commit 1299496996bc332f02218f926640a9255ed60310
author Mike Kelly <mike.kelly@arm.com> Thu Apr 21 11:57:09 2022 +0100
committer mike.kelly <mike.kelly@arm.com> Thu May 05 08:29:20 2022 +0000
tree 2d242e142bd8fe7387140bcf8cdf39cd13ffc9eb
parent 8272a7bda2974c39b6c06e3eb3a000f2bdb749f7

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

src/backends/backendsCommon/WorkloadData.cpp

diff --git a/src/backends/backendsCommon/WorkloadData.cpp b/src/backends/backendsCommon/WorkloadData.cpp
index 70d7641..a2dcd63 100644
--- a/src/backends/backendsCommon/WorkloadData.cpp
+++ b/src/backends/backendsCommon/WorkloadData.cpp
@@ -3860,38 +3860,20 @@
     {
         throw InvalidArgumentException(descriptorName + ": Invalid number of inputs.");
     }
-    if (workloadInfo.m_OutputTensorInfos.size() != 1)
+    if (workloadInfo.m_OutputTensorInfos.size() != 3)
     {
         throw InvalidArgumentException(descriptorName + ": Invalid number of outputs.");
     }
 
     std::vector<DataType> supportedTypes =
     {
-        DataType::Float32
+        DataType::Float32,
+        DataType::QAsymmS8
     };
 
     // check for supported type of one input and match them with all the other input and output
     ValidateDataTypes(workloadInfo.m_InputTensorInfos[0], supportedTypes, descriptorName);
 
-    // type matches all other inputs
-    for (uint32_t i = 1u; i < workloadInfo.m_InputTensorInfos.size(); ++i)
-    {
-        ValidateTensorDataTypesMatch(workloadInfo.m_InputTensorInfos[0],
-                                     workloadInfo.m_InputTensorInfos[i],
-                                     descriptorName,
-                                     "input_0",
-                                     "input_" + std::to_string(i));
-    }
-    // type matches all other outputs
-    for (uint32_t i = 0u; i < workloadInfo.m_OutputTensorInfos.size(); ++i)
-    {
-        ValidateTensorDataTypesMatch(workloadInfo.m_InputTensorInfos[0],
-                                     workloadInfo.m_OutputTensorInfos[i],
-                                     "LstmQueueDescriptor",
-                                     "input_0",
-                                     "output_" + std::to_string(i));
-    }
-
     // Making sure clipping parameters have valid values.
     // == 0 means no clipping
     //  > 0 means clipping
@@ -3936,7 +3918,7 @@
                                 descriptorName + " input_2");
 
     // outputTensor
-    ValidateTensorNumDimNumElem(workloadInfo.m_OutputTensorInfos[0], 3, (timeStep * n_batch * n_output),
+    ValidateTensorNumDimNumElem(workloadInfo.m_OutputTensorInfos[2], 3, (timeStep * n_batch * n_output),
                                 descriptorName + " output_0");
 
     // check that dimensions of inputs/outputs and QueueDescriptor data match with each other