From 14ca2ed21a50939fb5fefc75de8265f316666d65 Mon Sep 17 00:00:00 2001
From: Hyeongseok Oh <hseok82.oh@samsung.com>
Date: Thu, 29 Aug 2024 16:13:54 +0900
Subject: [PATCH] [onert] Remove layout in ITensor

This commit removes layout access from ITensor and derived classes.
Tensor classes for external buffer (IOTensor, UserTensor, EdgeTensor) and unittest remains layout information.

ONE-DCO-1.0-Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
---
 runtime/onert/backend/acl_common/IACLTensor.cc              | 2 --
 runtime/onert/backend/acl_common/IACLTensor.h               | 1 -
 runtime/onert/backend/cpu/KernelGenerator.cc                | 6 ------
 runtime/onert/backend/cpu/ops/OperationUtils.cc             | 1 -
 runtime/onert/backend/cpu/ops/OperationUtils.h              | 3 ---
 runtime/onert/backend/ruy/KernelGenerator.cc                | 6 ------
 runtime/onert/backend/ruy/ops/OperationUtils.h              | 3 ---
 runtime/onert/backend/train/KernelGenerator.cc              | 5 -----
 runtime/onert/backend/train/ops/OperationUtils.cc           | 3 ---
 runtime/onert/backend/train/ops/PoolLayer.cc                | 4 ----
 runtime/onert/backend/train/optimizer/Optimizers.test.cc    | 4 ----
 runtime/onert/backend/trix/Convert.h                        | 2 +-
 runtime/onert/backend/xnnpack/KernelGenerator.cc            | 6 ------
 runtime/onert/backend/xnnpack/ops/ConvolutionLayer.cc       | 3 ---
 .../onert/backend/xnnpack/ops/DepthwiseConvolutionLayer.cc  | 3 ---
 runtime/onert/backend/xnnpack/ops/FullyConnectedLayer.cc    | 3 ---
 runtime/onert/core/include/backend/ITensor.h                | 1 -
 runtime/onert/core/include/backend/basic/Tensor.h           | 6 ++----
 .../core/include/backend/basic/train/TrainableTensor.h      | 1 -
 runtime/onert/core/src/backend/builtin/IOTensor.cc          | 1 -
 runtime/onert/core/src/backend/builtin/IOTensor.h           | 2 +-
 runtime/onert/core/src/backend/builtin/UserTensor.h         | 2 +-
 .../onert/core/src/backend/builtin/kernel/PermuteLayer.cc   | 4 ++--
 runtime/onert/core/src/exec/EdgeTensor.h                    | 2 +-
 runtime/onert/core/src/exec/IPermuteFunction.cc             | 3 ---
 runtime/onert/core/src/exec/IPermuteFunction.h              | 4 ++--
 runtime/onert/core/src/exec/IPermuteFunction.test.cc        | 2 +-
 runtime/onert/core/src/exec/feature/MockTensor.test.h       | 2 +-
 runtime/onert/core/src/exec/feature/nchw/Reader.h           | 2 --
 runtime/onert/core/src/exec/feature/nhwc/Reader.h           | 2 --
 30 files changed, 12 insertions(+), 77 deletions(-)

diff --git a/runtime/onert/backend/acl_common/IACLTensor.cc b/runtime/onert/backend/acl_common/IACLTensor.cc
index 9920750fc24..50fef459daf 100644
--- a/runtime/onert/backend/acl_common/IACLTensor.cc
+++ b/runtime/onert/backend/acl_common/IACLTensor.cc
@@ -41,8 +41,6 @@ size_t IACLTensor::calcOffset(const ir::Coordinates &coords) const
   return info()->offset_element_in_bytes(acl_coords);
 }
 
-ir::Layout IACLTensor::layout() const { return acl_common::asRuntimeLayout(info()->data_layout()); }
-
 ir::DataType IACLTensor::data_type() const
 {
   return acl_common::asRuntimeDataType(info()->data_type());
diff --git a/runtime/onert/backend/acl_common/IACLTensor.h b/runtime/onert/backend/acl_common/IACLTensor.h
index 7ea6327a7c6..e43a4cbdac8 100644
--- a/runtime/onert/backend/acl_common/IACLTensor.h
+++ b/runtime/onert/backend/acl_common/IACLTensor.h
@@ -49,7 +49,6 @@ class IACLTensor : public ITensor
   uint8_t *buffer() const final { return handle()->buffer(); }
   size_t total_size() const final { return info()->total_size(); }
   size_t calcOffset(const ir::Coordinates &coords) const final;
-  ir::Layout layout() const final;
   ir::DataType data_type() const final;
   float data_scale() const override;
   int32_t data_zero_point() const override;
diff --git a/runtime/onert/backend/cpu/KernelGenerator.cc b/runtime/onert/backend/cpu/KernelGenerator.cc
index 598ec422688..75c78d886cd 100644
--- a/runtime/onert/backend/cpu/KernelGenerator.cc
+++ b/runtime/onert/backend/cpu/KernelGenerator.cc
@@ -260,12 +260,6 @@ std::unique_ptr<exec::FunctionSequence> KernelGenerator::generate(ir::OperationI
 
