Port CpuTranspose to new API
Partially Resolves: COMPMID-4277 (2/2)
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Id8ee520081fe905cb796d4376864fa84ac384caa
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/303714
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: bsgcomp <bsgcomp@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5217
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
index ec782fc..0a5318a 100644
--- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -24,6 +24,7 @@
#include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h"
#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/ITensorPack.h"
#include "arm_compute/core/Size2D.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
@@ -39,9 +40,8 @@
#include "src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.h"
#include "src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h"
#include "src/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
-#include "src/core/NEON/kernels/NETransposeKernel.h"
+#include "src/core/cpu/kernels/CpuTransposeKernel.h"
-#include <algorithm>
#include <cmath>
namespace arm_compute
@@ -142,16 +142,39 @@
}
} // namespace
+struct NEFullyConnectedLayerReshapeWeights::Impl
+{
+ const ITensor *src{ nullptr };
+ ITensor *dst{ nullptr };
+ std::unique_ptr<cpu::kernels::CpuTransposeKernel> op{ nullptr };
+};
+
+NEFullyConnectedLayerReshapeWeights::NEFullyConnectedLayerReshapeWeights()
+ : _impl(std::make_unique<Impl>())
+{
+}
+
+NEFullyConnectedLayerReshapeWeights::~NEFullyConnectedLayerReshapeWeights() = default;
+
void NEFullyConnectedLayerReshapeWeights::configure(const ITensor *input, ITensor *output)
{
- auto k = std::make_unique<NETransposeKernel>();
- k->configure(input, output);
- _kernel = std::move(k);
+ _impl->op = std::make_unique<cpu::kernels::CpuTransposeKernel>();
+ _impl->op->configure(input->info(), output->info());
+ _impl->src = input;
+ _impl->dst = output;
}
Status NEFullyConnectedLayerReshapeWeights::validate(const ITensorInfo *input, const ITensorInfo *output)
{
- return NETransposeKernel::validate(input, output);
+ return cpu::kernels::CpuTransposeKernel::validate(input, output);
+}
+
+void NEFullyConnectedLayerReshapeWeights::run()
+{
+ ITensorPack pack{};
+ pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+ pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+ NEScheduler::get().schedule_op(_impl->op.get(), Window::DimY, _impl->op->window(), pack);
}
NEFullyConnectedLayer::~NEFullyConnectedLayer() = default;
diff --git a/src/runtime/NEON/functions/NEPermute.cpp b/src/runtime/NEON/functions/NEPermute.cpp
index 257c1a2..f707fad 100644
--- a/src/runtime/NEON/functions/NEPermute.cpp
+++ b/src/runtime/NEON/functions/NEPermute.cpp
@@ -40,10 +40,6 @@
{
}
-NEPermute::NEPermute(NEPermute &&) = default;
-
-NEPermute &NEPermute::operator=(NEPermute &&) = default;
-
NEPermute::~NEPermute() = default;
void NEPermute::configure(const ITensor *input, ITensor *output, const PermutationVector &perm)
diff --git a/src/runtime/NEON/functions/NETranspose.cpp b/src/runtime/NEON/functions/NETranspose.cpp
index aaa52e3..3b3023f 100644
--- a/src/runtime/NEON/functions/NETranspose.cpp
+++ b/src/runtime/NEON/functions/NETranspose.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -23,21 +23,48 @@
*/
#include "arm_compute/runtime/NEON/functions/NETranspose.h"
-#include "src/core/NEON/kernels/NETransposeKernel.h"
-
-#include <utility>
+#include "arm_compute/core/Validate.h"
+#include "src/runtime/cpu/operators/CpuTranspose.h"
namespace arm_compute
{
+struct NETranspose::Impl
+{
+ const ITensor *src{ nullptr };
+ ITensor *dst{ nullptr };
+ std::unique_ptr<cpu::CpuTranspose> op{ nullptr };
+};
+
+NETranspose::NETranspose()
+ : _impl(std::make_unique<Impl>())
+{
+}
+
+NETranspose::~NETranspose() = default;
+
void NETranspose::configure(const ITensor *input, ITensor *output)
{
- auto k = std::make_unique<NETransposeKernel>();
- k->configure(input, output);
- _kernel = std::move(k);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+ _impl->src = input;
+ _impl->dst = output;
+ _impl->op = std::make_unique<cpu::CpuTranspose>();
+ _impl->op->configure(input->info(), output->info());
}
Status NETranspose::validate(const ITensorInfo *input, const ITensorInfo *output)
{
- return NETransposeKernel::validate(input, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+ ARM_COMPUTE_RETURN_ON_ERROR(cpu::CpuTranspose::validate(input, output));
+ return Status{};
}
+
+void NETranspose::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