeOffset (int b, int h, int w, int height, int width) |
int | CountLeadingZeros (uint32_t integer_input) |
void | GetInvSqrtQuantizedMultiplierExp (int32_t input, int reverse_shift, int32_t *output_inv_sqrt, int *output_shift) |
int | SubscriptToIndex (const NdArrayDesc< 4 > &desc, int i0, int i1, int i2, int i3) |
template<int N> | |
int | SubscriptToIndexGeneric (const NdArrayDesc< N > *desc, int *iter) |
template<int N> | |
void | CopyDimsToDesc (const Shape &input_shape, NdArrayDesc< N > *desc_out) |
template<int N> | |
void | NdArrayDescsForElementwiseBroadcast (const Shape &input0_shape, const Shape &input1_shape, NdArrayDesc< N > *desc0_out, NdArrayDesc< N > *desc1_out) |
template<int N> | |
void | NdArrayDescsForElementwiseBroadcast (const Shape &input0_shape, const Shape &input1_shape, const Shape &input2_shape, NdArrayDesc< N > *desc0_out, NdArrayDesc< N > *desc1_out, NdArrayDesc< N > *desc2_out) |
bool | NextIndex (const int num_dims, const int *dims, int *current) |
size_t | ReducedOutputOffset (const int num_dims, const int *dims, const int *index, const int num_axis, const int *axis) |
template<typename T > | |
void | optimized_ops_preload_l1_keep (const T *ptr) |
std::ostream & | operator<< (std::ostream &os, const Shape &shape) |
using nnfw::cker::ComparisonFn = typedef bool (*)(T, T) |
Definition at line 37 of file Comparison.h.
typedef Eigen::ThreadPoolDevice nnfw::cker::CPUDevice |
Definition at line 51 of file RandomOpCpu.h.
typedef TTypes<float,1>::Tensor32Bit::Index nnfw::cker::Index32 |
using nnfw::cker::LabelCounts = typedef std::vector<int32_t> |
using nnfw::cker::Labels = typedef std::vector<int32_t> |
using nnfw::cker::LabelToDimSizes = typedef std::vector<int32_t> |
using nnfw::cker::MatrixMap = typedef typename std::conditional< std::is_const<Scalar>::value, Eigen::Map< const Eigen::Matrix<typename std::remove_const<Scalar>::type, Eigen::Dynamic, Eigen::Dynamic> >, Eigen::Map<Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> >>::type |
using nnfw::cker::OperandLabelCounts = typedef std::vector<LabelCounts> |
using nnfw::cker::OperandLabels = typedef std::vector<Labels> |
using nnfw::cker::ShapeVec = typedef std::vector<int32_t> |
using nnfw::cker::VectorMap = typedef typename std::conditional< std::is_const<Scalar>::value, Eigen::Map<const Eigen::Matrix<typename std::remove_const<Scalar>::type, Eigen::Dynamic, 1> >, Eigen::Map<Eigen::Matrix<Scalar, Eigen::Dynamic, 1> >>::type |
|
strong |
|
strong |
|
strong |
Enumerator | |
---|---|
kNeverCache | |
kCacheIfLargeSpeedup | |
kAlwaysCache |
Definition at line 425 of file Types.h.
|
strong |
Enumerator | |
---|---|
Equal | |
NotEqual | |
Greater | |
GreaterEqual | |
Less | |
LessEqual |
Definition at line 56 of file Types.h.
Enumerator | |
---|---|
kBroadcasting | |
kBatch | |
kFree | |
kContract | |
kReduce |
Definition at line 116 of file Einsum.h.
|
strong |
Enumerator | |
---|---|
kTfLiteLSTMFullKernel | |
kTfLiteLSTMBasicKernel |
|
strong |
|
strong |
|
strong |
Enumerator | |
---|---|
kFloatingPoint | |
kIntegerWithUniformMultiplier | |
kIntegerWithPerRowMultiplier |
Definition at line 474 of file Types.h.
|
strong |
|
inline |
Definition at line 51 of file Elementwise.h.
References MapAsVector(), and output_shape.
|
inline |
Definition at line 43 of file Utils.h.
Referenced by AveragePool< float >(), BiasAndClamp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::Conv(), nnfw::cker::reference_integer_ops::DepthwiseConvHybridPerChannel(), nnfw::cker::reference::HybridConvPerChannel(), InstanceNorm(), and MaxPool< float >().
void nnfw::cker::AddN | ( | const Shape & | input_shape, |
const size_t | num_inputs, | ||
const T ** | input_data, | ||
T * | output_data | ||
) |
Definition at line 29 of file AddN.h.
References nnfw::cker::Shape::FlatSize(), and size.
|
inline |
Definition at line 109 of file TensorUtils.h.
References PortableApplyActivationToVector().
Referenced by CalculateLstmGateFloat(), CalculateLstmOutputFloat(), FullyConnected(), FullyConnectedHybrid(), FullyConnectedSparseWeight16x1(), and FullyConnectedSparseWeightRandom().
void nnfw::cker::ArgMinMax | ( | const Shape & | input1_shape, |
const T1 * | input1_data, | ||
const Shape & | output_shape, | ||
T2 * | output_data, | ||
int32_t | axis, | ||
const Cmp & | cmp | ||
) |
Definition at line 29 of file ArgMinMax.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and output_shape.
void nnfw::cker::AveragePool | ( | const PoolParams & | , |
const Shape & | , | ||
const T * | , | ||
const Shape & | , | ||
T * | |||
) |
Definition at line 36 of file AveragePool.h.
|
inline |
Definition at line 106 of file AveragePool.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PaddingValues::height, MatchingDim(), Offset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::quantized_activation_max, nnfw::cker::PoolParams::quantized_activation_min, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
Referenced by AveragePool< uint8_t >().
|
inline |
Definition at line 242 of file AveragePool.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PaddingValues::height, MatchingDim(), Offset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::quantized_activation_max, nnfw::cker::PoolParams::quantized_activation_min, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
Referenced by AveragePool< uint8_t >().
void nnfw::cker::AveragePool< float > | ( | const PoolParams & | params, |
const Shape & | input_shape, | ||
const float * | input_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 44 of file AveragePool.h.
References ActivationFunctionWithMinMax(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PoolParams::float_activation_max, nnfw::cker::PoolParams::float_activation_min, nnfw::cker::PaddingValues::height, MapAsMatrixWithLastDimAsRows(), MatchingDim(), NodeOffset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
void nnfw::cker::AveragePool< int8_t > | ( | const PoolParams & | params, |
const Shape & | input_shape, | ||
const int8_t * | input_data, | ||
const Shape & | output_shape, | ||
int8_t * | output_data | ||
) |
Definition at line 399 of file AveragePool.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PaddingValues::height, MatchingDim(), Offset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::quantized_activation_max, nnfw::cker::PoolParams::quantized_activation_min, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
void nnfw::cker::AveragePool< uint8_t > | ( | const PoolParams & | params, |
const Shape & | input_shape, | ||
const uint8_t * | input_data, | ||
const Shape & | output_shape, | ||
uint8_t * | output_data | ||
) |
Definition at line 384 of file AveragePool.h.
References AveragePool16(), AveragePool32(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, and output_shape.
|
inline |
Definition at line 50 of file BatchToSpaceND.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), GetIndexRange(), Offset(), output_shape, nnfw::cker::Shape::SetDim(), and UNUSED.
|
inline |
Definition at line 88 of file ShapeIterator.h.
|
inline |
Definition at line 29 of file Common.h.
References ActivationFunctionWithMinMax().
Referenced by nnfw::cker::optimized::AddBiasAndEvalActivationFunction(), and nnfw::cker::detail::GemmImplUsingEigen::Run().
|
inline |
Definition at line 257 of file BinaryArithmeticOps.h.
References nnfw::cker::optimized::Add(), ADD, nnfw::cker::optimized::Div(), DIV, nnfw::cker::optimized::Mul(), MUL, output_shape, nnfw::cker::optimized::Sub(), and SUB.
|
inline |
Definition at line 213 of file BinaryArithmeticOps.h.