   for (auto &&ind : (op.getInputs() | ir::Remove::UNDEFINED) + op.getOutputs())
   {
-    auto portable_tensor = _tensor_reg->getPortableTensor(ind);
-    if (portable_tensor)
-    {
-      assert(portable_tensor->layout() == ir::Layout::NHWC);
-    }
-
     auto tensor = _tensor_reg->getNativeTensor(ind);
     if (tensor)
     {
diff --git a/runtime/onert/backend/cpu/ops/OperationUtils.cc b/runtime/onert/backend/cpu/ops/OperationUtils.cc
index 686865af281..5e27f175e5f 100644
--- a/runtime/onert/backend/cpu/ops/OperationUtils.cc
+++ b/runtime/onert/backend/cpu/ops/OperationUtils.cc
@@ -286,7 +286,6 @@ std::vector<int32_t> getReducerAxes(const IPortableTensor *axes)
   std::vector<int32_t> ret;
 
   auto axes_vals = (axes->getShape().rank() == 0) ? 1 : axes->getShape().dim(0);
-  assert(axes->layout() == ir::Layout::NHWC);
   assert(static_cast<size_t>(axes_vals) == axes->getShape().num_elements());
   switch (axes->data_type())
   {
diff --git a/runtime/onert/backend/cpu/ops/OperationUtils.h b/runtime/onert/backend/cpu/ops/OperationUtils.h
index 39ef0cf6c54..544d07b8785 100644
--- a/runtime/onert/backend/cpu/ops/OperationUtils.h
+++ b/runtime/onert/backend/cpu/ops/OperationUtils.h
@@ -97,9 +97,6 @@ inline nnfw::cker::Shape getShape(const IPortableTensor *tensor)
     return nnfw::cker::Shape();
 
   const ir::Shape &shape = tensor->get_info().shape();
-
-  assert(tensor->layout() == ir::Layout::NHWC);
-
   auto rank = shape.rank();
   nnfw::cker::Shape ret(rank);
   auto data = ret.DimsData();
diff --git a/runtime/onert/backend/ruy/KernelGenerator.cc b/runtime/onert/backend/ruy/KernelGenerator.cc
index 8e2b12d4024..ef80452d5b4 100644
--- a/runtime/onert/backend/ruy/KernelGenerator.cc
+++ b/runtime/onert/backend/ruy/KernelGenerator.cc
@@ -57,12 +57,6 @@ std::unique_ptr<exec::FunctionSequence> KernelGenerator::generate(ir::OperationI
 
   for (const auto &ind : (op.getInputs() | ir::Remove::UNDEFINED) + op.getOutputs())
   {
-    auto portable_tensor = _tensor_reg->getPortableTensor(ind);
-    if (portable_tensor)
-    {
-      assert(portable_tensor->layout() == ir::Layout::NHWC);
-    }
-
     auto tensor = _tensor_reg->getNativeTensor(ind);
     if (tensor)
     {
diff --git a/runtime/onert/backend/ruy/ops/OperationUtils.h b/runtime/onert/backend/ruy/ops/OperationUtils.h
index 716400c1f14..5f00a240871 100644
--- a/runtime/onert/backend/ruy/ops/OperationUtils.h
+++ b/runtime/onert/backend/ruy/ops/OperationUtils.h
@@ -45,9 +45,6 @@ inline nnfw::ruy::Shape getTensorShape(const IPortableTensor *tensor)
     return nnfw::ruy::Shape();
 
   const ir::Shape &shape = tensor->get_info().shape();
-
-  assert(tensor->layout() == ir::Layout::NHWC);
-
   auto rank = shape.rank();
   nnfw::ruy::Shape ret(rank);
   auto data = ret.DimsData();
diff --git a/runtime/onert/backend/train/KernelGenerator.cc b/runtime/onert/backend/train/KernelGenerator.cc
index f3df39735e0..aaaa50f1b72 100644
--- a/runtime/onert/backend/train/KernelGenerator.cc
+++ b/runtime/onert/backend/train/KernelGenerator.cc
@@ -133,11 +133,6 @@ std::unique_ptr<exec::train::TrainableFnSequence> KernelGenerator::generate(ir::
 
