TIM-VX/src/tim/vx/ops/maxpoolgrad.cc

202 lines
8.0 KiB
C++

/****************************************************************************
*
* Copyright (c) 2020-2023 Vivante Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
*****************************************************************************/
#include "tim/vx/ops.h"
#include "vsi_nn_pub.h"
#include "op_impl.h"
#include <array>
#ifdef VSI_FEAT_OP_MAXPOOLWITHARGMAX
namespace tim {
namespace vx {
namespace ops {
class MaxpoolGradImpl : public OpImpl {
public:
enum {
POOL_INPUT_TENSOR = 0,
GRADIENT_TENSOR = 1,
INPUT_CNT = 2,
UPDATED_TENSOR = 0,
OUTPUT_CNT = 1,
};
MaxpoolGradImpl(Graph* graph, PadType padding,
const std::array<uint32_t, 2>& ksize,
const std::array<uint32_t, 2>& stride,
int input_cnt, int output_cnt,
RoundType round_type,
DataLayout layout)
: OpImpl(graph, -1, input_cnt, output_cnt, layout),
padding_(padding),
ksize_(ksize),
stride_(stride),
round_type_(round_type) {
maxpoolwithargmax2_ =
graph->CreateOperation<tim::vx::ops::MaxpoolWithArgmax2>(
padding_, ksize_, stride_, round_type_, layout_);
}
~MaxpoolGradImpl() {}
MaxpoolGradImpl& BindInput(const std::shared_ptr<Tensor>& tensor) override {
in_tensors_[input_tensor_index] = tensor;
if (this->input_tensor_index == INPUT_CNT - 1) {
tim::vx::ShapeType in_shape = in_tensors_[POOL_INPUT_TENSOR]->GetShape();
tim::vx::ShapeType grad_shape = in_tensors_[GRADIENT_TENSOR]->GetShape();
tim::vx::ShapeType idx_flattened_shape({CalFlattenedShape(grad_shape)});
tim::vx::ShapeType out_flattened_shape({CalFlattenedShape(in_shape)});
auto in_type = in_tensors_[POOL_INPUT_TENSOR]->GetDataType();
auto in_quant = in_tensors_[POOL_INPUT_TENSOR]->GetQuantization();
if (in_quant.Type() != tim::vx::QuantType::NONE) {
VSILOGW("MaxPoolGrad deal with quantization tensor not validate yet!");
}
tim::vx::TensorSpec pool_out_spec_values(in_type,
grad_shape, tim::vx::TensorAttribute::TRANSIENT, in_quant);
tim::vx::TensorSpec pool_out_spec_indices(tim::vx::DataType::INT32,
grad_shape, tim::vx::TensorAttribute::TRANSIENT);
tim::vx::TensorSpec idx_flattened_spec(tim::vx::DataType::INT32,
idx_flattened_shape,tim::vx::TensorAttribute::TRANSIENT);
tim::vx::TensorSpec upd_flattened_spec(in_type,
idx_flattened_shape, tim::vx::TensorAttribute::TRANSIENT, in_quant);
tim::vx::TensorSpec out_flattened_spec(in_type,
out_flattened_shape, tim::vx::TensorAttribute::TRANSIENT, in_quant);
auto pool_out_values_tensor = graph_->CreateTensor(pool_out_spec_values);
auto pool_out_indices_tensor =
graph_->CreateTensor(pool_out_spec_indices);
auto idx_flattened_tensor = graph_->CreateTensor(idx_flattened_spec);
auto upd_flattened_tensor = graph_->CreateTensor(upd_flattened_spec);
auto out_flattened_tensor = graph_->CreateTensor(out_flattened_spec);
(*maxpoolwithargmax2_).BindInput(in_tensors_[POOL_INPUT_TENSOR])
.BindOutputs({pool_out_values_tensor, pool_out_indices_tensor});
// eliminate pool out of maxpoolwithargmax begin
tim::vx::TensorSpec sliced_spec(in_type,
{1, 1, 1, 1}, tim::vx::TensorAttribute::TRANSIENT, in_quant);
auto sliced_tensor = graph_->CreateTensor(sliced_spec);