References nnfw::cker::reference::BinaryArithmeticOp(), and output_shape.
|
inline |
Definition at line 223 of file BinaryArithmeticOps.h.
References nnfw::cker::reference::BinaryArithmeticOp(), and output_shape.
|
inline |
Definition at line 233 of file BinaryArithmeticOps.h.
References nnfw::cker::optimized::Add(), ADD, DIV, nnfw::cker::optimized::Mul(), MUL, output_shape, and SUB.
|
inline |
Definition at line 325 of file BinaryArithmeticOps.h.
References ADD, nnfw::cker::optimized::BroadcastAddDispatch(), nnfw::cker::optimized::BroadcastDivDispatch(), nnfw::cker::optimized::BroadcastMulDispatch(), nnfw::cker::optimized::BroadcastSubDispatch(), DIV, MUL, output_shape, POW, and SUB.
|
inline |
Definition at line 289 of file BinaryArithmeticOps.h.
References nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), and output_shape.
|
inline |
Definition at line 300 of file BinaryArithmeticOps.h.
References ADD, nnfw::cker::optimized::BroadcastAddDispatch(), nnfw::cker::optimized::BroadcastMulDispatch(), DIV, MUL, output_shape, POW, and SUB.
|
inline |
Definition at line 123 of file Comparison.h.
References output_shape.
|
inline |
Definition at line 91 of file Comparison.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 132 of file Comparison.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::ComparisonParams::input1_multiplier, nnfw::cker::ComparisonParams::input1_offset, nnfw::cker::ComparisonParams::input1_shift, nnfw::cker::ComparisonParams::input2_multiplier, nnfw::cker::ComparisonParams::input2_offset, nnfw::cker::ComparisonParams::input2_shift, nnfw::cker::ComparisonParams::left_shift, MultiplyByQuantizedMultiplierSmallerThanOneExp(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
void nnfw::cker::BroadcastSelect4DSlow | ( | const Shape & | input_condition_shape, |
const D * | input_condition_data, | ||
const Shape & | input_x_shape, | ||
const T * | input_x_data, | ||
const Shape & | input_y_shape, | ||
const T * | input_y_data, | ||
const Shape & | output_shape, | ||
T * | output_data | ||
) |
Definition at line 63 of file Select.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 132 of file BroadcastTo.h.
References nnfw::cker::Tensor::buffer, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), nnfw::cker::Shape::FlatSize(), nnfw::cker::BCast::FromShape(), nnfw::cker::eigen_support::GetThreadPoolDevice(), nnfw::cker::BCastList< N >::IsValid(), nnfw::cker::BCast::output_shape(), output_shape, nnfw::cker::Shape::ReplaceWith(), nnfw::cker::Tensor::scalar(), nnfw::cker::Tensor::shape, and nnfw::cker::BCast::ToShape().
Referenced by nnfw::cker::train::BinaryArithmeticGrad(), and nnfw::cker::train::MeanGrad().
|
inline |
Definition at line 195 of file StridedSlice.h.
References begin, nnfw::cker::StridedSliceParams::begin_mask, nnfw::cker::StridedSliceParams::ellipsis_mask, end(), nnfw::cker::StridedSliceParams::end_mask, nnfw::cker::StridedSliceParams::new_axis_mask, nnfw::cker::StridedSliceParams::shrink_axis_mask, nnfw::cker::StridedSliceParams::start_indices, nnfw::cker::StridedSliceParams::start_indices_count, nnfw::cker::StridedSliceParams::stop_indices, nnfw::cker::StridedSliceParams::stop_indices_count, nnfw::cker::StridedSliceParams::strides, and nnfw::cker::StridedSliceParams::strides_count.
|
inline |
Definition at line 62 of file LSTM.h.
References ApplyActivationToVector(), MatrixBatchVectorMultiplyAccumulate(), MeanStddevNormalization(), VectorBatchVectorAdd(), VectorBatchVectorAssign(), VectorBatchVectorCwiseProduct(), and VectorBatchVectorCwiseProductAccumulate().
Referenced by LstmStepFloat().
void nnfw::cker::CalculateLstmOutputFloat | ( | int | n_batch, |
int | n_cell, | ||
int | n_output, | ||
const float * | cell_state, | ||
const float * | output_gate, | ||
FusedActivationFunctionType | activation, | ||
const float * | projection_weights, | ||
const float * | projection_bias, | ||
const float | proj_clip, | ||
float * | output_state, | ||
float * | scratch | ||
) |
Definition at line 183 of file LSTM.h.
References ApplyActivationToVector(), CwiseClipping(), MatrixBatchVectorMultiplyAccumulate(), VectorBatchVectorAssign(), and VectorVectorCwiseProduct().
Referenced by LstmStepFloat().
|
inline |
Definition at line 268 of file Shape.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and nnfw::cker::Shape::FlatSize().
Referenced by MatchingFlatSize().
void nnfw::cker::checkOutputSize | ( | const StridedSliceParams & | op_params, |
const Shape & | input_shape, | ||
const Shape & | output_shape, | ||
uint32_t | rank | ||
) |
Definition at line 224 of file StridedSlice.h.
References begin, end(), output_shape, nnfw::cker::StridedSliceParams::shrink_axis_mask, StartForAxis(), StopForAxis(), and nnfw::cker::StridedSliceParams::strides.
|
inline |
Definition at line 32 of file StridedSlice.h.
Referenced by StartForAxis(), and StopForAxis().
|
inline |
Definition at line 53 of file Comparison.h.
References output_shape.
|
inline |
Definition at line 40 of file Comparison.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 62 of file Comparison.h.
References nnfw::cker::ComparisonParams::input1_multiplier, nnfw::cker::ComparisonParams::input1_offset, nnfw::cker::ComparisonParams::input1_shift, nnfw::cker::ComparisonParams::input2_multiplier, nnfw::cker::ComparisonParams::input2_offset, nnfw::cker::ComparisonParams::input2_shift, nnfw::cker::ComparisonParams::left_shift, MatchingFlatSize(), MultiplyByQuantizedMultiplierSmallerThanOneExp(), and output_shape.
|
inline |
Definition at line 40 of file BCast.h.
Referenced by nnfw::cker::BCastList< N >::BCastList().
|
inline |
Definition at line 100 of file ResizeBilinear.h.
Referenced by ResizeBilinear(), ResizeBilinearGeneric(), and ResizeBilinearGenericSmallChannel().
|
inline |
Definition at line 268 of file ResizeBilinear.h.
|
inline |
Definition at line 33 of file Concatenation.h.
References nnfw::cker::ConcatenationParams::axis, nnfw::cker::Shape::Dims(), nnfw::cker::ConcatenationParams::inputs_count, MatchingDim(), and output_shape.
|
inline |
Definition at line 83 of file Concatenation.h.
References nnfw::cker::ConcatenationParams::axis, nnfw::cker::Shape::Dims(), nnfw::cker::ConcatenationParams::input_scale, nnfw::cker::ConcatenationParams::input_zeropoint, nnfw::cker::ConcatenationParams::inputs_count, nnfw::cker::ConcatenationParams::output_scale, output_shape, and nnfw::cker::ConcatenationParams::output_zeropoint.
Referenced by onert::backend::cpu::ops::ConcatLayer::concatenationQuant8().
|
inline |
Definition at line 277 of file Utils.h.
References nnfw::cker::Shape::Dims(), nnfw::cker::NdArrayDesc< N >::extents, and nnfw::cker::NdArrayDesc< N >::strides.
void nnfw::cker::CopyMultipleTimes | ( | const T * | in_data, |
int32_t | in_size, | ||
M | multiplier, | ||
T * | out_data | ||
) |
Definition at line 29 of file Tile.h.
Referenced by TileOneDimension().
|
inline |
Definition at line 41 of file Elementwise.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 152 of file Utils.h.
Referenced by GetInvSqrtQuantizedMultiplierExp().
|
inline |
Definition at line 34 of file TensorUtils.h.
