Skip to content

Commit

Permalink
revert dconv
Browse files Browse the repository at this point in the history
  • Loading branch information
@zetwhite
zetwhite committed Aug 7, 2024
1 parent dc9b748 commit bae9407
Show file tree
Hide file tree
Showing 2 changed files with 18 additions and 71 deletions.
77 changes: 13 additions & 64 deletions runtime/onert/backend/train/ops/DepthwiseConvolutionLayer.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,11 +56,9 @@ void DepthwiseConvolutionLayer::configureBackward(IPortableTensor *back_prop_inp

if (activation != ir::Activation::NONE)
{
/*
_act_back_prop_output = std::make_unique<BackPropTensor>(_back_prop_output->get_info());
_act_back_prop_output->setBuffer(
std::make_shared<basic::Allocator>(_act_back_prop_output->total_size()));
*/
}

const int64_t k_packet_size = [&]() {
Expand All @@ -77,20 +75,20 @@ void DepthwiseConvolutionLayer::configureBackward(IPortableTensor *back_prop_inp
}();

const auto incoming_shape = getShape(_back_prop_output);
// const auto filter_shape = getShape(_kernel);
// const int batch = incoming_shape.Dims(0);
const auto filter_shape = getShape(_kernel);
const int batch = incoming_shape.Dims(0);
const int out_depth = incoming_shape.Dims(3);
// const int filter_rows = filter_shape.Dims(1);
// const int filter_cols = filter_shape.Dims(2);
const int filter_rows = filter_shape.Dims(1);
const int filter_cols = filter_shape.Dims(2);

// const int filter_spatial_size = filter_rows * filter_cols;
// const int padded_filter_inner_dim_size =
// ((out_depth + k_packet_size - 1) / k_packet_size) * k_packet_size;
const int filter_spatial_size = filter_rows * filter_cols;
const int padded_filter_inner_dim_size =
((out_depth + k_packet_size - 1) / k_packet_size) * k_packet_size;

_use_padded_filter = (out_depth % k_packet_size) == 0 ? false : true;

// prepare padded_filter buffer for cker
/* auto padded_filter_info = ir::OperandInfo(_kernel->get_info());
auto padded_filter_info = ir::OperandInfo(_kernel->get_info());
padded_filter_info.shape({batch, filter_spatial_size, padded_filter_inner_dim_size});
_padded_filter = std::make_unique<Tensor>(padded_filter_info);
_padded_filter->setBuffer(std::make_shared<basic::Allocator>(_padded_filter->total_size()));
Expand All @@ -110,55 +108,6 @@ void DepthwiseConvolutionLayer::configureBackward(IPortableTensor *back_prop_inp
_filter_dim_buffers = std::make_unique<Tensor>(filter_dim_buffers_info);
_filter_dim_buffers->setBuffer(
std::make_shared<basic::Allocator>(_filter_dim_buffers->total_size()));
*/
}

ExtraTensorRequests DepthwiseConvolutionLayer::requestExtraTensors()
{
ExtraTensorRequests reqs;

reqs.push_back(ExtraTensorRequest::createLike(_back_prop_output, &_act_back_prop_output));

const auto incoming_shape = getShape(_back_prop_output);
const auto batch = incoming_shape.Dims(0);
const auto depth = incoming_shape.Dims(3);

const auto filter_shape = getShape(_kernel);
const int filter_rows = filter_shape.Dims(1);
const int filter_cols = filter_shape.Dims(2);
const int filter_spatial_size = filter_rows * filter_cols;

const auto k_packet_size = _dconv_kernel->kPacketSize<float>();
const int padded_filter_inner_dim_size =
((depth + k_packet_size - 1) / k_packet_size) * k_packet_size;

const int thread_count = _dconv_kernel->getThreadCount();

// _padded_filter
{
auto type_info = _kernel->get_info().typeInfo();
ir::Shape shape({batch, filter_spatial_size, padded_filter_inner_dim_size});
auto info = ir::OperandInfo::createStaticInfo(shape, type_info);
reqs.emplace_back(info, ExtraTensorLifeTime::BACKWARD, &_padded_filter);
}

