commit bef7fa27b0d231a8649952f60808132d109b6345
author Sang-Hoon Park <sang-hoon.park@arm.com> Wed Oct 21 15:58:54 2020 +0100
committer Sang-Hoon Park <sang-hoon.park@arm.com> Sat Nov 07 08:07:22 2020 +0000
tree 7543c66a473d90e28b4860986fad77afa5115043
parent b9531540dadce8331a703c32456f3c9defdfefa9

COMPMID-3639: (3RDPARTY_UPDATE) Move CL kernels to src

Change-Id: I10d27db788e5086adae1841e3e2441cd9b76ef84
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4310
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp

diff --git a/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp b/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp
index 7ad017f..7af4290 100644
--- a/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp
@@ -28,6 +28,15 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
+#include "src/core/CL/kernels/CLFillBorderKernel.h"
+#include "src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h"
+#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h"
+#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h"
+#include "src/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
+#include "src/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
+#include "src/core/CL/kernels/CLWinogradFilterTransformKernel.h"
+#include "src/core/CL/kernels/CLWinogradOutputTransformKernel.h"
+#include "support/MemorySupport.h"
 
 using namespace arm_compute;
 
@@ -90,11 +99,13 @@
 } // namespace
 
 CLWinogradConvolutionLayer::CLWinogradConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_group(memory_manager), _batched_mm(memory_manager), _input_transform(), _filter_transform(), _output_transform(), _input0(), _input1(), _batched_mm_output(), _original_weights(nullptr),
-      _is_prepared(false)
+    : _memory_group(memory_manager), _batched_mm(memory_manager), _input_transform(), _filter_transform(support::cpp14::make_unique<CLWinogradFilterTransformKernel>()),
+      _output_transform(support::cpp14::make_unique<CLWinogradOutputTransformKernel>()), _input0(), _input1(), _batched_mm_output(), _original_weights(nullptr), _is_prepared(false)
 {
 }
 
+CLWinogradConvolutionLayer::~CLWinogradConvolutionLayer() = default;
+
 void CLWinogradConvolutionLayer::configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info,
                                            bool enable_fast_math)
 {
@@ -139,7 +150,7 @@
     _input_transform.configure(compile_context, input, &_input0, winograd_info);
 
     // Configure filter transform
-    _filter_transform.configure(compile_context, weights, &_input1, winograd_info);
+    _filter_transform->configure(compile_context, weights, &_input1, winograd_info);
 
     // Configure batched matrix multiply
     _batched_mm.configure(compile_context, &_input0, &_input1, nullptr, &_batched_mm_output, 1.0f, 0.0f, GEMMInfo(false, false, true /* Reshape weights only for the first run*/, 0, false, false,
@@ -147,7 +158,7 @@
                                                                                                                   (input->info()->data_type() == DataType::F16)));
 
     // Configure output transform
-    _output_transform.configure(compile_context, &_batched_mm_output, biases, output, winograd_info, act_info);
+    _output_transform->configure(compile_context, &_batched_mm_output, biases, output, winograd_info, act_info);
 
     // Allocate temporary tensors
     _input0.allocator()->allocate();
@@ -218,7 +229,7 @@
     _batched_mm.run();
 
     // Run output transform
-    CLScheduler::get().enqueue(_output_transform);
+    CLScheduler::get().enqueue(*_output_transform);
 }
 
 void CLWinogradConvolutionLayer::prepare()
@@ -227,7 +238,7 @@
     {
         // Run filter transform and mark original weights as unused
         _input1.allocator()->allocate();
-        CLScheduler::get().enqueue(_filter_transform, false);
+        CLScheduler::get().enqueue(*_filter_transform, false);
         _original_weights->mark_as_unused();
 
         // Prepare GEMM and release reshaped weights if marked unused by CLGEMM