aboutsummaryrefslogtreecommitdiff
path: root/src/core
diff options
context:
space:
mode:
Diffstat (limited to 'src/core')
-rw-r--r--src/core/NEON/kernels/NEBatchConcatenateLayerKernel.cpp21
-rw-r--r--src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp21
-rw-r--r--src/core/NEON/kernels/NEHeightConcatenateLayerKernel.cpp39
-rw-r--r--src/core/NEON/kernels/NEWidthConcatenateLayerKernel.cpp34
4 files changed, 59 insertions, 56 deletions
diff --git a/src/core/NEON/kernels/NEBatchConcatenateLayerKernel.cpp b/src/core/NEON/kernels/NEBatchConcatenateLayerKernel.cpp
index 65789160f6..c597afd804 100644
--- a/src/core/NEON/kernels/NEBatchConcatenateLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEBatchConcatenateLayerKernel.cpp
@@ -141,21 +141,19 @@ Status validate_arguments(const ITensorInfo *input, unsigned int batch_offset, c
} // namespace
NEBatchConcatenateLayerKernel::NEBatchConcatenateLayerKernel()
- : _func(nullptr), _input(nullptr), _output(nullptr), _batch_offset(0)
+ : _func(nullptr), _batch_offset(0)
{
}
-void NEBatchConcatenateLayerKernel::configure(const ITensor *input, unsigned int batch_offset, ITensor *output)
+void NEBatchConcatenateLayerKernel::configure(const ITensorInfo *input, unsigned int batch_offset, ITensorInfo *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), batch_offset, output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, batch_offset, output));
_func = nullptr;
- _input = input;
- _output = output;
_batch_offset = batch_offset;
- switch(input->info()->data_type())
+ switch(input->data_type())
{
case DataType::S8:
case DataType::U8:
@@ -178,10 +176,10 @@ void NEBatchConcatenateLayerKernel::configure(const ITensor *input, unsigned int
}
// Configure kernel window
- Window win = calculate_max_window(*output->info(), Steps());
+ Window win = calculate_max_window(*output, Steps());
Coordinates coord;
- coord.set_num_dimensions(output->info()->num_dimensions());
- output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+ coord.set_num_dimensions(output->num_dimensions());
+ output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
INEKernel::configure(win);
}
@@ -193,13 +191,14 @@ Status NEBatchConcatenateLayerKernel::validate(const arm_compute::ITensorInfo *i
return Status{};
}
-void NEBatchConcatenateLayerKernel::run(const Window &window, const ThreadInfo &info)
+void NEBatchConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+ const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
ARM_COMPUTE_ERROR_ON(_func == nullptr);
- (*_func)(_input, _output, _batch_offset, window);
+ (*_func)(inputs.at(TensorType::ACL_SRC), outputs.at(TensorType::ACL_DST), _batch_offset, window);
}
} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp b/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
index a95d711f43..49e10de94e 100644
--- a/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
@@ -142,21 +142,19 @@ Status validate_arguments(const ITensorInfo *input, unsigned int depth_offset, c
} // namespace
NEDepthConcatenateLayerKernel::NEDepthConcatenateLayerKernel()
- : _func(nullptr), _input(nullptr), _output(nullptr), _depth_offset(0)
+ : _func(nullptr), _depth_offset(0)
{
}
-void NEDepthConcatenateLayerKernel::configure(const ITensor *input, unsigned int depth_offset, ITensor *output)
+void NEDepthConcatenateLayerKernel::configure(const ITensorInfo *input, unsigned int depth_offset, ITensorInfo *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), depth_offset, output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, depth_offset, output));
_func = nullptr;
- _input = input;
- _output = output;
_depth_offset = depth_offset;
- switch(input->info()->data_type())
+ switch(input->data_type())
{
case DataType::QASYMM8:
_func = &depth_concat<uint8_t>;
@@ -175,11 +173,11 @@ void NEDepthConcatenateLayerKernel::configure(const ITensor *input, unsigned int
}
// Configure kernel window
- Window win = calculate_max_window(*output->info(), Steps());