References CwiseClipping(), and NEON_OR_PORTABLE.
Referenced by CalculateLstmOutputFloat(), CwiseClipping(), and UpdateLstmCellFloat().
|
inline |
Definition at line 30 of file DepthToSpace.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), Offset(), and output_shape.
|
inline |
Definition at line 124 of file DepthwiseConv.h.
References nnfw::cker::optimized::DepthwiseConvImpl(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::cpu_backend_threadpool::Execute(), HowManyConvThreads(), MultithreadAlongBatches(), and output_shape.
void nnfw::cker::DepthwiseConvOp | ( | const DepthwiseConvParams & | params, |
const Shape & | input_shape, | ||
const float * | input_data, | ||
const Shape & | filter_shape, | ||
const float * | filter_data, | ||
const Shape & | bias_shape, | ||
const float * | bias_data, | ||
float * | padded_filter_data, | ||
bool | pad_filter, | ||
float * | filter_buffers_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 185 of file DepthwiseConv.h.
References nnfw::cker::DepthwiseConvParams::depth_multiplier, nnfw::cker::DepthwiseConvParams::dilation_height_factor, nnfw::cker::DepthwiseConvParams::dilation_width_factor, nnfw::cker::Shape::Dims(), nnfw::cker::DepthwiseConvParams::float_activation_max, nnfw::cker::DepthwiseConvParams::float_activation_min, nnfw::cker::PaddingValues::height, MatchingDim(), output_shape, nnfw::cker::DepthwiseConvParams::padding_values, nnfw::cker::DepthwiseConvParams::stride_height, nnfw::cker::DepthwiseConvParams::stride_width, and nnfw::cker::PaddingValues::width.
Referenced by onert::backend::cpu::ops::DepthwiseConvolutionLayer::convFloat32().
|
inline |
Definition at line 115 of file Dequantize.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 80 of file Dequantize.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 44 of file Dequantize.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 30 of file ELU.h.
References MatchingFlatSize(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
|
inline |
Definition at line 89 of file ShapeIterator.h.
References nnfw::cker::ShapeIterator::end_iterator().
Referenced by buildStridedSliceParams(), checkOutputSize(), nnfw::cker::xent_ops::functor::XentFunctorBase< Device, T >::operator()(), nnfw::cker::training_ops::functor::ApplyAdamNonCuda< Device, T >::operator()(), operator<<(), and StridedSlice().
|
inline |
Definition at line 30 of file Comparison.h.
|
inline |
Definition at line 29 of file Erf.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 30 of file Exp.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 28 of file Fill.h.
References output_shape.
void nnfw::cker::Fill | ( | random::PhiloxRandom | random, |
Tensor * | output | ||
) |
Definition at line 64 of file StatelessRandomUniform.h.
|
inline |
Definition at line 253 of file Shape.h.
References nnfw::cker::Shape::DimensionsCount(), and nnfw::cker::Shape::DimsData().
Referenced by nnfw::cker::optimized_integer_ops::depthwise_conv::DepthwiseConvGeneral(), nnfw::cker::optimized::depthwise_conv::DepthwiseConvGeneral(), nnfw::cker::optimized::DepthwiseConvImpl(), FullyConnected(), FullyConnectedSparseWeight16x1(), FullyConnectedSparseWeightRandom(), nnfw::cker::optimized_integer_ops::HowManyConvThreads(), MapAsMatrixWithLastDimAsRows(), MatchingFlatSizeSkipDim(), nnfw::cker::train::MSE(), and nnfw::cker::train::MSEGrad().
|
inline |
Definition at line 90 of file Elementwise.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 30 of file FloorDiv.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 66 of file FloorDiv.h.
References nnfw::cker::Shape::FlatSize().
|
inline |
Definition at line 34 of file FloorMod.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 82 of file FloorMod.h.
References nnfw::cker::Shape::FlatSize().
|
inline |
Definition at line 98 of file FullyConnected.h.
References nnfw::cker::FullyConnectedParams::activation, ApplyActivationToVector(), nnfw::cker::Shape::Dims(), nnfw::cker::Shape::FlatSize(), kNone, MatrixBatchVectorMultiplyAccumulate(), VectorBatchVectorAssign(), and ZeroVector().
Referenced by onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedFloat32(), and onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedQuant8().
|
inline |
Definition at line 131 of file FullyConnected.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), FlatSizeSkipDim(), nnfw::cker::FullyConnectedParams::input_offset, MatchingDim(), MultiplyByQuantizedMultiplier(), nnfw::cker::FullyConnectedParams::output_multiplier, nnfw::cker::FullyConnectedParams::output_offset, output_shape, nnfw::cker::FullyConnectedParams::output_shift, nnfw::cker::FullyConnectedParams::quantized_activation_max, nnfw::cker::FullyConnectedParams::quantized_activation_min, and nnfw::cker::FullyConnectedParams::weights_offset.
|
inline |
Definition at line 183 of file FullyConnected.h.
References nnfw::cker::FCTempArena::accum_scratch, nnfw::cker::FullyConnectedParams::activation, ApplyActivationToVector(), nnfw::cker::Shape::Dims(), nnfw::cker::Shape::FlatSize(), nnfw::cker::FCTempArena::input_quantized, IsZeroVector(), kNone, MatrixBatchVectorMultiplyAccumulate(), offset(), output_shape, nnfw::cker::FCTempArena::scaling_factors, SymmetricQuantizeFloats(), VectorBatchVectorAssign(), nnfw::cker::FullyConnectedParams::weights_scale, and ZeroVector().
Referenced by onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedHybrid().
|
inline |
Definition at line 57 of file FullyConnectedSparse16x1.h.
References nnfw::cker::FullyConnectedParams::activation, ApplyActivationToVector(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), FlatSizeSkipDim(), kNone, MatchingDim(), output_shape, VectorBatchVectorAssign(), and ZeroVector().
Referenced by onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedSparseWeight().
|
inline |
Definition at line 250 of file FullyConnected.h.
References nnfw::cker::FullyConnectedParams::activation, ApplyActivationToVector(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), FlatSizeSkipDim(), kNone, MatchingDim(), output_shape, VectorBatchVectorAssign(), and ZeroVector().
Referenced by onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedSparseWeight().
|
inline |
Definition at line 31 of file Gather.h.
References nnfw::cker::GatherParams::axis, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and nnfw::cker::Shape::FlatSize().
void nnfw::cker::GenerateKey | ( | Tensor | seed, |
random::PhiloxRandom::Key * | out_key, | ||
random::PhiloxRandom::ResultType * | out_counter | ||
) |
Definition at line 37 of file StatelessRandomUniform.h.
References nnfw::cker::Tensor::flat().
Referenced by StatelessRandomUniform().
|
inline |
Definition at line 37 of file BatchToSpaceND.h.
Referenced by BatchToSpaceND().
|
inline |
Definition at line 164 of file Utils.h.
References CountLeadingZeros().
Referenced by L2NormalizeQuant8().
|
inline |
|
inline |
Definition at line 30 of file Range.h.
References size.
|
inline |
Definition at line 33 of file Comparison.h.
|
inline |
Definition at line 32 of file Comparison.h.
|
inline |
Definition at line 81 of file DepthwiseConv.h.
References nnfw::cker::Shape::Dims(), and output_shape.
Referenced by DepthwiseConv().
|
inline |
Definition at line 208 of file Reduce.h.
References nnfw::cker::Shape::DimensionsCount(), and nnfw::cker::Shape::DimsData().
Referenced by nnfw::cker::ReduceMean::PrepareforReduce(), and nnfw::cker::Reduce::ReduceGeneric().
|
inline |
Definition at line 31 of file InstanceNorm.h.
References ActivationFunctionWithMinMax(), nnfw::cker::InstanceNormParams::epsilon, nnfw::cker::InstanceNormParams::float_activation_max, nnfw::cker::InstanceNormParams::float_activation_min, MatchingDim(), Offset(), output_shape, and size.
|
inline |
Definition at line 104 of file TensorUtils.h.
References IsZeroVector(), and NEON_OR_PORTABLE.
