diff options
| author | Mike Kelly <mike.kelly@arm.com> | 2022-04-21 11:57:09 +0100 |
|---|---|---|
| committer | mike.kelly <mike.kelly@arm.com> | 2022-05-05 08:29:20 +0000 |
| commit | 1299496996bc332f02218f926640a9255ed60310 (patch) | |
| tree | 2d242e142bd8fe7387140bcf8cdf39cd13ffc9eb /src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp | |
| parent | 8272a7bda2974c39b6c06e3eb3a000f2bdb749f7 (diff) | |
| download | armnn-1299496996bc332f02218f926640a9255ed60310.tar.gz | |
IVGCVSW-6806 Add Unidirectional Sequence Lstm support to Neon
* Corrected TensorInfo order for IsUnidirectionalSequenceLstmSupported
* outputStateOut TensorInfo is not optional.
* cellStateOut TensorInfo is not optional.
* TensorInfo Order matches other QLSTM/LSTM layers.
* Added missing parameters to UnidirectionalSequenceLstmOperator for
delegate.
* Added quantized UnidirectionalSequenceLstm support to Neon
!android-nn-driver:7457
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I26dde1bb96793dd25eb9081ca5ae5f63752288c4
Diffstat (limited to 'src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp')
| -rw-r--r-- | src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp b/src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp index cc9aea8486..ac24120804 100644 --- a/src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp +++ b/src/backends/cl/workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp @@ -41,17 +41,17 @@ ClUnidirectionalSequenceLstmFloatWorkload::ClUnidirectionalSequenceLstmFloatWork GetGuid()); const arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor(); - arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor(); + arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[2])->GetTensor(); TensorInfo inputInfo = info.m_InputTensorInfos[0]; - TensorInfo outputInfo = info.m_OutputTensorInfos[0]; + TensorInfo outputInfo = info.m_OutputTensorInfos[2]; arm_compute::DataType armComputeDataType = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetDataType(); armnn::DataType armnnDataType = GetArmNNDataType(armComputeDataType); TensorShape inputLayerShape = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetShape(); TensorShape cellStateLayerShape = static_cast<IClTensorHandle*>(m_Data.m_Inputs[2])->GetShape(); - TensorShape outputLayerShape = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetShape(); + TensorShape outputLayerShape = static_cast<IClTensorHandle*>(m_Data.m_Outputs[2])->GetShape(); unsigned int maxTime = m_Data.m_Parameters.m_TimeMajor ? inputLayerShape[0] : inputLayerShape[1]; unsigned int batchSize = m_Data.m_Parameters.m_TimeMajor ? inputLayerShape[1] : inputLayerShape[0]; |