+ Window win = calculate_max_window(*output, Steps());
Coordinates coord;
- coord.set_num_dimensions(output->info()->num_dimensions());
+ coord.set_num_dimensions(output->num_dimensions());
- output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+ output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
INEKernel::configure(win);
}
@@ -191,13 +189,14 @@ Status NEDepthConcatenateLayerKernel::validate(const arm_compute::ITensorInfo *i
return Status{};
}
-void NEDepthConcatenateLayerKernel::run(const Window &window, const ThreadInfo &info)
+void NEDepthConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+ const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
ARM_COMPUTE_ERROR_ON(_func == nullptr);
- (*_func)(_input, _output, _depth_offset, window);
+ (*_func)(inputs.at(TensorType::ACL_SRC), outputs.at(TensorType::ACL_DST), _depth_offset, window);
}
} // namespace arm_compute
diff --git a/src/core/NEON/kernels/NEHeightConcatenateLayerKernel.cpp b/src/core/NEON/kernels/NEHeightConcatenateLayerKernel.cpp
index 0adf996cca..d4043e02b7 100644
--- a/src/core/NEON/kernels/NEHeightConcatenateLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEHeightConcatenateLayerKernel.cpp
@@ -58,24 +58,23 @@ Status validate_arguments(const ITensorInfo *input, unsigned int height_offset,
} // namespace
NEHeightConcatenateLayerKernel::NEHeightConcatenateLayerKernel()
- : _input(nullptr), _output(nullptr), _height_offset(0)
+ : _height_offset(0)
{
}
-void NEHeightConcatenateLayerKernel::configure(const ITensor *input, unsigned int height_offset, ITensor *output)
+void NEHeightConcatenateLayerKernel::configure(const ITensorInfo *input, unsigned int height_offset, ITensorInfo *output)
{
+ ARM_COMPUTE_UNUSED(input);
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), height_offset, output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, height_offset, output));
- _input = input;
- _output = output;
_height_offset = height_offset;
// Configure kernel window
- Window win = calculate_max_window(*output->info(), Steps());
+ Window win = calculate_max_window(*output, Steps());
Coordinates coord;
- coord.set_num_dimensions(output->info()->num_dimensions());
- output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+ coord.set_num_dimensions(output->num_dimensions());
+ output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
INEKernel::configure(win);
}
@@ -85,30 +84,34 @@ Status NEHeightConcatenateLayerKernel::validate(const ITensorInfo *input, unsign
return Status{};
}
-void NEHeightConcatenateLayerKernel::run(const Window &window, const ThreadInfo &info)
+void NEHeightConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+ const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
+ const auto src = inputs.at(TensorType::ACL_SRC);
+ auto dst = outputs.at(TensorType::ACL_DST);
+
// Offset output pointer to the correct position
- uint8_t *output_ptr = _output->buffer() + _output->info()->offset_first_element_in_bytes() + _height_offset * _output->info()->strides_in_bytes()[Window::DimY];
+ uint8_t *output_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + _height_offset * dst->info()->strides_in_bytes()[Window::DimY];
const auto window_start_x = static_cast<int>(window.x().start());
- const auto window_end_x = static_cast<int>(window.x().end()) * static_cast<int>(_output->info()->element_size());
- const int window_step_x = 16;
+ const auto window_end_x = static_cast<int>(window.x().end()) * static_cast<int>(dst->info()->element_size());
+ const int window_step_x = 16;
Window win{ window };
win.set(Window::DimX, Window::Dimension(0, 1, 1));
- win.set(Window::DimY, Window::Dimension(0, _input->info()->tensor_shape().y(), 1));
+ win.set(Window::DimY, Window::Dimension(0, src->info()->tensor_shape().y(), 1));
// Create iterators
- Iterator input(_input, win);
- Iterator output(_output, win);
+ Iterator input(src, win);
+ Iterator output(dst, win);
