aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp')
-rw-r--r--src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp80
1 files changed, 51 insertions, 29 deletions
diff --git a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
index 2728958add..5b72354abe 100644
--- a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
+++ b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/Validate.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -44,16 +45,20 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, u
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::S32, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32, DataType::S64);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(op != ReductionOperation::ARG_IDX_MAX && op != ReductionOperation::ARG_IDX_MIN, "Only ARG_IDX_MAX and ARG_IDX_MIN are supported");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions, "Reduction axis greater than max number of dimensions");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(op != ReductionOperation::ARG_IDX_MAX && op != ReductionOperation::ARG_IDX_MIN,
+ "Only ARG_IDX_MAX and ARG_IDX_MIN are supported");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions,
+ "Reduction axis greater than max number of dimensions");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis");
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U32, DataType::S32, DataType::S64, DataType::U64);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U32, DataType::S32, DataType::S64,
+ DataType::U64);
}
return Status{};
@@ -66,22 +71,34 @@ CLArgMinMaxLayerKernel::CLArgMinMaxLayerKernel()
_type = CLKernelType::ELEMENTWISE;
}
-void CLArgMinMaxLayerKernel::configure(const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLArgMinMaxLayerKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, axis, op);
}
-void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- TensorShape output_shape{ input->info()->tensor_shape() };
+ TensorShape output_shape{input->info()->tensor_shape()};
output_shape.set(axis, 1);
- auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape).set_data_type(DataType::S32).reset_padding().set_is_resizable(true));
+ auto_init_if_empty(*output->info(), input->info()
+ ->clone()
+ ->set_tensor_shape(output_shape)
+ .set_data_type(DataType::S32)
+ .reset_padding()
+ .set_is_resizable(true));
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), axis, op));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -90,11 +107,14 @@ void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context,
// Set build options
const auto adjusted_vector_size = adjust_vec_size(16U, input->info()->dimension(0));
- const auto vector_size = (adjusted_vector_size == 3U && axis == 0U) ? 2U : adjusted_vector_size; // the opencl kernel only supports sizes 2, 4, 8 and 16.
+ const auto vector_size = (adjusted_vector_size == 3U && axis == 0U)
+ ? 2U
+ : adjusted_vector_size; // the opencl kernel only supports sizes 2, 4, 8 and 16.
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->info()->dimension(0) % vector_size));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" +
+ support::cpp11::to_string(input->info()->dimension(0) % vector_size));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vector_size));
build_opts.add_option_if(is_data_type_float(input->info()->data_type()), "-DFLOAT_DATA_TYPE");
build_opts.add_option_if_else(op == ReductionOperation::ARG_IDX_MAX, "-DARG_MAX", "-DARG_MIN");
@@ -104,7 +124,7 @@ void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context,
// Create kernel
std::string kernel_axis_name;
- switch(axis)
+ switch (axis)
{
case 0:
build_opts.add_option("-DWIDTH=" + support::cpp11::to_string(input->info()->dimension(0)));
@@ -135,7 +155,10 @@ void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context,
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLArgMinMaxLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op)
+Status CLArgMinMaxLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, axis, op));
return Status{};
@@ -146,7 +169,7 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
- switch(_reduction_axis)
+ switch (_reduction_axis)
{
case 0:
{
@@ -154,7 +177,8 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
Window out_window(window);
Window in_window(window);
out_window.set(Window::DimX, Window::Dimension(0, 0, 0));
- in_window.set(Window::DimX, Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
+ in_window.set(Window::DimX,
+ Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
in_window.set(Window::DimY, Window::Dimension(0, _input->info()->dimension(1), 1u));
// Get first input and output slices
@@ -166,15 +190,15 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
add_2D_tensor_argument(idx, _input, in_slice);
add_2D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(in_window.slide_window_slice_2D(in_slice) && out_window.slide_window_slice_2D(out_slice));
+ } while (in_window.slide_window_slice_2D(in_slice) && out_window.slide_window_slice_2D(out_slice));
}
break;
case 1:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimY, Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
+ Window window_in{window};
+ window_in.set(Window::DimY,
+ Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
Window in_slice = window_in.first_slice_window_2D();
Window out_slice = window.first_slice_window_2D();
@@ -184,15 +208,15 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
add_2D_tensor_argument(idx, _input, in_slice);
add_2D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
+ } while (window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
}
break;
case 2:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimZ, Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
+ Window window_in{window};
+ window_in.set(Window::DimZ,
+ Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
Window in_slice = window_in.first_slice_window_3D();
Window out_slice = window.first_slice_window_3D();
@@ -202,14 +226,13 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
add_3D_tensor_argument(idx, _input, in_slice);
add_3D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
+ } while (window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
}
break;
case 3:
{
// Get first input and output slices
- Window window_in{ window };
+ Window window_in{window};
window_in.set(3, Window::Dimension(0, 1, 1));
Window in_slice = window_in.first_slice_window_4D();
Window out_slice = window.first_slice_window_4D();
@@ -220,8 +243,7 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue)
add_4D_tensor_argument(idx, _input, in_slice);
add_4D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
+ } while (window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
}
break;
default:
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-20 01:25:07 +0000