Referenced by onert::backend::cpu::ops::FullyConnectedLayer::fullyConnectedHybrid(), FullyConnectedHybrid(), IsZeroVector(), LstmStepFloat(), and onert::backend::cpu::ops::FullyConnectedLayer::prepare().
void nnfw::cker::L2NormalizeFloat32 | ( | const Shape & | input_shape, |
const float * | input_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 30 of file L2Normalize.h.
References nnfw::cker::Shape::DimensionsCount(), MatchingDim(), MatchingFlatSizeSkipDim(), and output_shape.
Referenced by onert::backend::cpu::ops::L2NormLayer::run().
void nnfw::cker::L2NormalizeQuant8 | ( | L2NormParams & | params, |
const Shape & | input_shape, | ||
const uint8_t * | input_data, | ||
const Shape & | output_shape, | ||
uint8_t * | output_data | ||
) |
Definition at line 56 of file L2Normalize.h.
References nnfw::cker::Shape::DimensionsCount(), GetInvSqrtQuantizedMultiplierExp(), nnfw::cker::L2NormParams::input_zero_point, MatchingDim(), MatchingFlatSizeSkipDim(), MultiplyByQuantizedMultiplierSmallerThanOneExp(), and output_shape.
Referenced by onert::backend::cpu::ops::L2NormLayer::run().
|
inline |
Definition at line 31 of file LeakyReLU.h.
References nnfw::cker::LeakyReluParams::alpha, MatchingFlatSize(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
|
inline |
Definition at line 35 of file Comparison.h.
|
inline |
Definition at line 34 of file Comparison.h.
|
inline |
Definition at line 80 of file Elementwise.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 30 of file LogicalAnd.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 65 of file LogicalAnd.h.
References nnfw::cker::Shape::FlatSize().
|
inline |
Definition at line 28 of file LogicalNot.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 30 of file LogicalOr.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
|
inline |
Definition at line 65 of file LogicalOr.h.
References nnfw::cker::Shape::FlatSize().
|
inline |
Definition at line 32 of file Logistic.h.
References MapAsVector(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
|
inline |
Definition at line 34 of file LogSoftMax.h.
References nnfw::cker::SoftmaxParams::axis, nnfw::cker::SoftmaxParams::beta, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), MatchingDim(), and output_shape.
Referenced by onert::backend::cpu::ops::LogSoftMaxLayer::logsoftmaxFloat32(), and onert::backend::cpu::ops::LogSoftMaxLayer::logsoftmaxQuant8().
|
inline |
Definition at line 80 of file LogSoftMax.h.
References nnfw::cker::SoftmaxParams::axis, nnfw::cker::SoftmaxParams::beta, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), MatchingDim(), output_shape, nnfw::cker::SoftmaxParams::scale, nnfw::cker::SoftmaxParams::table, and nnfw::cker::SoftmaxParams::zero_point.
|
inline |
Definition at line 187 of file StridedSlice.h.
Referenced by StridedSlice().
|
inline |
Definition at line 285 of file LSTM.h.
References nnfw::cker::LSTMParams::activation, CalculateLstmGateFloat(), CalculateLstmOutputFloat(), nnfw::cker::LSTMParams::cell_clip, IsZeroVector(), kSigmoid, nnfw::cker::LSTMParams::proj_clip, and UpdateLstmCellFloat().
MatrixMap< Scalar > nnfw::cker::MapAsMatrixWithLastDimAsRows | ( | Scalar * | data, |
const Shape & | shape | ||
) |
Definition at line 60 of file Utils.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and FlatSizeSkipDim().
Referenced by nnfw::cker::train::AveragePool2DGrad(), AveragePool< float >(), nnfw::cker::train::CategoricalCrossEntropy(), nnfw::cker::train::CategoricalCrossEntropyGrad(), nnfw::cker::train::FullyConnectedBiasGrad(), nnfw::cker::train::MaxPool2D(), nnfw::cker::train::MaxPool2DGrad(), MaxPool< float >(), nnfw::cker::train::MeanGrad(), and Softmax().
VectorMap< Scalar > nnfw::cker::MapAsVector | ( | Scalar * | data, |
const Shape & | shape | ||
) |
Definition at line 43 of file Utils.h.
References nnfw::cker::Shape::FlatSize(), and size.
Referenced by Abs(), nnfw::cker::train::BinaryArithmeticGrad(), nnfw::cker::train::CategoricalCrossEntropy(), Logistic(), ReLU(), ReLU6(), nnfw::cker::train::ReLU6Grad(), nnfw::cker::train::ReLUGrad(), and Tanh().
|
inline |
Definition at line 220 of file Shape.h.
References nnfw::cker::Shape::Dims().
Referenced by AveragePool16(), nnfw::cker::train::AveragePool2DGrad(), AveragePool32(), AveragePool< float >(), AveragePool< int8_t >(), nnfw::cker::train::backpropFilter(), nnfw::cker::train::backpropInput(), Concatenation(), nnfw::cker::reference::Conv(), nnfw::cker::multithreaded::Conv(), nnfw::cker::reference::Conv(), nnfw::cker::reference::Conv(), nnfw::cker::train::ConvFilterGrad(), nnfw::cker::train::ConvInputGrad(), nnfw::cker::optimized_integer_ops::depthwise_conv::DepthwiseConvGeneral(), nnfw::cker::optimized::depthwise_conv::DepthwiseConvGeneral(), nnfw::cker::reference_integer_ops::DepthwiseConvHybridPerChannel(), nnfw::cker::optimized::DepthwiseConvImpl(), DepthwiseConvOp(), nnfw::cker::reference_integer_ops::DepthwiseConvPerChannel(), nnfw::cker::optimized_integer_ops::DepthwiseConvWithRounding(), nnfw::cker::optimized::DepthwiseConvWithRounding(), nnfw::cker::optimized::DilatedIm2col(), FullyConnected(), FullyConnectedSparseWeight16x1(), FullyConnectedSparseWeightRandom(), nnfw::cker::reference::HybridConvPerChannel(), nnfw::cker::optimized::Im2col(), InstanceNorm(), L2NormalizeFloat32(), L2NormalizeQuant8(), LogSoftmax(), LogSoftmax(), MatchingDim(), nnfw::cker::train::MaxPool2D(), nnfw::cker::train::MaxPool2DGrad(), MaxPool< float >(), MaxPool< uint8_t >(), RankOneSelect(), ResizeBilinear(), ResizeBilinear(), ResizeBilinear(), RmsNorm(), RoPE(), nnfw::cker::reference::Softmax(), Softmax(), SplitV(), TransposeConv(), and nnfw::cker::reference::TransposeImpl().
int nnfw::cker::MatchingDim | ( | const Shape & | shape1, |
int | index1, | ||
const Shape & | shape2, | ||
int | index2, | ||
Args... | args | ||
) |
Definition at line 228 of file Shape.h.
References nnfw::cker::Shape::Dims(), and MatchingDim().
|
inline |
Definition at line 333 of file Shape.h.
References nnfw::cker::Shape::FlatSize().
Referenced by nnfw::cker::optimized::Add(), nnfw::cker::optimized::Add(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::reference::BinaryArithmeticOp(), nnfw::cker::optimized::Div(), nnfw::cker::optimized::Mul(), nnfw::cker::optimized::Mul(), and nnfw::cker::optimized::Sub().
|
inline |
Definition at line 297 of file Shape.h.
References checkMatching(), and nnfw::cker::Shape::FlatSize().
Referenced by ComparisonImpl(), ComparisonWithScaling(), Cos(), Dequantize(), Dequantize(), Dequantize(), ELU(), Erf(), Exp(), Floor(), LeakyReLU(), Log(), LogicalNot(), Neg(), powImpl(), Quantize(), Quantize(), Quantize(), Quantize(), Round(), Rsqrt(), Select(), Sin(), Softmax(), nnfw::cker::train::SoftMaxGrad(), Sqrt(), and Square().
|
inline |
Definition at line 304 of file Shape.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and FlatSizeSkipDim().