- const DataType dt = _input->info()->data_type();
- const UniformQuantizationInfo &input_qinfo = _input->info()->quantization_info().uniform();
- const UniformQuantizationInfo &output_qinfo = _output->info()->quantization_info().uniform();
+ const DataType dt = src->info()->data_type();
+ const UniformQuantizationInfo &input_qinfo = src->info()->quantization_info().uniform();
+ const UniformQuantizationInfo &output_qinfo = dst->info()->quantization_info().uniform();
if(dt == DataType::QASYMM8 && input_qinfo != output_qinfo)
{
execute_window_loop(win, [&](const Coordinates &)
diff --git a/src/core/NEON/kernels/NEWidthConcatenateLayerKernel.cpp b/src/core/NEON/kernels/NEWidthConcatenateLayerKernel.cpp
index f5bdfae5d6..1b32e3614e 100644
--- a/src/core/NEON/kernels/NEWidthConcatenateLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEWidthConcatenateLayerKernel.cpp
@@ -58,24 +58,22 @@ Status validate_arguments(const ITensorInfo *input, unsigned int width_offset, c
} // namespace
NEWidthConcatenateLayerKernel::NEWidthConcatenateLayerKernel()
- : _input(nullptr), _output(nullptr), _width_offset(0)
+ : _width_offset(0)
{
}
-void NEWidthConcatenateLayerKernel::configure(const ITensor *input, unsigned int width_offset, ITensor *output)
+void NEWidthConcatenateLayerKernel::configure(const ITensorInfo *input, unsigned int width_offset, ITensorInfo *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), width_offset, output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, width_offset, output));
- _input = input;
- _output = output;
_width_offset = width_offset;
// Configure kernel window
- Window win = calculate_max_window(*input->info(), Steps());
+ Window win = calculate_max_window(*input, Steps());
Coordinates coord;
- coord.set_num_dimensions(output->info()->num_dimensions());
- output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+ coord.set_num_dimensions(output->num_dimensions());
+ output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
INEKernel::configure(win);
}
@@ -86,28 +84,32 @@ Status NEWidthConcatenateLayerKernel::validate(const ITensorInfo *input, unsigne
return Status{};
}
-void NEWidthConcatenateLayerKernel::run(const Window &window, const ThreadInfo &info)
+void NEWidthConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+ const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
+ const auto src = inputs.at(TensorType::ACL_SRC);
+ auto dst = outputs.at(TensorType::ACL_DST);
+
// Offset output pointer to the correct position
- uint8_t *output_ptr = _output->buffer() + _output->info()->offset_first_element_in_bytes() + _width_offset * _output->info()->strides_in_bytes()[0];
+ uint8_t *output_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + _width_offset * dst->info()->strides_in_bytes()[0];
const auto window_start_x = static_cast<int>(window.x().start());
- const auto window_end_x = static_cast<int>(window.x().end()) * static_cast<int>(_output->info()->element_size());
+ const auto window_end_x = static_cast<int>(window.x().end()) * static_cast<int>(dst->info()->element_size());
constexpr int window_step_x = 16;
Window win{ window };
win.set(Window::DimX, Window::Dimension(0, 1, 1));
// Create iterators
- Iterator input(_input, win);
- Iterator output(_output, win);
- const DataType dt = _input->info()->data_type();
- const UniformQuantizationInfo &input_qinfo = _input->info()->quantization_info().uniform();
- const UniformQuantizationInfo &output_qinfo = _output->info()->quantization_info().uniform();
+ Iterator input(src, win);
+ Iterator output(dst, win);
+ const DataType dt = src->info()->data_type();
+ const UniformQuantizationInfo &input_qinfo = src->info()->quantization_info().uniform();
+ const UniformQuantizationInfo &output_qinfo = dst->info()->quantization_info().uniform();
if(dt == DataType::QASYMM8 && input_qinfo != output_qinfo)
{
execute_window_loop(win, [&](const Coordinates &)
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-18 07:34:55 +0000