aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp')
-rw-r--r--src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp49
1 files changed, 29 insertions, 20 deletions
diff --git a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
index c688951d57..9ed9d7c5b0 100644
--- a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
+++ b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -29,11 +29,12 @@
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
#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"
-
#include "support/StringSupport.h"
namespace arm_compute
@@ -42,7 +43,8 @@ namespace
{
constexpr int max_input_tensor_dim = 3;
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
+Status
+validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
{
ARM_COMPUTE_UNUSED(epsilon);
@@ -52,14 +54,15 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, cons
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis > 2, "Actual axis greater than 2 is not supported");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis >= TensorShape::num_max_dimensions, "Actual normalization axis greater than max number of dimensions");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis >= TensorShape::num_max_dimensions,
+ "Actual normalization axis greater than max number of dimensions");
// Reduce shape on axis
TensorShape sum_shape = input->tensor_shape();
sum_shape.set(actual_axis, 1);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(sum->tensor_shape(), sum_shape);
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
@@ -74,18 +77,25 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, cons
CLL2NormalizeLayerKernel::CLL2NormalizeLayerKernel()
: _input(nullptr), _sum(nullptr), _output(nullptr), _actual_axis(0), _epsilon(1e-12)
{
+ _type = CLKernelType::ELEMENTWISE;
}
-void CLL2NormalizeLayerKernel::configure(const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
+void CLL2NormalizeLayerKernel::configure(
+ const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
{
configure(CLKernelLibrary::get().get_compile_context(), input, sum, output, axis, epsilon);
}
-void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
+void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *sum,
+ ICLTensor *output,
+ int axis,
+ float epsilon)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, sum, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), sum->info(), output->info(), axis, epsilon));
- auto padding_info = get_padding_info({ input, sum, output });
+ auto padding_info = get_padding_info({input, sum, output});
_input = input;
_sum = sum;
@@ -93,8 +103,9 @@ void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context
_actual_axis = wrap_around(axis, max_input_tensor_dim);
_epsilon = epsilon;
- const unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
- const int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
+ const unsigned int vec_size_x =
+ adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
+ const int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
// Set build options
CLBuildOptions build_opts;
@@ -105,7 +116,7 @@ void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context
// Create kernel
std::string kernel_name;
unsigned int idx = 0;
- switch(_actual_axis)
+ switch (_actual_axis)
{
case 0:
kernel_name = "l2_normalize_x";
@@ -125,7 +136,7 @@ void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Set epsilon argument
- if(input->info()->data_type() == DataType::F32)
+ if (input->info()->data_type() == DataType::F32)
{
_kernel.setArg<cl_float>(idx, _epsilon);
}
@@ -144,7 +155,8 @@ void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLL2NormalizeLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
+Status CLL2NormalizeLayerKernel::validate(
+ const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, sum, output, axis, epsilon));
return Status{};
@@ -157,7 +169,7 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue
Window window_sum(window);
- switch(_actual_axis)
+ switch (_actual_axis)
{
case 0:
{
@@ -171,8 +183,7 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue
add_2D_tensor_argument(idx, _sum, sum_slice);
add_2D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
+ } while (window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
}
break;
case 1:
@@ -187,8 +198,7 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue
add_2D_tensor_argument(idx, _sum, sum_slice);
add_2D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
+ } while (window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
}
break;
case 2:
@@ -203,8 +213,7 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue
add_3D_tensor_argument(idx, _sum, sum_slice);
add_3D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(sum_slice));
+ } while (window.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(sum_slice));
}
break;
default:
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-03 21:13:47 +0000