/external/tensorflow/tensorflow/lite/tools/optimize/calibration/custom_logging_ops/ |
D | lstm.cc | 63 const TfLiteLSTMParams* params, int n_batch, int n_cell, int n_input, in LstmStepWithAuxInput() argument 80 std::fill_n(input_gate_scratch, n_cell * n_batch, 0.0f); in LstmStepWithAuxInput() 82 std::fill_n(forget_gate_scratch, n_cell * n_batch, 0.0f); in LstmStepWithAuxInput() 83 std::fill_n(cell_scratch, n_cell * n_batch, 0.0f); in LstmStepWithAuxInput() 84 std::fill_n(output_gate_scratch, n_cell * n_batch, 0.0f); in LstmStepWithAuxInput() 88 n_batch, input_gate_scratch); in LstmStepWithAuxInput() 90 tensor_utils::VectorBatchVectorAssign(forget_gate_bias_ptr, n_cell, n_batch, in LstmStepWithAuxInput() 92 tensor_utils::VectorBatchVectorAssign(cell_bias_ptr, n_cell, n_batch, in LstmStepWithAuxInput() 94 tensor_utils::VectorBatchVectorAssign(output_gate_bias_ptr, n_cell, n_batch, in LstmStepWithAuxInput() 101 input_to_input_weights_ptr, n_cell, n_input, input_ptr, n_batch, in LstmStepWithAuxInput() [all …]
|
/external/tensorflow/tensorflow/lite/kernels/internal/reference/ |
D | portable_tensor_utils.h | 60 int n_batch, float* result) { in MatrixBatchVectorMultiplyAccumulate() argument 62 n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 69 int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 72 scaling_factors, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 78 int n_batch, float* __restrict__ result, const float* per_channel_scale, in MatrixBatchVectorMultiplyAccumulate() argument 82 matrix, m_rows, m_cols, vectors, scaling_factors, n_batch, result, in MatrixBatchVectorMultiplyAccumulate() 91 int n_batch, int32_t* scratch, in MatrixBatchVectorMultiplyAccumulate() argument 95 scaling_factors, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 101 const float* __restrict__ vector, int n_batch, float* __restrict__ result) { in SparseMatrixBatchVectorMultiplyAccumulate1x4() argument 103 matrix, segments, indices, m_rows, m_cols, vector, n_batch, result); in SparseMatrixBatchVectorMultiplyAccumulate1x4() [all …]
|
D | portable_tensor_utils_impl.h | 60 int n_batch, float* result); 65 int n_batch, float* __restrict__ result); 70 int n_batch, float* __restrict__ result, const float* per_channel_scale, 77 int n_batch, int32_t* scratch, float* __restrict__ result, 83 const float* __restrict__ vector, int n_batch, float* __restrict__ result); 87 int m_rows, int m_cols, const float* __restrict__ vector, int n_batch, 94 int n_batch, const int32_t input_offset, const int32_t output_multiplier, 102 const float* scaling_factors, int n_batch, float* __restrict__ result); 110 int v_size, int n_batch, 114 const int16_t* vector, int v_size, const int16_t* batch_vector, int n_batch, [all …]
|
