diff options
Diffstat (limited to 'src/core/CL/kernels/CLReductionOperationKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLReductionOperationKernel.cpp | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/core/CL/kernels/CLReductionOperationKernel.cpp b/src/core/CL/kernels/CLReductionOperationKernel.cpp index 9f498b8273..db4850f14e 100644 --- a/src/core/CL/kernels/CLReductionOperationKernel.cpp +++ b/src/core/CL/kernels/CLReductionOperationKernel.cpp @@ -47,7 +47,14 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, u { ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); + if(input->num_channels() == 1) + { + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); + } + else + { + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); + } ARM_COMPUTE_RETURN_ERROR_ON_MSG(op == ReductionOperation::SUM_SQUARE && input->data_type() == DataType::QASYMM8, "Not supported reduction operation for QASYMM8"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions, "Reduction axis greater than max number of dimensions"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis"); @@ -77,7 +84,7 @@ std::tuple<Status, Window> validate_and_configure_window(ITensorInfo *input, ITe output_shape.set(axis, 1); const bool is_arg_min_max = (op == ReductionOperation::ARG_IDX_MIN || op == ReductionOperation::ARG_IDX_MAX); DataType output_data_type = is_arg_min_max ? DataType::U32 : input->data_type(); - auto_init_if_empty(*output, output_shape, 1, output_data_type, input->quantization_info()); + auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_data_type(output_data_type).reset_padding().set_is_resizable(true)); const unsigned int num_elems_processed_per_iteration = (is_data_type_quantized(input->data_type()) && (axis == 0)) ? 1 : 16; Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration)); @@ -160,6 +167,7 @@ void CLReductionOperationKernel::configure(const ICLTensor *input, ICLTensor *ou build_opts.add_option_if(op == ReductionOperation::ARG_IDX_MAX, "-DARG_MAX"); build_opts.add_option_if(op == ReductionOperation::ARG_IDX_MIN, "-DARG_MIN"); build_opts.add_option_if(op == ReductionOperation::PROD, "-DPROD"); + build_opts.add_option_if(input->info()->num_channels() == 2, "-DCOMPLEX"); switch(op) { |