// _filter_buffers
{
auto type_info = _kernel->get_info().typeInfo();
ir::Shape shape({thread_count, filter_spatial_size, padded_filter_inner_dim_size});
auto info = ir::OperandInfo::createStaticInfo(shape, type_info);
reqs.emplace_back(info, ExtraTensorLifeTime::BACKWARD, &_filter_buffers);
}

// _filter_dim_buffers
{
auto type = _back_prop_input->get_info().typeInfo();
ir::Shape shape({thread_count, padded_filter_inner_dim_size});
auto info = ir::OperandInfo::createStaticInfo(shape, type);
reqs.emplace_back(info, ExtraTensorLifeTime::BACKWARD, &_filter_dim_buffers);
}

return reqs;
}

void DepthwiseConvolutionLayer::forward(bool) { cpu::ops::DepthwiseConvolutionLayer::run(); }
Expand Down Expand Up @@ -187,7 +136,7 @@ void DepthwiseConvolutionLayer::backwardFloat32()
try
{
backprop_act =
backpropActivation(_activation, _output, _back_prop_output, _act_back_prop_output);
backpropActivation(_activation, _output, _back_prop_output, _act_back_prop_output.get());
}
catch (const std::exception &e)
{
Expand All @@ -207,15 +156,15 @@ void DepthwiseConvolutionLayer::backwardFloat32()
// Calculate gradient for input
nnfw::cker::train::backpropInput(
dconv_params, getShape(backprop_act), getBuffer<float>(backprop_act), getShape(_kernel),
getBuffer<float>(_kernel), getBuffer<float>(_padded_filter), getShape(_back_prop_input),
getBuffer<float>(_back_prop_input), _use_padded_filter, getBuffer<float>(_filter_buffers),
getBuffer<float>(_filter_dim_buffers));
getBuffer<float>(_kernel), getBuffer<float>(_padded_filter.get()), getShape(_back_prop_input),
getBuffer<float>(_back_prop_input), _use_padded_filter, getBuffer<float>(_filter_buffers.get()),
getBuffer<float>(_filter_dim_buffers.get()));

// Calculate gradient for weights
nnfw::cker::train::backpropFilter(
dconv_params, getShape(backprop_act), getBuffer<float>(backprop_act), getShape(_input),
getBuffer<float>(_input), getShape(_grad_weights), getBuffer<float>(_grad_weights),
getBuffer<float>(_padded_filter), getBuffer<float>(_filter_buffers));
getBuffer<float>(_padded_filter.get()), getBuffer<float>(_filter_buffers.get()));

// Calculate gradient for bias
if (_bias)
Expand Down
12 changes: 5 additions & 7 deletions runtime/onert/backend/train/ops/DepthwiseConvolutionLayer.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -42,8 +42,6 @@ class DepthwiseConvolutionLayer : public ::onert::exec::train::ITrainableFunctio
void configureBackward(IPortableTensor *back_prop_input, IPortableTensor *grad_weights,
IPortableTensor *grad_bias, const IPortableTensor *back_prop_output,
const ir::Activation activation);

ExtraTensorRequests requestExtraTensors() override;
void forward(bool training) override;
void backward() override;

Expand All @@ -56,12 +54,12 @@ class DepthwiseConvolutionLayer : public ::onert::exec::train::ITrainableFunctio
IPortableTensor *_back_prop_input;
const IPortableTensor *_back_prop_output;

ExtraTensor *_act_back_prop_output;

// TODO Consider if these tensors should be built in TensorBuilder
std::unique_ptr<BackPropTensor> _act_back_prop_output;
bool _use_padded_filter;
ExtraTensor *_padded_filter;
ExtraTensor *_filter_buffers;
ExtraTensor *_filter_dim_buffers;
std::unique_ptr<Tensor> _padded_filter;
std::unique_ptr<Tensor> _filter_buffers;
std::unique_ptr<Tensor> _filter_dim_buffers;
};

} // namespace ops
Expand Down

0 comments on commit bae9407

Please sign in to comment.