D | portable_tensor_utils.cc | 122 int n_batch, float* result) { in PortableMatrixBatchVectorMultiplyAccumulate() argument 124 for (int b = 0; b < n_batch; b++) { in PortableMatrixBatchVectorMultiplyAccumulate() 141 int n_batch, float* __restrict__ result) { in PortableMatrixBatchVectorMultiplyAccumulate() argument 142 for (int batch = 0; batch < n_batch; ++batch, vectors += m_cols) { in PortableMatrixBatchVectorMultiplyAccumulate() 166 int n_batch, float* __restrict__ result, const float* per_channel_scale, in PortableMatrixBatchVectorMultiplyAccumulate() argument 171 matrix, m_rows, m_cols, vectors, scaling_factors, n_batch, result); in PortableMatrixBatchVectorMultiplyAccumulate() 181 for (int batch = 0; batch < n_batch; ++batch, vectors += m_cols) { in PortableMatrixBatchVectorMultiplyAccumulate() 209 const float* __restrict__ vector, int n_batch, float* __restrict__ result) { in PortableSparseMatrixBatchVectorMultiplyAccumulate1x4() argument 212 for (int batch = 0; batch < n_batch; batch++) { in PortableSparseMatrixBatchVectorMultiplyAccumulate1x4() 234 int n_batch, const int32_t input_offset, const int32_t output_multiplier, in PortableSparseMatrixBatchVectorMultiplyAccumulate1x16() argument [all …]
|
D | svdf.h | 77 const int n_batch = input_shape.Dims(0); in EvalIntegerSVDF() local 86 std::copy(state_data + 1, state_data + n_batch * n_memory * n_filter, in EvalIntegerSVDF() 95 for (int b = 0; b < n_batch; b++) { in EvalIntegerSVDF() 120 for (int b = 0; b < n_batch; ++b) { in EvalIntegerSVDF() 133 n_batch * n_unit, n_rank); in EvalIntegerSVDF() 136 tensor_utils::VectorBatchVectorAdd(bias_data, n_unit, n_batch, in EvalIntegerSVDF() 142 for (int i = 0; i < n_batch * n_unit; ++i) { in EvalIntegerSVDF()
|
/external/tensorflow/tensorflow/lite/kernels/internal/optimized/ |
D | sse_tensor_utils.h | 39 int n_batch, float* result) { in MatrixBatchVectorMultiplyAccumulate() argument 42 n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 45 vector, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 52 const float* __restrict__ scaling_factors, int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 55 vectors, scaling_factors, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 61 int n_batch, float* __restrict__ result, const float* per_channel_scale, in MatrixBatchVectorMultiplyAccumulate() argument 65 vectors, scaling_factors, n_batch, result, per_channel_scale, in MatrixBatchVectorMultiplyAccumulate() 72 const float* __restrict__ scaling_factors, int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 76 vectors, scaling_factors, n_batch, scratch, result, context); in MatrixBatchVectorMultiplyAccumulate() 82 const float* __restrict__ vector, int n_batch, float* __restrict__ result) { in SparseMatrixBatchVectorMultiplyAccumulate1x4() argument [all …]
|
D | neon_tensor_utils.h | 29 int n_batch, float* result) { in MatrixBatchVectorMultiplyAccumulate() argument 31 vector, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 38 int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 41 vectors, scaling_factors, n_batch, result); in MatrixBatchVectorMultiplyAccumulate() 48 int n_batch, int32_t* scratch, in MatrixBatchVectorMultiplyAccumulate() argument 52 vectors, scaling_factors, n_batch, scratch, result, context); in MatrixBatchVectorMultiplyAccumulate() 58 int n_batch, float* __restrict__ result, const float* per_channel_scale, in MatrixBatchVectorMultiplyAccumulate() argument 62 vectors, scaling_factors, n_batch, result, per_channel_scale, in MatrixBatchVectorMultiplyAccumulate() 69 const float* __restrict__ vector, int n_batch, float* __restrict__ result) { in SparseMatrixBatchVectorMultiplyAccumulate1x4() argument 71 segments, indices, m_rows, m_cols, vector, n_batch, result); in SparseMatrixBatchVectorMultiplyAccumulate1x4() [all …]