   for (auto &&ind : (op.getInputs() | ir::Remove::UNDEFINED) + op.getOutputs())
   {
-    auto portable_tensor = _tensor_reg->getPortableTensor(ind);
-    if (portable_tensor)
-    {
-      assert(portable_tensor->layout() == ir::Layout::NHWC);
-    }
     auto tensor = _tensor_reg->getNonConstTensor(ind);
     if (tensor)
     {
diff --git a/runtime/onert/backend/train/ops/OperationUtils.cc b/runtime/onert/backend/train/ops/OperationUtils.cc
index 9736d5ba5d6..94a2f52491a 100644
--- a/runtime/onert/backend/train/ops/OperationUtils.cc
+++ b/runtime/onert/backend/train/ops/OperationUtils.cc
@@ -37,9 +37,6 @@ nnfw::cker::Shape getShape(const IPortableTensor *tensor)
   assert(!tensor->is_dynamic() && "Dynamic tensor is not supported yet");
 
   const ir::Shape &shape = tensor->get_info().shape();
-
-  assert(tensor->layout() == ir::Layout::NHWC);
-
   auto rank = shape.rank();
   nnfw::cker::Shape ret(rank);
   auto data = ret.DimsData();
diff --git a/runtime/onert/backend/train/ops/PoolLayer.cc b/runtime/onert/backend/train/ops/PoolLayer.cc
index f77d58e6517..e98a33050ba 100644
--- a/runtime/onert/backend/train/ops/PoolLayer.cc
+++ b/runtime/onert/backend/train/ops/PoolLayer.cc
@@ -77,8 +77,6 @@ class MaxPool2D final : public TrainingKernelRegistry
 public:
   void forward(const IPortableTensor *in, IPortableTensor *out)
   {
-    assert(in->layout() == ir::Layout::NHWC);
-
     auto out_shape = getShape(out);
     auto out_data = getBuffer<float>(out);
     auto arg_max_index = _arg_max_index.get();
@@ -90,8 +88,6 @@ class MaxPool2D final : public TrainingKernelRegistry
 
   void backward(const IPortableTensor *back_prop_out, IPortableTensor *back_prop_in)
   {
-    assert(back_prop_out->layout() == ir::Layout::NHWC);
-
     // activation backward
     try
     {
diff --git a/runtime/onert/backend/train/optimizer/Optimizers.test.cc b/runtime/onert/backend/train/optimizer/Optimizers.test.cc
index f24138c0481..2876fe430cc 100644
--- a/runtime/onert/backend/train/optimizer/Optimizers.test.cc
+++ b/runtime/onert/backend/train/optimizer/Optimizers.test.cc
@@ -54,8 +54,6 @@ class MockUpTensor : public IPortableTensor
 
   template <typename T> const std::vector<T> &data() const { return _data; }
 
-  ir::Layout layout() const override { return ir::Layout::NHWC; }
-
 private:
   using ITensor::setShape;
   using ITensor::set_dynamic;
@@ -89,8 +87,6 @@ class MockUpTrainableTensor : public backend::train::ITrainableTensor
     return const_cast<uint8_t *>(_data.data());
   }
 
-  ir::Layout layout() const override { return ir::Layout::NHWC; }
-
 public:
   std::vector<ITensor *> optVars() override
   {
diff --git a/runtime/onert/backend/trix/Convert.h b/runtime/onert/backend/trix/Convert.h
index 662ed44b6b7..9359f0a5084 100644
--- a/runtime/onert/backend/trix/Convert.h
+++ b/runtime/onert/backend/trix/Convert.h
@@ -61,7 +61,7 @@ void setDataInfo(const std::vector<T *> &tensors, tensors_data_info *info)
 
   for (uint32_t idx = 0; idx < info->num_info; ++idx)
   {
-    info->info[idx].layout = convertDataLayout(tensors[idx]->layout());
+    info->info[idx].layout = DATA_LAYOUT_NHWC;
     info->info[idx].type = convertDataType(tensors[idx]->data_type());
   }
 }
diff --git a/runtime/onert/backend/xnnpack/KernelGenerator.cc b/runtime/onert/backend/xnnpack/KernelGenerator.cc
index 01408561f87..abdfbe914c1 100644
--- a/runtime/onert/backend/xnnpack/KernelGenerator.cc
+++ b/runtime/onert/backend/xnnpack/KernelGenerator.cc
@@ -70,12 +70,6 @@ std::unique_ptr<exec::FunctionSequence> KernelGenerator::generate(ir::OperationI
 