Referenced by L2NormalizeFloat32(), L2NormalizeQuant8(), MatchingFlatSizeSkipDim(), RankOneSelect(), nnfw::cker::reference::Softmax(), and Softmax().
|
inline |
Definition at line 318 of file Shape.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and MatchingFlatSizeSkipDim().
void nnfw::cker::MatrixBandPart | ( | const T | num_lower_diags, |
const T | num_upper_diags, | ||
const Shape & | input_shape, | ||
const float * | input_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 30 of file MatrixBandPart.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and output_shape.
|
inline |
Definition at line 136 of file TensorUtils.h.
References MatrixBatchVectorMultiplyAccumulate(), and NEON_OR_PORTABLE.
|
inline |
Definition at line 127 of file TensorUtils.h.
References MatrixBatchVectorMultiplyAccumulate(), and NEON_OR_PORTABLE.
Referenced by CalculateLstmGateFloat(), CalculateLstmOutputFloat(), FullyConnected(), FullyConnectedHybrid(), MatrixBatchVectorMultiplyAccumulate(), MatrixBatchVectorMultiplyAccumulate(), and MatrixBatchVectorMultiplyAccumulate().
|
inline |
Definition at line 144 of file TensorUtils.h.
References MatrixBatchVectorMultiplyAccumulate(), and NEON_OR_PORTABLE.
|
inline |
Definition at line 47 of file MaxMin.h.
References desc1, desc2, nnfw::cker::Shape::DimensionsCount(), NdArrayDescsForElementwiseBroadcast(), Offset(), output_shape, and SubscriptToIndex().
void nnfw::cker::MaxPool | ( | const PoolParams & | , |
const Shape & | , | ||
const T * | , | ||
const Shape & | , | ||
T * | |||
) |
void nnfw::cker::MaxPool< float > | ( | const PoolParams & | params, |
const Shape & | input_shape, | ||
const float * | input_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 42 of file MaxPool.h.
References ActivationFunctionWithMinMax(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PoolParams::float_activation_max, nnfw::cker::PoolParams::float_activation_min, nnfw::cker::PaddingValues::height, MapAsMatrixWithLastDimAsRows(), MatchingDim(), NodeOffset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
void nnfw::cker::MaxPool< uint8_t > | ( | const PoolParams & | params, |
const Shape & | input_shape, | ||
const uint8_t * | input_data, | ||
const Shape & | output_shape, | ||
uint8_t * | output_data | ||
) |
Definition at line 98 of file MaxPool.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PoolParams::filter_height, nnfw::cker::PoolParams::filter_width, nnfw::cker::PaddingValues::height, MatchingDim(), Offset(), output_shape, nnfw::cker::PoolParams::padding_values, nnfw::cker::PoolParams::quantized_activation_max, nnfw::cker::PoolParams::quantized_activation_min, nnfw::cker::PoolParams::stride_height, nnfw::cker::PoolParams::stride_width, and nnfw::cker::PaddingValues::width.
void nnfw::cker::Mean | ( | const Shape & | input_shape, |
const In * | input_data, | ||
const Shape & | output_shape, | ||
Out * | output_data, | ||
const std::vector< int > & | axes | ||
) |
Definition at line 211 of file ReduceMean.h.
References nnfw::cker::Shape::DimensionsCount(), mean_reducer(), output_shape, and nnfw::cker::ReduceMean::ReduceOp().
Referenced by onert::backend::cpu::ops::MeanLayer::MeanFloat32().
Out nnfw::cker::mean_reducer | ( | const Out | data1, |
const In | data2, | ||
int | normalizer | ||
) |
Definition at line 41 of file ReduceMean.h.
Referenced by Mean().
void nnfw::cker::MeanAxis1And2 | ( | const Shape & | input_shape, |
const In * | input_data, | ||
const Shape & | output_shape, | ||
Out * | output_data | ||
) |
Definition at line 233 of file ReduceMean.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), Offset(), and output_shape.
Referenced by onert::backend::cpu::ops::MeanLayer::MeanFloat32().
void nnfw::cker::MeanQ8Asymm | ( | const Shape & | input_shape, |
const In * | input_data, | ||
float | input_scale, | ||
int32_t | input_offset, | ||
const Shape & | output_shape, | ||
Out * | output_data, | ||
float | output_scale, | ||
int32_t | output_offset, | ||
const std::vector< int > & | axes | ||
) |
Definition at line 221 of file ReduceMean.h.
References nnfw::cker::Shape::DimensionsCount(), output_shape, nnfw::cker::ReduceMean::ReduceOp(), and sum_reducer().
Referenced by onert::backend::cpu::ops::MeanLayer::MeanQuant8().
|
inline |
Definition at line 154 of file TensorUtils.h.
References PortableMeanStddevNormalization().
Referenced by CalculateLstmGateFloat().
|
inline |
Definition at line 96 of file Utils.h.
Referenced by nnfw::cker::reference::Conv(), nnfw::cker::reference::Conv(), nnfw::cker::optimized::depthwise_conv::DepthwiseConvGeneral(), nnfw::cker::reference_integer_ops::DepthwiseConvPerChannel(), FullyConnected(), nnfw::cker::optimized::MulElementwise(), nnfw::cker::optimized::MulElementwise(), nnfw::cker::optimized::MulSimpleBroadcast(), nnfw::cker::optimized::quant8_mul(), Quantize(), Requantize< int8_t, uint8_t >(), and Requantize< uint8_t, int8_t >().
|
inline |
|
inline |
Definition at line 111 of file Utils.h.
Referenced by nnfw::cker::optimized::AddElementwise(), nnfw::cker::optimized::AddElementwise(), nnfw::cker::optimized::AddScalarBroadcast(), BroadcastComparison4DSlowWithScaling(), ComparisonWithScaling(), L2NormalizeQuant8(), and nnfw::cker::optimized::quant8_sum().
|
inline |
Definition at line 95 of file DepthwiseConv.h.
Referenced by DepthwiseConv().
|
inline |
Definition at line 329 of file Utils.h.
References nnfw::cker::NdArrayDesc< N >::extents, and nnfw::cker::NdArrayDesc< N >::strides.
|
inline |
Definition at line 290 of file Utils.h.
References nnfw::cker::NdArrayDesc< N >::extents, and nnfw::cker::NdArrayDesc< N >::strides.
Referenced by nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), BroadcastComparison4DSlowImpl(), BroadcastComparison4DSlowWithScaling(), BroadcastSelect4DSlow(), FloorDivBroadcast(), FloorModBroadcast(), LogicalAndBroadcast(), LogicalOrBroadcast(), and MaximumMinimumBroadcast4DSlow().
|
inline |
Definition at line 70 of file Elementwise.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 387 of file Utils.h.
Referenced by ReduceImpl(), ReduceMeanImpl(), ReduceSumQuantImpl(), and SqDiffImpl().
|
inline |
Definition at line 147 of file Utils.h.
Referenced by nnfw::cker::train::AveragePool2DGrad(), AveragePool< float >(), nnfw::cker::train::MaxPool2D(), and MaxPool< float >().
|
inline |
Definition at line 31 of file Comparison.h.
|
inline |
|
inline |
Definition at line 237 of file Shape.h.
References nnfw::cker::Shape::DimensionsCount(), and nnfw::cker::Shape::DimsDataUpTo4D().