|
D | neon_tensor_utils_impl.h | 34 int n_batch, float* result); 41 int n_batch, 50 int n_batch, int32_t* scratch, 58 int n_batch, float* __restrict__ result, const float* per_channel_scale, 65 int n_batch, int n_input, int16_t* output); 67 void NeonApplySigmoid(const int16_t* input, int32_t n_batch, int32_t n_input, 70 void NeonApplyTanh(int32_t integer_bits, const int16_t* input, int32_t n_batch, 73 void NeonCwiseMul(const int16_t* input_1, const int16_t* input_2, int n_batch, 77 int32_t multiplier, int shift, int n_batch, int n_input, 80 void NeonCwiseAdd(const int16_t* input_1, const int16_t* input_2, int n_batch, [all …]
|
D | neon_tensor_utils.cc | 227 int n_batch, float* result) { in NeonMatrixBatchVectorMultiplyAccumulate() argument 234 for (int b = 0; b < n_batch; b++) { in NeonMatrixBatchVectorMultiplyAccumulate() 276 const int8_t* ShuffleVectors(const int8_t* vectors, const int n_batch, in ShuffleVectors() argument 279 kNeonVectorAlignment, n_batch * m_cols, shuffled_vectors_free)); in ShuffleVectors() 281 for (int i = 0; i < n_batch; i += 4) { in ShuffleVectors() 331 const int8_t* vectors, const float* scaling_factors, int n_batch, in DotprodMatrixBatchFourVectorMultiplyAccumulate() argument 336 ShuffleVectors(vectors, n_batch, m_cols, &shuffled_vectors_free); in DotprodMatrixBatchFourVectorMultiplyAccumulate() 339 for (int batch = 0; batch < n_batch; batch += 4) { in DotprodMatrixBatchFourVectorMultiplyAccumulate() 459 const int8_t* vectors, const float* scaling_factors, int n_batch, in DotprodMatrixBatchFourVectorMultiplyAccumulate() argument 464 ShuffleVectors(vectors, n_batch, m_cols, &shuffled_vectors_free); in DotprodMatrixBatchFourVectorMultiplyAccumulate() [all …]
|
D | sse_tensor_utils_impl.h | 33 const float* __restrict__ vector, int n_batch, float* __restrict__ result); 39 const float* __restrict__ scaling_factors, int n_batch, 51 const float* __restrict__ scaling_factors, int n_batch, 59 const float* __restrict__ scaling_factors, int n_batch, int32_t* scratch, 66 const float* __restrict__ scaling_factors, int n_batch, 76 const float* __restrict__ scaling_factors, int n_batch,
|
D | sse_tensor_utils.cc | 161 const float* __restrict__ vector, int n_batch, float* __restrict__ result) { in Avx2MatrixBatchVectorMultiplyAccumulateImpl() argument 168 for (int b = 0; b < n_batch; ++b) { in Avx2MatrixBatchVectorMultiplyAccumulateImpl() 202 const float* __restrict__ scaling_factors, int n_batch, in Avx2MatrixBatchVectorMultiplyAccumulateImpl() argument 205 for (std::intptr_t batch = 0; batch < n_batch; ++batch) { in Avx2MatrixBatchVectorMultiplyAccumulateImpl() 294 const float* __restrict__ scaling_factors, int n_batch, in SseMatrixBatchVectorMultiplyAccumulateImpl() argument 299 matrix, m_rows, m_cols, vectors, scaling_factors, n_batch, result, in SseMatrixBatchVectorMultiplyAccumulateImpl() 303 for (std::intptr_t batch = 0; batch < n_batch; ++batch) { in SseMatrixBatchVectorMultiplyAccumulateImpl() 379 const int8_t* input_to_gate_weights, int32_t n_batch, in SseCpuBackendGemm() argument 395 rhs_params.cols = n_batch; in SseCpuBackendGemm() 400 dst_params.cols = n_batch; in SseCpuBackendGemm() [all …]
