diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2017-11-22 12:53:45 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
| commit | ee8be2d09573d4ec9960d8f2b947ef38716f0c62 (patch) | |
| tree | 516110d2f92398705b4d79f8af12114e2bc088fd /src/runtime/CL/functions/CLSoftmaxLayer.cpp | |
| parent | 4d55e0a3e848db25496b31529f4405bee7115cf8 (diff) | |
| download | ComputeLibrary-ee8be2d09573d4ec9960d8f2b947ef38716f0c62.tar.gz | |
COMPMID-556: Set CLSoftmaxLayer temp buffers to resizable.
Auto-padding configuration lead the clone of input tensor info to be not
resizable in computer_vision.
Explicitly sets the tensor info clones to resizable.
Change-Id: I8ec3d430127e6ed5cca7884c844654b6e71b7ee5
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110194
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/CL/functions/CLSoftmaxLayer.cpp')
| -rw-r--r-- | src/runtime/CL/functions/CLSoftmaxLayer.cpp | 15 |
1 files changed, 9 insertions, 6 deletions
diff --git a/src/runtime/CL/functions/CLSoftmaxLayer.cpp b/src/runtime/CL/functions/CLSoftmaxLayer.cpp index d39335b6e5..f74b778e95 100644 --- a/src/runtime/CL/functions/CLSoftmaxLayer.cpp +++ b/src/runtime/CL/functions/CLSoftmaxLayer.cpp @@ -41,17 +41,20 @@ CLSoftmaxLayer::CLSoftmaxLayer(std::shared_ptr<IMemoryManager> memory_manager) void CLSoftmaxLayer::configure(const ICLTensor *input, ICLTensor *output, float beta) { - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32); + // Perform validation step + ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); + ARM_COMPUTE_ERROR_THROW_ON(CLSoftmaxLayer::validate(input->info(), output->info())); // Create intermediate tensors shapes - DataType tmp_data_type = is_data_type_quantized_asymmetric(input->info()->data_type()) ? DataType::S32 : input->info()->data_type(); - TensorInfo tensor_info_tmp(input->info()->clone()->set_data_type(tmp_data_type).reset_padding()); + const TensorInfo input_info = input->info()->clone()->reset_padding().set_is_resizable(true); + DataType tmp_data_type = is_data_type_quantized_asymmetric(input->info()->data_type()) ? DataType::S32 : input->info()->data_type(); + TensorInfo tensor_info_tmp(input_info.clone()->set_data_type(tmp_data_type)); _tmp.allocator()->init(tensor_info_tmp); TensorShape max_sum_shape = input->info()->tensor_shape(); max_sum_shape.set(0, 1); - _max.allocator()->init(input->info()->clone()->set_tensor_shape(max_sum_shape).reset_padding()); - _sum.allocator()->init(input->info()->clone()->set_tensor_shape(max_sum_shape).set_data_type(tmp_data_type).reset_padding()); + _max.allocator()->init(input_info.clone()->set_tensor_shape(max_sum_shape)); + _sum.allocator()->init(input_info.clone()->set_tensor_shape(max_sum_shape).set_data_type(tmp_data_type)); // Set GPU target to kernels _max_shift_exp_sum_kernel.set_target(CLScheduler::get().target()); @@ -83,7 +86,7 @@ void CLSoftmaxLayer::configure(const ICLTensor *input, ICLTensor *output, float Error CLSoftmaxLayer::validate(const ITensorInfo *input, const ITensorInfo *output) { - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input); + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); // Create intermediate tensor info DataType tmp_data_type = is_data_type_quantized_asymmetric(input->data_type()) ? DataType::S32 : input->data_type(); |