Referenced by AveragePool16(), AveragePool32(), AveragePool< int8_t >(), BatchToSpaceND(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), BroadcastComparison4DSlowImpl(), BroadcastComparison4DSlowWithScaling(), BroadcastSelect4DSlow(), nnfw::cker::reference::Conv(), nnfw::cker::reference::Conv(), nnfw::cker::reference::Conv(), DepthToSpace(), nnfw::cker::reference_integer_ops::DepthwiseConvHybridPerChannel(), nnfw::cker::reference_integer_ops::DepthwiseConvPerChannel(), nnfw::cker::optimized::DilatedIm2col(), nnfw::cker::optimized::ExtractPatchIntoBufferColumn(), FloorDivBroadcast(), FloorModBroadcast(), nnfw::cker::reference::HybridConvPerChannel(), InstanceNorm(), LogicalAndBroadcast(), LogicalOrBroadcast(), MaximumMinimumBroadcast4DSlow(), MaxPool< uint8_t >(), MeanAxis1And2(), Offset(), ResizeBilinear(), ResizeBilinearGeneric(), ResizeBilinearGenericSmallChannel(), ResizeBilinearKernel2x2(), RmsNorm(), RoPE(), Slice(), SpaceToBatchND(), SpaceToDepth(), StridedSlice(), TransposeConv(), and nnfw::cker::reference::TransposeImpl().
void nnfw::cker::OneHot | ( | const int32_t | depth, |
const T | on_value, | ||
const T | off_value, | ||
int32_t | axis, | ||
const Shape & | indices_shape, | ||
const TI * | indices_data, | ||
const Shape & | , | ||
T * | output_data | ||
) |
Definition at line 29 of file OneHot.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and nnfw::cker::Shape::FlatSize().
|
inline |
Definition at line 486 of file Utils.h.
void nnfw::cker::optimized_ops_preload_l1_keep | ( | const T * | ptr | ) |
Definition at line 455 of file Utils.h.
Referenced by Transpose2D().
|
inline |
Definition at line 30 of file Pack.h.
References nnfw::cker::PackParams::axis, nnfw::cker::PackParams::inputs_count, and output_shape.
|
inline |
List of padding information
Definition at line 30 of file Pad.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and output_shape.
|
inline |
Definition at line 148 of file SoftMax.h.
Referenced by onert::backend::cpu::ops::SoftMaxLayer::configure().
|
inline |
Definition at line 103 of file PortableTensorUtils.h.
Referenced by ApplyActivationToVector().
|
inline |
Definition at line 147 of file PortableTensorUtils.h.
References offset(), and size.
Referenced by onert::backend::cpu::ops::DepthwiseConvolutionLayer::convQ8iHybridPerChannel().
void nnfw::cker::PortableCwiseClipping | ( | T * | vector, |
const int | v_size, | ||
const T | clipping_value | ||
) |
Definition at line 63 of file PortableTensorUtils.h.
|
inline |
Definition at line 93 of file PortableTensorUtils.h.
|
inline |
Definition at line 242 of file PortableTensorUtils.h.
|
inline |
Definition at line 231 of file PortableTensorUtils.h.
References PortableMatrixBatchVectorMultiplyAccumulate().
|
inline |
Definition at line 201 of file PortableTensorUtils.h.
Referenced by PortableMatrixBatchVectorMultiplyAccumulate().
|
inline |
Definition at line 264 of file PortableTensorUtils.h.
Referenced by MeanStddevNormalization().
|
inline |
Definition at line 113 of file PortableTensorUtils.h.
|
inline |
Definition at line 121 of file PortableTensorUtils.h.
References size.
|
inline |
Definition at line 80 of file PortableTensorUtils.h.
Referenced by VectorBatchVectorAdd().
|
inline |
Definition at line 71 of file PortableTensorUtils.h.
Referenced by VectorBatchVectorAssign().
|
inline |
Definition at line 293 of file PortableTensorUtils.h.
Referenced by ZeroVector().
|
inline |
Definition at line 31 of file Pow.h.
References MatchingFlatSize(), and output_shape.
Referenced by onert::backend::cpu::ops::PowLayer::powFloat32().
|
inline |
Definition at line 109 of file BinaryArithmeticOps.h.
References nnfw::cker::BinaryArithmeticOpParam::broadcast_category, nnfw::cker::BinaryArithmeticOpParam::broadcast_shape, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), kFirstInputBroadcastsFast, kGenericBroadcast, kNonBroadcast, and kSecondInputBroadcastsFast.
|
inline |
Definition at line 207 of file Quantize.h.
References MultiplyByQuantizedMultiplier().
|
inline |
Definition at line 156 of file Quantize.h.
References MatchingFlatSize(), output_shape, and RoundToNearest().
|
inline |
Definition at line 50 of file Quantize.h.
References MatchingFlatSize(), output_shape, and RoundToNearest().
|
inline |
Definition at line 103 of file Quantize.h.
References MatchingFlatSize(), output_shape, and RoundToNearest().
|
inline |
Definition at line 34 of file Quantize.h.
References MatchingFlatSize(), and output_shape.
Referenced by onert::backend::cpu::ops::affineQuantize(), and nnfw::cker::optimized_integer_ops::depthwise_conv::DepthwiseConvGeneral().
|
inline |
Definition at line 48 of file Utils.h.
Referenced by QuantizeMultiplierSmallerThanOneExp().
|
inline |
Definition at line 85 of file Utils.h.
References QuantizeMultiplier().
|
inline |
|
inline |
|
inline |
Definition at line 44 of file Range.h.
void nnfw::cker::RankOneSelect | ( | const Shape & | input_condition_shape, |
const D * | input_condition_data, | ||
const Shape & | input_x_shape, | ||
const T * | input_x_data, | ||
const Shape & | input_y_shape, | ||
const T * | input_y_data, | ||
const Shape & | output_shape, | ||
T * | output_data | ||
) |
Definition at line 45 of file Select.h.
References nnfw::cker::Shape::FlatSize(), MatchingDim(), MatchingFlatSizeSkipDim(), offset(), and output_shape.
|
inline |
Definition at line 420 of file Utils.h.
References offset().
Referenced by ReduceImpl(), ReduceMeanImpl(), and ReduceSumQuantImpl().
|
inline |
Definition at line 118 of file Reduce.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), NextIndex(), and ReducedOutputOffset().
|
inline |
Definition at line 52 of file ReduceMean.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), NextIndex(), and ReducedOutputOffset().
|
inline |
Definition at line 83 of file ReduceMean.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), NextIndex(), and ReducedOutputOffset().
|
inline |
Definition at line 32 of file ReLU.h.
References MapAsVector(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
|
inline |
Definition at line 32 of file ReLU6.h.
References MapAsVector(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
|
inline |
Definition at line 301 of file Quantize.h.
References size.
|
inline |
Definition at line 379 of file Quantize.h.
References MultiplyByQuantizedMultiplier(), and size.
Referenced by onert::backend::cpu::ops::QuantizeLayer::run().
|
inline |
Definition at line 311 of file Quantize.h.
References MultiplyByQuantizedMultiplier(), and size.
Referenced by onert::backend::cpu::ops::QuantizeLayer::run().
|
inline |
Definition at line 285 of file ResizeBilinear.h.
References nnfw::cker::ResizeBilinearParams::align_corners, ComputeInterpolationValues(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::ResizeBilinearParams::half_pixel_centers, MatchingDim(), Offset(), nnfw::cker::ResizeBilinearParams::output_height, output_shape, and nnfw::cker::ResizeBilinearParams::output_width.
void nnfw::cker::ResizeBilinear | ( | ResizeBilinearParams & | params, |
const Shape & | input_shape, | ||
const float * | input_data, | ||
const Shape & | output_shape, | ||
float * | output_data | ||
) |
Definition at line 213 of file ResizeBilinear.h.
References nnfw::cker::ResizeBilinearParams::align_corners, nnfw::cker::Shape::Dims(), nnfw::cker::ResizeBilinearParams::half_pixel_centers, MatchingDim(), nnfw::cker::ResizeBilinearParams::output_height, output_shape, nnfw::cker::ResizeBilinearParams::output_width, ResizeBilinear2x2(), and ResizeBilinearGeneric().
Referenced by onert::backend::cpu::ops::ResizeBilinearLayer::run().
void nnfw::cker::ResizeBilinear | ( | ResizeBilinearParams & | params, |
const Shape & | input_shape, | ||
const uint8_t * | input_data, | ||
const Shape & | output_shape, | ||
uint8_t * | output_data | ||
) |
Definition at line 247 of file ResizeBilinear.h.