|
/external/tensorflow/tensorflow/lite/kernels/internal/ |
D | portable_tensor_utils.h | 51 void VectorBatchVectorAdd(const T* vector, int v_size, int n_batch, in VectorBatchVectorAdd() argument 53 for (int b = 0; b < n_batch; b++) { in VectorBatchVectorAdd() 73 const T* batch_vector, int n_batch, in VectorBatchVectorCwiseProduct() argument 75 for (int b = 0; b < n_batch; b++) { in VectorBatchVectorCwiseProduct() 101 int n_batch, T* result) { in VectorBatchVectorCwiseProductAccumulate() argument 102 for (int b = 0; b < n_batch; b++) { in VectorBatchVectorCwiseProductAccumulate() 112 void VectorBatchVectorAssign(const T* vector, int v_size, int n_batch, in VectorBatchVectorAssign() argument 114 for (int b = 0; b < n_batch; b++) { in VectorBatchVectorAssign() 153 inline void BatchQuantizeFloats(const float* float_data_ptr, int n_batch, in BatchQuantizeFloats() argument 157 for (int b = 0; b < n_batch; ++b) { in BatchQuantizeFloats() [all …]
|
D | tensor_utils.h | 45 const float* __restrict__ scaling_factors, int n_batch, 53 int n_batch, float* __restrict__ result, const float* per_channel_scale, 63 const float* vector_scaling_factors, int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 68 for (int b = 0; b < n_batch; ++b) { in MatrixBatchVectorMultiplyAccumulate() 73 scaling_factor_scratch, n_batch, result, in MatrixBatchVectorMultiplyAccumulate() 107 int32_t n_batch, int32_t n_input, int32_t n_output, int32_t output_zp, 138 int32_t n_batch, int32_t n_input, int32_t n_output, int32_t output_zp,
|
/external/tensorflow/tensorflow/lite/kernels/ |
D | lstm_eval.cc | 47 float* output, int m_rows, int m_cols, int n_batch, in MatrixBatchVectorMultiplyAccumulate() argument 56 tflite::RuntimeShape input_shape({n_batch, m_cols}); in MatrixBatchVectorMultiplyAccumulate() 57 tflite::RuntimeShape output_shape({n_batch, m_rows}); in MatrixBatchVectorMultiplyAccumulate() 58 if (n_batch == 1) { in MatrixBatchVectorMultiplyAccumulate() 66 for (int i = 0; i < m_rows * n_batch; ++i) { in MatrixBatchVectorMultiplyAccumulate() 187 const int n_batch, const int n_input, const int n_aux_input, in CalculateLstmGateFloat() argument 198 std::fill_n(gate, n_cell * n_batch, 0.0f); in CalculateLstmGateFloat() 200 tensor_utils::VectorBatchVectorAssign(gate_bias, n_cell, n_batch, gate); in CalculateLstmGateFloat() 208 n_input, n_batch, context); in CalculateLstmGateFloat() 216 n_aux_input, n_batch, context); in CalculateLstmGateFloat() [all …]
|
D | bidirectional_sequence_lstm_test.cc | 32 BidirectionalLSTMOpModel(int n_batch, int n_input, int n_cell, int n_output, in BidirectionalLSTMOpModel() argument 40 : n_batch_(n_batch), in BidirectionalLSTMOpModel() 430 const int n_batch = 1; in TEST_P() local 441 n_batch, n_input, n_cell, n_output, sequence_length, /*use_cifg=*/false, in TEST_P() 447 {sequence_length, n_batch, n_input}, // input tensor in TEST_P() 495 {n_batch, n_output}, // activation_state tensor in TEST_P() 496 {n_batch, n_cell}, // cell_state tensor in TEST_P() 498 {n_batch, n_output}, // activation_state tensor in TEST_P() 499 {n_batch, n_cell}, // cell_state tensor in TEST_P() 501 {sequence_length, n_batch, 0}, // aux_input tensor in TEST_P() [all …]
