Port OpenCL Dequantization to new API

Partially resolves: COMPMID-4193

Change-Id: I4e14149d5b0a7f9c0dd3bfce800eaddca1e4d885
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5238
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 40 insertions, 13 deletions

diff --git a/src/runtime/CL/functions/CLDequantizationLayer.cpp b/src/runtime/CL/functions/CLDequantizationLayer.cpp
index d358813724..e0381f90ae 100644
--- a/src/runtime/CL/functions/CLDequantizationLayer.cpp
+++ b/src/runtime/CL/functions/CLDequantizationLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,10 +23,27 @@
  */
 #include "arm_compute/runtime/CL/functions/CLDequantizationLayer.h"
 
-#include "src/core/CL/kernels/CLDequantizationLayerKernel.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClDequantization.h"
 
 namespace arm_compute
 {
+struct CLDequantizationLayer::Impl
+{
+    const ICLTensor                          *src{ nullptr };
+    ICLTensor                                *dst{ nullptr };
+    std::unique_ptr<opencl::ClDequantization> op{ nullptr };
+};
+
+CLDequantizationLayer::CLDequantizationLayer()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLDequantizationLayer::~CLDequantizationLayer() = default;
+
 void CLDequantizationLayer::configure(const ICLTensor *input, ICLTensor *output)
 {
     configure(CLKernelLibrary::get().get_compile_context(), input, output);
@@ -34,13 +51,23 @@ void CLDequantizationLayer::configure(const ICLTensor *input, ICLTensor *output)
 
 void CLDequantizationLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output)
 {
-    auto k = std::make_unique<CLDequantizationLayerKernel>();
-    k->configure(compile_context, input, output);
-    _kernel = std::move(k);
+    _impl->src = input;
+    _impl->dst = output;
+
+    _impl->op = std::make_unique<opencl::ClDequantization>();
+    _impl->op->configure(compile_context, input->info(), output->info());
 }
 
 Status CLDequantizationLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
-    return CLDequantizationLayerKernel::validate(input, output);
+    return opencl::ClDequantization::validate(input, output);
+}
+
+void CLDequantizationLayer::run()
+{
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+    _impl->op->run(pack);
 }
 } // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
index fb698d5b88..81e24dba08 100644
--- a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
+++ b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
@@ -25,9 +25,9 @@
 
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/functions/CLDequantizationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLQuantizationLayer.h"
 #include "src/core/CL/kernels/CLBoundingBoxTransformKernel.h"
-#include "src/core/CL/kernels/CLDequantizationLayerKernel.h"
 #include "src/core/CL/kernels/CLGenerateProposalsLayerKernel.h"
 #include "src/core/CL/kernels/CLPadLayerKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -43,8 +43,8 @@ CLGenerateProposalsLayer::CLGenerateProposalsLayer(std::shared_ptr<IMemoryManage
       _compute_anchors_kernel(std::make_unique<CLComputeAllAnchorsKernel>()),
       _bounding_box_kernel(std::make_unique<CLBoundingBoxTransformKernel>()),
       _pad_kernel(std::make_unique<CLPadLayerKernel>()),
-      _dequantize_anchors(std::make_unique<CLDequantizationLayerKernel>()),
-      _dequantize_deltas(std::make_unique<CLDequantizationLayerKernel>()),
+      _dequantize_anchors(std::make_unique<CLDequantizationLayer>()),
+      _dequantize_deltas(std::make_unique<CLDequantizationLayer>()),
       _quantize_all_proposals(std::make_unique<CLQuantizationLayer>()),
       _cpp_nms(memory_manager),
       _is_nhwc(false),
@@ -261,10 +261,10 @@ Status CLGenerateProposalsLayer::validate(const ITensorInfo *scores, const ITens
     if(is_qasymm8)
     {
         TensorInfo all_anchors_f32_info(anchors->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32));
-        ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayerKernel::validate(&all_anchors_info, &all_anchors_f32_info));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayer::validate(&all_anchors_info, &all_anchors_f32_info));
 
         TensorInfo deltas_flattened_f32_info(deltas->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32));
-        ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayerKernel::validate(&deltas_flattened_info, &deltas_flattened_f32_info));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayer::validate(&deltas_flattened_info, &deltas_flattened_f32_info));
 
         TensorInfo proposals_4_roi_values_f32(deltas->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32));
         ARM_COMPUTE_RETURN_ON_ERROR(CLBoundingBoxTransformKernel::validate(&all_anchors_f32_info, &proposals_4_roi_values_f32, &deltas_flattened_f32_info,
@@ -363,8 +363,8 @@ void CLGenerateProposalsLayer::run()
 
     if(_is_qasymm8)
     {
-        CLScheduler::get().enqueue(*_dequantize_anchors, false);
-        CLScheduler::get().enqueue(*_dequantize_deltas, false);
+        _dequantize_anchors->run();
+        _dequantize_deltas->run();
     }
 
     // Build the boxes