auto one_zero_tensor = graph_->CreateTensor(sliced_spec);
auto grad_tensor = graph_->CreateTensor(pool_out_spec_values);
std::vector<int32_t> start = {0, 0, 0, 0};
std::vector<int32_t> length = {1, 1, 1, 1};
auto slice_one =
graph_->CreateOperation<tim::vx::ops::Slice>(0, start, length);
(*slice_one).BindInput(pool_out_values_tensor).BindOutput(sliced_tensor);
auto self_sub = graph_->CreateOperation<tim::vx::ops::Sub>();
(*self_sub).BindInputs({sliced_tensor, sliced_tensor})
.BindOutput(one_zero_tensor);
auto add_zeros = graph_->CreateOperation<tim::vx::ops::Add>();
(*add_zeros).BindInputs({one_zero_tensor, in_tensors_[GRADIENT_TENSOR]})
.BindOutput(grad_tensor);
// eliminate pool out of maxpoolwithargmax end
auto flatten_idx =
graph_->CreateOperation<tim::vx::ops::Reshape>(idx_flattened_shape);
(*flatten_idx).BindInput(pool_out_indices_tensor)
.BindOutput(idx_flattened_tensor);
auto flatten_upd =
graph_->CreateOperation<tim::vx::ops::Reshape>(idx_flattened_shape);
(*flatten_upd).BindInput(grad_tensor).BindOutput(upd_flattened_tensor);
auto scatternd =
graph_->CreateOperation<tim::vx::ops::ScatterND>(out_flattened_shape);
(*scatternd).BindInputs({idx_flattened_tensor, upd_flattened_tensor})
.BindOutput(out_flattened_tensor);
reshape_like_input_ =
graph_->CreateOperation<tim::vx::ops::Reshape>(in_shape);
(*reshape_like_input_).BindInput(out_flattened_tensor);
}
this->input_tensor_index++;
return *this;
}
MaxpoolGradImpl& BindOutput(const std::shared_ptr<Tensor>& tensor) override {
out_tensors_[output_tensor_index] = tensor;
if (this->output_tensor_index == OUTPUT_CNT - 1) {
(*reshape_like_input_).BindOutput(out_tensors_[UPDATED_TENSOR]);
}
this->output_tensor_index++;
return *this;
}
vsi_nn_node_t* node() override { return nullptr; }
std::vector<std::shared_ptr<Tensor>> InputsTensor() override {
return inputs_tensor_;
}
std::vector<std::shared_ptr<Tensor>> OutputsTensor() override {
return outputs_tensor_;
}
private:
const PadType padding_;
const std::array<uint32_t, 2> ksize_;
const std::array<uint32_t, 2> stride_;
const RoundType round_type_;
std::shared_ptr<tim::vx::Operation> maxpoolwithargmax2_;
std::shared_ptr<tim::vx::Operation> reshape_like_input_;
std::array<std::shared_ptr<tim::vx::Tensor>, INPUT_CNT> in_tensors_;
std::array<std::shared_ptr<tim::vx::Tensor>, OUTPUT_CNT> out_tensors_;
uint32_t CalFlattenedShape(const tim::vx::ShapeType& shape) {
uint32_t out = 1;
for(auto& x: shape) {
out *= x;
}
return out;
}
};
MaxpoolGrad::MaxpoolGrad(Graph* graph, PadType padding,
const std::array<uint32_t, 2>& ksize,
const std::array<uint32_t, 2>& stride,
RoundType round_type,
DataLayout layout)
: padding_(padding),
ksize_(ksize),
stride_(stride),
round_type_(round_type) {
impl_ = std::make_unique<MaxpoolGradImpl>(graph, padding, ksize, stride, 2, 1, round_type, layout);
}
std::shared_ptr<Operation> MaxpoolGrad::Clone(
std::shared_ptr<Graph>& graph) const {
return graph->CreateOperation<MaxpoolGrad>(
this->padding_, this->ksize_, this->stride_, this->round_type_,
this->impl_->layout_);
}
} // namespace ops
} // namespace vx
} // namespace tim
#endif //(VSI_FEAT_OP_MAXPOOLWITHARGMAX)