References nnfw::cker::ResizeBilinearParams::align_corners, nnfw::cker::Shape::Dims(), nnfw::cker::ResizeBilinearParams::half_pixel_centers, MatchingDim(), nnfw::cker::ResizeBilinearParams::output_height, output_shape, and nnfw::cker::ResizeBilinearParams::output_width.
|
inline |
Definition at line 69 of file ResizeBilinear.h.
References output_shape, and ResizeBilinearKernel2x2().
Referenced by ResizeBilinear().
|
inline |
Definition at line 118 of file ResizeBilinear.h.
References ComputeInterpolationValues(), Offset(), and ResizeBilinearKernel().
Referenced by ResizeBilinear().
|
inline |
Definition at line 171 of file ResizeBilinear.h.
References ComputeInterpolationValues(), and Offset().
|
inline |
Definition at line 89 of file ResizeBilinear.h.
Referenced by ResizeBilinearGeneric().
|
inline |
Definition at line 30 of file ResizeBilinear.h.
References nnfw::cker::Shape::Dims(), Offset(), and output_shape.
Referenced by ResizeBilinear2x2().
|
inline |
Definition at line 169 of file Reduce.h.
Referenced by nnfw::cker::ReduceMean::PrepareforReduce(), nnfw::cker::Reduce::QuantizedMeanOrSum(), and nnfw::cker::Reduce::ReduceGeneric().
void nnfw::cker::Reverse | ( | int | axis, |
const Shape & | input_shape, | ||
const Scalar * | input_data, | ||
const Shape & | , | ||
Scalar * | output_data | ||
) |
Definition at line 31 of file Reverse.h.
References nnfw::cker::Shape::DimensionsCount(), and nnfw::cker::Shape::Dims().
Referenced by nnfw::cker::BCastList< N >::BCastList().
|
inline |
Definition at line 32 of file RmsNorm.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::RmsNormParams::epsilon, MatchingDim(), Offset(), and output_shape.
Referenced by onert::backend::cpu::ops::RmsNormLayer::run().
|
inline |
Definition at line 32 of file RoPE.h.
References nnfw::cker::Shape::Dims(), kGptNeox, MatchingDim(), offset(), Offset(), and output_shape.
|
inline |
Definition at line 65 of file Round.h.
References MatchingFlatSize(), output_shape, and RoundToNearest().
float nnfw::cker::round_nearest | ( | float | value | ) |
Definition at line 29 of file ReduceMean.h.
Referenced by nnfw::cker::ReduceMean::ReduceOp().
|
inline |
Definition at line 31 of file Round.h.
Referenced by Quantize(), Quantize(), Quantize(), and Round().
|
inline |
Definition at line 59 of file Elementwise.h.
References MatchingFlatSize(), output_shape, and size.
void nnfw::cker::Select | ( | const Shape & | input_condition_shape, |
const D * | input_condition_data, | ||
const Shape & | input_x_shape, | ||
const T * | input_x_data, | ||
const Shape & | input_y_shape, | ||
const T * | input_y_data, | ||
const Shape & | output_shape, | ||
T * | output_data | ||
) |
Definition at line 32 of file Select.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 31 of file Elementwise.h.
References MatchingFlatSize(), output_shape, and size.
|
inline |
Definition at line 72 of file Slice.h.
References Slice().
|
inline |
Definition at line 31 of file Slice.h.
References nnfw::cker::SliceParams::begin, nnfw::cker::SliceParams::begin_count, begin_count, nnfw::cker::Shape::Dims(), Offset(), nnfw::cker::SliceParams::size, nnfw::cker::SliceParams::size_count, size_count, and nnfw::cker::SequentialTensorWriter< T >::WriteN().
Referenced by Slice().
|
inline |
Definition at line 79 of file SoftMax.h.
Referenced by onert::backend::cpu::ops::SoftMaxLayer::softmaxFloat32().
|
inline |
Definition at line 116 of file SoftMax.h.
References nnfw::cker::SoftmaxParams::beta, MapAsMatrixWithLastDimAsRows(), MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 159 of file SoftMax.h.
References nnfw::cker::Shape::DimensionsCount(), MatchingDim(), MatchingFlatSizeSkipDim(), output_shape, nnfw::cker::SoftmaxParams::scale, nnfw::cker::SoftmaxParams::table, and nnfw::cker::SoftmaxParams::zero_point.
|
inline |
Definition at line 31 of file SpaceToBatchND.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), Offset(), nnfw::cker::SpaceToBatchParams::output_offset, and output_shape.
|
inline |
Definition at line 30 of file SpaceToDepth.h.
References nnfw::cker::SpaceToDepthParams::block_size, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), Offset(), and output_shape.
void nnfw::cker::Split | ( | const SplitParams & | params, |
const Shape & | input_shape, | ||
const Scalar * | input_data, | ||
const Shape & | output_shape, | ||
Scalar *const * | output_data | ||
) |
Definition at line 30 of file Split.h.
References nnfw::cker::SplitParams::axis, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::SplitParams::num_split, and output_shape.
void nnfw::cker::SplitV | ( | const SplitVParams & | params, |
const Shape & | input_shape, | ||
const Scalar * | input_data, | ||
std::vector< nnfw::cker::Shape > & | output_shapes, | ||
Scalar *const * | output_data | ||
) |
Definition at line 30 of file SplitV.h.
References nnfw::cker::SplitVParams::axis, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), MatchingDim(), and nnfw::cker::SplitVParams::num_split.
void nnfw::cker::SqDiff | ( | const Shape & | input1_shape, |
const T * | input1_data, | ||
const Shape & | input2_shape, | ||
const T * | input2_data, | ||
const Shape & | output_shape, | ||
T * | output_data | ||
) |
Definition at line 63 of file SqDiff.h.
References nnfw::cker::Shape::DimensionsCount(), output_shape, and SQDIFF.
Referenced by onert::backend::cpu::ops::SqDiffLayer::SqDiffFloat32().
void nnfw::cker::SqDiffImpl | ( | const Shape & | input1_shape, |
const T * | input1_data, | ||
const Shape & | input2_shape, | ||
const T * | input2_data, | ||
const Shape & | output_shape, | ||
T * | output_data, | ||
NdArrayDesc< N > * | desc1_in, | ||
NdArrayDesc< N > * | desc2_in, | ||
NdArrayDesc< N > * | desc_out | ||
) |
Definition at line 40 of file SqDiff.h.
References NextIndex(), output_shape, and SubscriptToIndexGeneric().
|
inline |
Definition at line 101 of file Elementwise.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 112 of file Elementwise.h.
References MatchingFlatSize(), and output_shape.
|
inline |
Definition at line 83 of file StridedSlice.h.
References nnfw::cker::StridedSliceParams::begin_mask, Clamp(), nnfw::cker::Shape::Dims(), nnfw::cker::StridedSliceParams::start_indices, and nnfw::cker::StridedSliceParams::strides.
Referenced by checkOutputSize(), and StridedSlice().
|
inline |
Definition at line 75 of file StatelessRandomUniform.h.
References nnfw::cker::Tensor::buffer, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), GenerateKey(), output_shape, nnfw::cker::Shape::ReplaceWith(), and nnfw::cker::Tensor::shape.
Referenced by onert::backend::cpu::ops::StatelessRandomUniformLayer::StatelessRandomUniformFloat32().
|
inline |
Definition at line 126 of file StridedSlice.h.
References Clamp(), nnfw::cker::Shape::Dims(), nnfw::cker::StridedSliceParams::end_mask, nnfw::cker::StridedSliceParams::shrink_axis_mask, nnfw::cker::StridedSliceParams::stop_indices, and nnfw::cker::StridedSliceParams::strides.
Referenced by checkOutputSize(), and StridedSlice().
|
inline |
Definition at line 258 of file StridedSlice.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), end(), LoopCondition(), Offset(), output_shape, StartForAxis(), StopForAxis(), StridedSlicePadIndices(), nnfw::cker::StridedSliceParams::strides, and nnfw::cker::StridedSliceParams::strides_count.
|
inline |
Definition at line 42 of file StridedSlice.h.