   for (auto &&ind : (op.getInputs() | ir::Remove::UNDEFINED) + op.getOutputs())
   {
-    auto portable_tensor = _tensor_reg->getPortableTensor(ind);
-    if (portable_tensor)
-    {
-      assert(portable_tensor->layout() == ir::Layout::NHWC);
-    }
-
     auto tensor = _tensor_reg->getNativeTensor(ind);
     if (tensor)
     {
diff --git a/runtime/onert/backend/xnnpack/ops/ConvolutionLayer.cc b/runtime/onert/backend/xnnpack/ops/ConvolutionLayer.cc
index d08f7eac95d..e2e4e38253f 100644
--- a/runtime/onert/backend/xnnpack/ops/ConvolutionLayer.cc
+++ b/runtime/onert/backend/xnnpack/ops/ConvolutionLayer.cc
@@ -59,9 +59,6 @@ void ConvolutionLayer::configure(const IPortableTensor *input, const IPortableTe
   _activation = activation;
   _output = output;
 
-  // TODO Support not nhwc layer
-  assert(_input->layout() == ir::Layout::NHWC);
-
   assert(_activation == ir::Activation::NONE || _activation == ir::Activation::RELU ||
          _activation == ir::Activation::RELU1 || _activation == ir::Activation::RELU6);
 }
diff --git a/runtime/onert/backend/xnnpack/ops/DepthwiseConvolutionLayer.cc b/runtime/onert/backend/xnnpack/ops/DepthwiseConvolutionLayer.cc
index 87617e10c0a..1fc04b32b6d 100644
--- a/runtime/onert/backend/xnnpack/ops/DepthwiseConvolutionLayer.cc
+++ b/runtime/onert/backend/xnnpack/ops/DepthwiseConvolutionLayer.cc
@@ -60,9 +60,6 @@ void DepthwiseConvolutionLayer::configure(
   _activation = activation;
   _output = output;
 
-  // TODO Support not nhwc layer
-  assert(_input->layout() == ir::Layout::NHWC);
-
   assert(_activation == ir::Activation::NONE || _activation == ir::Activation::RELU ||
          _activation == ir::Activation::RELU1 || _activation == ir::Activation::RELU6);
 }
diff --git a/runtime/onert/backend/xnnpack/ops/FullyConnectedLayer.cc b/runtime/onert/backend/xnnpack/ops/FullyConnectedLayer.cc
index 93885c619af..71f02b11091 100644
--- a/runtime/onert/backend/xnnpack/ops/FullyConnectedLayer.cc
+++ b/runtime/onert/backend/xnnpack/ops/FullyConnectedLayer.cc
@@ -44,9 +44,6 @@ void FullyConnectedLayer::configure(const IPortableTensor *input, const IPortabl
   _activation = activation;
   _output = output;
 
-  // TODO Support not nhwc layer
-  assert(_input->layout() == ir::Layout::NHWC);
-
   assert(_activation == ir::Activation::NONE || _activation == ir::Activation::RELU ||
          _activation == ir::Activation::RELU1 || _activation == ir::Activation::RELU6);
 }
diff --git a/runtime/onert/core/include/backend/ITensor.h b/runtime/onert/core/include/backend/ITensor.h
index 56041626448..81113bf24fe 100644
--- a/runtime/onert/core/include/backend/ITensor.h
+++ b/runtime/onert/core/include/backend/ITensor.h
@@ -42,7 +42,6 @@ class ITensor
   virtual uint8_t *buffer() const = 0;
   virtual size_t total_size() const = 0;
   virtual size_t calcOffset(const ir::Coordinates &coords) const = 0;
-  virtual ir::Layout layout() const = 0;
   virtual ir::DataType data_type() const = 0;
   virtual float data_scale() const = 0;
   virtual int32_t data_zero_point() const = 0;
diff --git a/runtime/onert/core/include/backend/basic/Tensor.h b/runtime/onert/core/include/backend/basic/Tensor.h
index 92d98f82e13..5847dfdd82d 100644
--- a/runtime/onert/core/include/backend/basic/Tensor.h
+++ b/runtime/onert/core/include/backend/basic/Tensor.h
@@ -41,8 +41,8 @@ class Tensor : public IPortableTensor
 
