diff options
Diffstat (limited to 'src/backends/cl/workloads/ClReduceWorkload.cpp')
| -rw-r--r-- | src/backends/cl/workloads/ClReduceWorkload.cpp | 34 |
1 files changed, 19 insertions, 15 deletions
diff --git a/src/backends/cl/workloads/ClReduceWorkload.cpp b/src/backends/cl/workloads/ClReduceWorkload.cpp index 6f594ff7a9..18415c4cba 100644 --- a/src/backends/cl/workloads/ClReduceWorkload.cpp +++ b/src/backends/cl/workloads/ClReduceWorkload.cpp @@ -19,24 +19,28 @@ arm_compute::Status ClReduceWorkloadValidate(const TensorInfo& input, const TensorInfo& output, const ReduceDescriptor& desc) { - const arm_compute::TensorInfo aclInputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(input); - const arm_compute::TensorInfo aclOutputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(output); - if (!desc.m_vAxis.empty() && desc.m_vAxis.size() > 1) + if ( desc.m_vAxis.size()==1 || desc.m_vAxis.empty()) { - return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, - "ClReduceWorkload: Reduction is supported only on 1 axis."); - } - - arm_compute::Coordinates coords = BuildArmComputeReductionCoordinates(aclInputInfo.num_dimensions(), - input.GetNumDimensions(), - desc.m_vAxis); + const arm_compute::TensorInfo aclInputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(input); + const arm_compute::TensorInfo aclOutputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(output); + arm_compute::Coordinates coords = BuildArmComputeReductionCoordinates(aclInputInfo.num_dimensions(), + input.GetNumDimensions(), + desc.m_vAxis); - return arm_compute::CLReductionOperation::validate(&aclInputInfo, - &aclOutputInfo, - static_cast<unsigned int>(coords[0]), - ConvertReductionOperationToAcl(desc), - desc.m_KeepDims); + return arm_compute::CLReductionOperation::validate(&aclInputInfo, + &aclOutputInfo, + static_cast<unsigned int>(coords[0]), + ConvertReductionOperationToAcl(desc), + desc.m_KeepDims); + } + else + { + // Validate layer if there are multiple axes. + arm_compute::Status status; + IS_MULTI_AXES_REDUCE_SUPPORTED(NeonReduceWorkloadValidate, input, desc, status); + return status; + } } ClReduceWorkload::ClReduceWorkload(const ReduceQueueDescriptor& descriptor, const WorkloadInfo& info) |