References nnfw::cker::StridedSliceParams::begin_mask, nnfw::cker::StridedSliceParams::ellipsis_mask, nnfw::cker::StridedSliceParams::end_mask, nnfw::cker::StridedSliceParams::new_axis_mask, nnfw::cker::StridedSliceParams::shrink_axis_mask, nnfw::cker::StridedSliceParams::start_indices, nnfw::cker::StridedSliceParams::start_indices_count, nnfw::cker::StridedSliceParams::stop_indices, nnfw::cker::StridedSliceParams::stop_indices_count, nnfw::cker::StridedSliceParams::strides, and nnfw::cker::StridedSliceParams::strides_count.
Referenced by StridedSlice().
|
inline |
Definition at line 115 of file TensorUtils.h.
References NEON_OR_PORTABLE, and Sub1Vector().
Referenced by Sub1Vector(), and UpdateLstmCellFloat().
|
inline |
Definition at line 255 of file Utils.h.
References nnfw::cker::NdArrayDesc< N >::extents, and nnfw::cker::NdArrayDesc< N >::strides.
Referenced by nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), nnfw::cker::reference::BroadcastBinaryArithmeticOpSlow(), BroadcastComparison4DSlowImpl(), BroadcastComparison4DSlowWithScaling(), BroadcastSelect4DSlow(), FloorDivBroadcast(), FloorModBroadcast(), LogicalAndBroadcast(), LogicalOrBroadcast(), and MaximumMinimumBroadcast4DSlow().
|
inline |
Definition at line 264 of file Utils.h.
References nnfw::cker::NdArrayDesc< N >::extents, and nnfw::cker::NdArrayDesc< N >::strides.
Referenced by SqDiffImpl().
int nnfw::cker::sum_reducer | ( | const int | data1, |
const In | data2 | ||
) |
Definition at line 46 of file ReduceMean.h.
Referenced by MeanQ8Asymm().
|
inline |
Definition at line 120 of file TensorUtils.h.
References NEON_OR_PORTABLE, size, and SymmetricQuantizeFloats().
Referenced by FullyConnectedHybrid(), and SymmetricQuantizeFloats().
|
inline |
Definition at line 31 of file Tanh.h.
References MapAsVector(), and output_shape.
Referenced by onert::backend::cpu::ops::ElementwiseActivationLayer::configure().
nnfw::cker::TFLITE_COMPARISON_OP | ( | Equal | ) |
nnfw::cker::TFLITE_COMPARISON_OP | ( | Greater | ) |
nnfw::cker::TFLITE_COMPARISON_OP | ( | GreaterEqual | ) |
nnfw::cker::TFLITE_COMPARISON_OP | ( | Less | ) |
nnfw::cker::TFLITE_COMPARISON_OP | ( | LessEqual | ) |
nnfw::cker::TFLITE_COMPARISON_OP | ( | NotEqual | ) |
std::pair< int, int > nnfw::cker::TileOneDimension | ( | const Shape & | in_dimensions, |
const T * | in_data, | ||
const M * | multipliers, | ||
T * | out_data, | ||
int | dimension | ||
) |
Definition at line 41 of file Tile.h.
References CopyMultipleTimes(), nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), and TileOneDimension().
Referenced by TileOneDimension().
TTypes< typenameTensorType::Scalar, TensorType::NumIndices >::Tensor32Bit nnfw::cker::To32Bit | ( | TensorType | in | ) |
Definition at line 178 of file Tensor.h.
References To32BitDims().
Referenced by nnfw::cker::functor::BroadcastTo< Device, T >::DoBCast32Bit().
Eigen::DSizes< Index32, DSizes::count > nnfw::cker::To32BitDims | ( | const DSizes & | in | ) |
void nnfw::cker::Transpose | ( | const TransposeParams & | unshrunk_params, |
const Shape & | unshrunk_input_shape, | ||
const T * | input_data, | ||
const Shape & | unshrunk_output_shape, | ||
T * | output_data | ||
) |
Definition at line 509 of file Transpose.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::FlatSize(), nnfw::cker::TransposeParams::perm, nnfw::cker::TransposeParams::perm_count, and TransposeImpl().
Referenced by onert::backend::cpu::ops::TransposeLayer::transpose().
|
inline |
Definition at line 297 of file Transpose.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::DimsData(), optimized_ops_preload_l1_keep(), and output_shape.
Referenced by TransposeImpl().
|
inline |
Definition at line 405 of file Transpose.h.
References nnfw::cker::Shape::Dims(), and nnfw::cker::TransposeParams::perm.
Referenced by TransposeImpl().
|
inline |
Definition at line 30 of file TransposeConv.h.
References nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::PaddingValues::height, MatchingDim(), Offset(), output_shape, nnfw::cker::TransposeConvParams::padding_values, nnfw::cker::TransposeConvParams::stride_height, nnfw::cker::TransposeConvParams::stride_width, and nnfw::cker::PaddingValues::width.
void nnfw::cker::TransposeImpl | ( | const TransposeParams & | params, |
const Shape & | input_shape, | ||
const T * | input_data, | ||
const Shape & | output_shape, | ||
T * | output_data | ||
) |
Definition at line 475 of file Transpose.h.
References nnfw::cker::Shape::DimensionsCount(), output_shape, nnfw::cker::reference::Transpose(), Transpose2D(), and Transpose3D().
Referenced by Transpose().
void nnfw::cker::Unpack | ( | const UnpackParams & | params, |
const Shape & | input_shape, | ||
const Scalar * | input_data, | ||
const Shape & | output_shape, | ||
Scalar *const * | output_datas | ||
) |
Definition at line 30 of file Unpack.h.
References nnfw::cker::UnpackParams::axis, nnfw::cker::Shape::DimensionsCount(), nnfw::cker::Shape::Dims(), nnfw::cker::UnpackParams::num_split, and output_shape.
void nnfw::cker::UpdateLstmCellFloat | ( | int | n_batch, |
int | n_cell, | ||
float * | cell_state, | ||
const float * | input_gate, | ||
float * | forget_gate, | ||
const float * | cell_gate, | ||
bool | use_cifg, | ||
float | clip | ||
) |
Definition at line 135 of file LSTM.h.
References CwiseClipping(), Sub1Vector(), VectorVectorCwiseProduct(), and VectorVectorCwiseProductAccumulate().
Referenced by LstmStepFloat().
void nnfw::cker::ValidateGemmParams | ( | const GemmParams< AccumScalar, DstScalar, quantization_flavor > & | params | ) |
Definition at line 544 of file Types.h.
References kFloatingPoint, kIntegerWithPerRowMultiplier, and kIntegerWithUniformMultiplier.
|
inline |
Definition at line 39 of file TensorUtils.h.
References PortableVectorBatchVectorAdd().
Referenced by CalculateLstmGateFloat().
|
inline |
Definition at line 44 of file TensorUtils.h.
References PortableVectorBatchVectorAssign().
Referenced by CalculateLstmGateFloat(), CalculateLstmOutputFloat(), FullyConnected(), FullyConnectedHybrid(), FullyConnectedSparseWeight16x1(), and FullyConnectedSparseWeightRandom().
|
inline |
Definition at line 76 of file TensorUtils.h.
References VectorVectorCwiseProduct().
Referenced by CalculateLstmGateFloat().
|
inline |
Definition at line 92 of file TensorUtils.h.
References VectorVectorCwiseProductAccumulate().
Referenced by CalculateLstmGateFloat().
|
inline |
Definition at line 52 of file TensorUtils.h.
Referenced by CalculateLstmOutputFloat(), UpdateLstmCellFloat(), and VectorBatchVectorCwiseProduct().
|
inline |
Definition at line 64 of file TensorUtils.h.
Referenced by UpdateLstmCellFloat(), and VectorBatchVectorCwiseProductAccumulate().
|
inline |
Definition at line 160 of file TensorUtils.h.
References PortableZeroVector().
Referenced by FullyConnected(), FullyConnectedHybrid(), FullyConnectedSparseWeight16x1(), and FullyConnectedSparseWeightRandom().