COMPMID-3386: Support memory injection in CLConcatenate functions/kernels
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I611adf4f506d406540e920b0bd6befb4b5108918
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3601
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
index 10aa615..0bae901 100644
--- a/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
+++ b/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
@@ -28,9 +28,9 @@
#include "arm_compute/core/CL/CLValidate.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/Cast.h"
#include "support/StringSupport.h"
@@ -73,34 +73,27 @@
} // namespace
CLDepthConcatenateLayerKernel::CLDepthConcatenateLayerKernel()
- : _input(nullptr), _output(nullptr), _depth_offset(0)
+ : _depth_offset(0)
{
}
-void CLDepthConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int depth_offset, ICLTensor *output)
-{
- configure(CLKernelLibrary::get().get_compile_context(), input, depth_offset, output);
-}
-
-void CLDepthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int depth_offset, ICLTensor *output)
+void CLDepthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, 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));
- _input = input;
- _output = output;
_depth_offset = depth_offset;
- const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size();
+ const unsigned int num_elems_processed_per_iteration = 16 / input->element_size();
// Add build options
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+ build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->data_type()));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
- if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
+ if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
{
- const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
- const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
+ const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
+ const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
@@ -112,13 +105,13 @@
_kernel = create_kernel(compile_context, "concatenate", build_opts.options());
// Configure kernel window
- auto win_config = validate_and_configure_window(input->info(), depth_offset, output->info());
+ auto win_config = validate_and_configure_window(input, depth_offset, output);
ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
ICLKernel::configure_internal(std::get<1>(win_config));
// Set output valid region
- output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+ output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
}
Status CLDepthConcatenateLayerKernel::validate(const arm_compute::ITensorInfo *input,
@@ -130,14 +123,18 @@
return Status{};
}
-void CLDepthConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLDepthConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+ const Window &window, cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+ const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC));
+ auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
Window slice = window.first_slice_window_3D();
- const int offset_to_first_elements_in_bytes = _depth_offset * _output->info()->strides_in_bytes()[2];
+ const int offset_to_first_elements_in_bytes = _depth_offset * dst->info()->strides_in_bytes()[2];
unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the input and output parameters
_kernel.setArg<cl_int>(idx, offset_to_first_elements_in_bytes);
@@ -145,8 +142,8 @@
do
{
unsigned int idx = 0;
- add_3D_tensor_argument(idx, _input, slice);
- add_3D_tensor_argument(idx, _output, slice);
+ add_3D_tensor_argument(idx, src, slice);
+ add_3D_tensor_argument(idx, dst, slice);
enqueue(queue, *this, slice, lws_hint());
}
while(window.slide_window_slice_3D(slice));