|
D | unidirectional_sequence_lstm_test.cc | 36 int n_batch, int n_input, int n_cell, int n_output, int sequence_length, in HybridUnidirectionalLSTMOpModel() argument 42 n_batch, n_input, n_cell, n_output, sequence_length, time_major, in HybridUnidirectionalLSTMOpModel() 248 const int n_batch = 1; in TEST_F() local 256 n_batch, n_input, n_cell, n_output, sequence_length, in TEST_F() 262 {sequence_length, n_batch, n_input}, // input tensor in TEST_F() 286 {n_batch, n_output}, // output_state tensor in TEST_F() 287 {n_batch, n_cell}, // cell_state tensor in TEST_F() 310 const int n_batch = 1; in TEST_F() local 318 n_batch, n_input, n_cell, n_output, sequence_length, in TEST_F() 324 {sequence_length, n_batch, n_input}, // input tensor in TEST_F() [all …]
|
D | unidirectional_sequence_gru_test.cc | 34 explicit GRUOpModel(int n_batch, int n_input, int n_output, in GRUOpModel() argument 37 : n_batch_(n_batch), n_input_(n_input), n_output_(n_output) { in GRUOpModel() 40 AddVariableInput(TensorData{TensorType_FLOAT32, {n_batch, n_output}}); in GRUOpModel() 101 const int n_batch = 2; in TEST() local 105 GRUOpModel m(n_batch, n_input, n_output, in TEST() 106 {{n_time, n_batch, n_input}, in TEST() 107 {n_batch, n_output}, in TEST() 133 EXPECT_THAT(m.GetOutputShape(), ElementsAre(n_time, n_batch, n_output)); in TEST()
|
D | gru_cell.cc | 45 const int n_batch = input_shape.Dims(0); in GruCell() local 70 auto r = ru.block(0 * n_output, 0, n_output, n_batch); in GruCell() 71 auto u = ru.block(1 * n_output, 0, n_output, n_batch); in GruCell() 76 auto hr = xh.block(n_input, 0, n_output, n_batch); in GruCell() 88 memcpy(output_state, output, n_batch * n_output * sizeof(float)); in GruCell()
|
D | unidirectional_sequence_gru.cc | 40 const int n_batch = input->dims->data[1]; in GruImpl() local 43 const int n_batch_input = n_batch * n_input; in GruImpl() 44 const int n_batch_output = n_batch * n_output; in GruImpl() 45 const RuntimeShape input_shape({n_batch, n_input}); in GruImpl() 58 const RuntimeShape output_shape = RuntimeShape({n_batch, n_output}); in GruImpl() 142 const int n_batch = input->dims->data[1]; in Prepare() local 150 TF_LITE_ENSURE_EQ(context, input_state->dims->data[0], n_batch); in Prepare() 189 output_size->data[1] = n_batch; in Prepare() 213 activation_size->data[0] = n_batch; in Prepare() 225 concat_size->data[0] = n_batch; in Prepare()
|
D | lstm_test.cc | 39 LSTMOpModel(int n_batch, int n_input, int n_cell, int n_output, bool use_cifg, in LSTMOpModel() argument 46 n_batch_(n_batch), in LSTMOpModel() 48 input_ = AddInput({TensorType_FLOAT32, {n_batch, n_input}}); in LSTMOpModel() 104 AddVariableInput({TensorType_FLOAT32, {n_batch, n_output}}); in LSTMOpModel() 105 AddVariableInput({TensorType_FLOAT32, {n_batch, n_cell}}); in LSTMOpModel() 130 output_ = AddOutput({TensorType_FLOAT32, {n_batch, n_output}}); in LSTMOpModel() 413 const int n_batch = 1; in TEST_P() local 475 LSTMOpModel lstm(n_batch, n_input, n_cell, n_output, in TEST_P() 490 const int n_batch = 1; in TEST_P() local 546 LSTMOpModel lstm(n_batch, n_input, n_cell, n_output, in TEST_P() [all …]
