diff options
Diffstat (limited to 'libs/kernel/acl/src/neon/Softmax.cpp')
-rw-r--r-- | libs/kernel/acl/src/neon/Softmax.cpp | 77 |
1 files changed, 77 insertions, 0 deletions
diff --git a/libs/kernel/acl/src/neon/Softmax.cpp b/libs/kernel/acl/src/neon/Softmax.cpp new file mode 100644 index 000000000..79d614418 --- /dev/null +++ b/libs/kernel/acl/src/neon/Softmax.cpp @@ -0,0 +1,77 @@ +/* + * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <OperationsUtils.h> +#include <NeuralNetworks.h> + +#include <arm_compute/core/TensorShape.h> +#include <arm_compute/core/TensorInfo.h> +#include "../IO_accessor.h" +#include "../shape.h" +#include "../util.h" +#include "../NEUniqueTensor.h" + +namespace nnfw { +namespace kernel { +namespace acl { +namespace neon { + +bool softmaxFloat32(const float* inputData, const nnfw::rt::Shape& inputShape, + const float beta, + float* outputData, const nnfw::rt::Shape& outputShape) +{ + arm_compute::TensorShape input_shape = util::fromNNShape(inputShape); + arm_compute::TensorShape output_shape = util::fromNNShape(outputShape); + + NEUniqueTensor input(arm_compute::TensorInfo(input_shape, arm_compute::Format::F32)); + NEUniqueTensor output(arm_compute::TensorInfo(output_shape, arm_compute::Format::F32)); + + auto softmax_f = std::make_shared<arm_compute::NESoftmaxLayer>(); + softmax_f->configure(input.ptr(), output.ptr(), beta); + + input.allocate(); + output.allocate(); + + if (inputShape.dimensions.size() == 4) + { + TensorAccess<InputAccessor>(input.ref(), inputData, inputShape); + + softmax_f->run(); + + TensorAccess<OutputAccessor>(output.ref(), outputData, outputShape); + } + else if (inputShape.dimensions.size() == 2) + { + // Softmax comes with 1xN matrix and this is translated to N vector in arm_compute::TensorShape + TensorAccess<VectorInputAccessor>(input.ref(), inputData, inputShape); + + softmax_f->run(); + + TensorAccess<VectorOutputAccessor>(output.ref(), outputData, outputShape); + } + else + { + assert("undefined dimension of input" && 0); + return false; + } + + return true; +} + +} // namespace neon +} // namespace acl +} // namespace kernel +} // namespace nnfw |