Port OpenCL Quantization to new API

Partially resolves: COMPMID-4193

Change-Id: Ie8367769c690442a0e30383c67851b50ab7c6742
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5231
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 36 insertions, 10 deletions

diff --git a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
index 365f95243f..fb698d5b88 100644
--- a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
+++ b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
@@ -25,11 +25,11 @@
 
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Types.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/CL/kernels/CLQuantizationLayerKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
 
 namespace arm_compute
@@ -45,7 +45,7 @@ CLGenerateProposalsLayer::CLGenerateProposalsLayer(std::shared_ptr<IMemoryManage
       _pad_kernel(std::make_unique<CLPadLayerKernel>()),
       _dequantize_anchors(std::make_unique<CLDequantizationLayerKernel>()),
       _dequantize_deltas(std::make_unique<CLDequantizationLayerKernel>()),
-      _quantize_all_proposals(std::make_unique<CLQuantizationLayerKernel>()),
+      _quantize_all_proposals(std::make_unique<CLQuantizationLayer>()),
       _cpp_nms(memory_manager),
       _is_nhwc(false),
       _is_qasymm8(false),
@@ -270,7 +270,7 @@ Status CLGenerateProposalsLayer::validate(const ITensorInfo *scores, const ITens
         ARM_COMPUTE_RETURN_ON_ERROR(CLBoundingBoxTransformKernel::validate(&all_anchors_f32_info, &proposals_4_roi_values_f32, &deltas_flattened_f32_info,
                                                                            BoundingBoxTransformInfo(info.im_width(), info.im_height(), 1.f)));
 
-        ARM_COMPUTE_RETURN_ON_ERROR(CLQuantizationLayerKernel::validate(&proposals_4_roi_values_f32, &proposals_4_roi_values_quantized));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLQuantizationLayer::validate(&proposals_4_roi_values_f32, &proposals_4_roi_values_quantized));
         proposals_4_roi_values_to_use = &proposals_4_roi_values_quantized;
     }
     else
@@ -372,7 +372,7 @@ void CLGenerateProposalsLayer::run()
 
     if(_is_qasymm8)
     {
-        CLScheduler::get().enqueue(*_quantize_all_proposals, false);
+        _quantize_all_proposals->run();
     }
 
     // Non maxima suppression
diff --git a/src/runtime/CL/functions/CLQuantizationLayer.cpp b/src/runtime/CL/functions/CLQuantizationLayer.cpp
index cb8cabef87..1f6ddb6014 100644
--- a/src/runtime/CL/functions/CLQuantizationLayer.cpp
+++ b/src/runtime/CL/functions/CLQuantizationLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,10 +23,26 @@
  */
 #include "arm_compute/runtime/CL/functions/CLQuantizationLayer.h"
 
-#include "src/core/CL/kernels/CLQuantizationLayerKernel.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClQuantization.h"
 
 namespace arm_compute
 {
+struct CLQuantizationLayer::Impl
+{
+    const ICLTensor                        *src{ nullptr };
+    ICLTensor                              *dst{ nullptr };
+    std::unique_ptr<opencl::ClQuantization> op{ nullptr };
+};
+
+CLQuantizationLayer::CLQuantizationLayer()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLQuantizationLayer::~CLQuantizationLayer() = default;
+
 void CLQuantizationLayer::configure(const ICLTensor *input, ICLTensor *output)
 {
     configure(CLKernelLibrary::get().get_compile_context(), input, output);
@@ -34,13 +50,23 @@ void CLQuantizationLayer::configure(const ICLTensor *input, ICLTensor *output)
 
 void CLQuantizationLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output)
 {
-    auto k = std::make_unique<CLQuantizationLayerKernel>();
-    k->configure(compile_context, input, output);
-    _kernel = std::move(k);
+    _impl->src = input;
+    _impl->dst = output;
+
+    _impl->op = std::make_unique<opencl::ClQuantization>();
+    _impl->op->configure(compile_context, input->info(), output->info());
 }
 
 Status CLQuantizationLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
-    return CLQuantizationLayerKernel::validate(input, output);
+    return opencl::ClQuantization::validate(input, output);
+}
+
+void CLQuantizationLayer::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