|
D | optional_tensor_test.cc | 31 LSTMOpModel(int n_batch, int n_input, int n_cell, int n_output, bool use_cifg, in LSTMOpModel() argument 35 : n_batch_(n_batch), in LSTMOpModel() 229 const int n_batch = 1; in TEST() local 235 LSTMOpModel lstm(n_batch, n_input, n_cell, n_output, in TEST() 241 {n_batch, n_input}, // input tensor in TEST()
|
D | bidirectional_sequence_lstm.cc | 438 const int n_batch = time_major ? input->dims->data[1] : input->dims->data[0]; in Prepare() local 546 n_batch * n_fw_output); in Prepare() 547 TF_LITE_ENSURE_EQ(context, NumElements(fw_cell_state), n_batch * n_fw_cell); in Prepare() 551 fw_output_size->data[0] = time_major ? max_time : n_batch; in Prepare() 552 fw_output_size->data[1] = time_major ? n_batch : max_time; in Prepare() 585 fw_scratch_buffer_size->data[0] = n_batch; in Prepare() 618 bw_output_size->data[0] = time_major ? max_time : n_batch; in Prepare() 619 bw_output_size->data[1] = time_major ? n_batch : max_time; in Prepare() 629 n_batch * n_bw_output); in Prepare() 630 TF_LITE_ENSURE_EQ(context, NumElements(bw_cell_state), n_batch * n_bw_cell); in Prepare() [all …]
|
/external/tensorflow/tensorflow/lite/tools/optimize/calibration/builtin_logging_ops/ |
D | lstm.cc | 46 const int n_batch, const int n_input, const int n_aux_input, in CalculateLstmGateFloat() argument 58 std::fill_n(gate, n_cell * n_batch, 0.0f); in CalculateLstmGateFloat() 60 tensor_utils::VectorBatchVectorAssign(gate_bias, n_cell, n_batch, gate); in CalculateLstmGateFloat() 66 input_to_gate_weights, n_cell, n_input, input, n_batch, gate); in CalculateLstmGateFloat() 73 aux_input, n_batch, gate); in CalculateLstmGateFloat() 77 recurrent_to_gate_weights, n_cell, n_output, output_state, n_batch, gate); in CalculateLstmGateFloat() 81 cell_to_gate_weights, n_cell, cell_state, n_batch, gate); in CalculateLstmGateFloat() 86 n_cell * n_batch, error_reporter); in CalculateLstmGateFloat() 88 tensor_utils::MeanStddevNormalization(gate, gate, n_cell, n_batch); in CalculateLstmGateFloat() 90 gate, n_batch, gate); in CalculateLstmGateFloat() [all …]
|
/external/tensorflow/tensorflow/lite/delegates/gpu/cl/kernels/ |
D | lstm_full_test.cc | 35 LSTMOpModel(int n_batch, int n_input, int n_cell, int n_output, bool use_cifg, in LSTMOpModel() argument 63 n_batch_(n_batch), in LSTMOpModel() 65 input_ = AddInput({TensorType_FLOAT32, {n_batch, n_input}}); in LSTMOpModel() 124 AddVariableInput({TensorType_FLOAT32, {n_batch, n_output}}); in LSTMOpModel() 125 AddVariableInput({TensorType_FLOAT32, {n_batch, n_cell}}); in LSTMOpModel() 150 output_ = AddOutput({TensorType_FLOAT32, {n_batch, n_output}}); in LSTMOpModel() 265 const int n_batch = 1; in TEST_P() local 319 n_batch, n_input, n_cell, n_output, in TEST_P() 339 const int n_batch = 1; in TEST_P() local 387 n_batch, n_input, n_cell, n_output, in TEST_P() [all …]
|
/external/armnn/src/backends/reference/workloads/ |
D | LstmUtils.cpp | 43 uint32_t n_batch, in MeanStddevNormalization() argument 46 for (uint32_t batch = 0; batch < n_batch; ++batch) { in MeanStddevNormalization() 72 output_vector -= v_size * n_batch; in MeanStddevNormalization() 73 input_vector -= v_size * n_batch; in MeanStddevNormalization()
|