 public:
   Tensor(const ir::OperandInfo &info, DynamicMemoryManager *dynamic_mem_mgr)
-    : IPortableTensor(info), _layout(ir::Layout::NHWC), _buffer(nullptr), _size(info.total_size()),
-      _num_references(0), _dynamic_mem_mgr(dynamic_mem_mgr), _allocator(nullptr)
+    : IPortableTensor(info), _buffer(nullptr), _size(info.total_size()), _num_references(0),
+      _dynamic_mem_mgr(dynamic_mem_mgr), _allocator(nullptr)
   {
     // DO NOTHING
   }
@@ -71,7 +71,6 @@ class Tensor : public IPortableTensor
 
 public:
   uint8_t *buffer() const override { return _buffer; }
-  ir::Layout layout() const override { return _layout; }
   void set_dynamic() override { _info.setDynamic(); }
   bool applyShape(const ir::Shape &new_shape) override;
 
@@ -126,7 +125,6 @@ class Tensor : public IPortableTensor
   void setShape(const ir::Shape &new_shape) override;
 
 protected:
-  const ir::Layout _layout;
   uint8_t *_buffer;
   size_t _size;
   int32_t _num_references;
diff --git a/runtime/onert/core/include/backend/basic/train/TrainableTensor.h b/runtime/onert/core/include/backend/basic/train/TrainableTensor.h
index 263b32962c8..6a5c41782a4 100644
--- a/runtime/onert/core/include/backend/basic/train/TrainableTensor.h
+++ b/runtime/onert/core/include/backend/basic/train/TrainableTensor.h
@@ -51,7 +51,6 @@ class TrainableTensor : public backend::train::ITrainableTensor
 
 public:
   uint8_t *buffer() const override { return _tensor.buffer(); }
-  ir::Layout layout() const override { return _tensor.layout(); }
 
 public:
   std::vector<ITensor *> optVars() override;
diff --git a/runtime/onert/core/src/backend/builtin/IOTensor.cc b/runtime/onert/core/src/backend/builtin/IOTensor.cc
index a467888a832..2097566fa81 100644
--- a/runtime/onert/core/src/backend/builtin/IOTensor.cc
+++ b/runtime/onert/core/src/backend/builtin/IOTensor.cc
@@ -40,7 +40,6 @@ void IOTensor::setTensor(IPortableTensor *tensor)
 {
   assert(tensor);
   assert(tensor != this);
-  assert(tensor->layout() == _orig->layout()); // Changing layout is not considered yet
   _tensor = tensor;
   if (_info.shape() != tensor->getShape())
   {
diff --git a/runtime/onert/core/src/backend/builtin/IOTensor.h b/runtime/onert/core/src/backend/builtin/IOTensor.h
index cb8307791dd..1467ca0d8ff 100644
--- a/runtime/onert/core/src/backend/builtin/IOTensor.h
+++ b/runtime/onert/core/src/backend/builtin/IOTensor.h
@@ -57,7 +57,7 @@ class IOTensor : public IPortableTensor
 
 public:
   uint8_t *buffer() const override { return _tensor->buffer(); }
-  ir::Layout layout() const override { return _orig->layout(); }
+  ir::Layout layout() const { return _orig->layout(); }
   void set_dynamic() override
   {
     _info.setDynamic();
diff --git a/runtime/onert/core/src/backend/builtin/UserTensor.h b/runtime/onert/core/src/backend/builtin/UserTensor.h
index b7f6ce091d1..9b6ffbd41a6 100644
--- a/runtime/onert/core/src/backend/builtin/UserTensor.h
+++ b/runtime/onert/core/src/backend/builtin/UserTensor.h
@@ -45,7 +45,7 @@ class UserTensor : public IPortableTensor
 
 public:
   uint8_t *buffer() const override { return _buffer; }
-  ir::Layout layout() const override { return _layout; }
+  ir::Layout layout() const { return _layout; }
   void set_dynamic() override { _info.setDynamic(); }
   void setShape(const ir::Shape &new_shape) override { _info.shape(new_shape); }
   bool applyShape(const ir::Shape &) override;
diff --git a/runtime/onert/core/src/backend/builtin/kernel/PermuteLayer.cc b/runtime/onert/core/src/backend/builtin/kernel/PermuteLayer.cc
index 7d06af10fbe..2f0ce5e4ecf 100644
--- a/runtime/onert/core/src/backend/builtin/kernel/PermuteLayer.cc
+++ b/runtime/onert/core/src/backend/builtin/kernel/PermuteLayer.cc
@@ -265,7 +265,7 @@ void PermuteLayer::run()
         // If dst is subtensor, we have to use clEnqueueMapBuffer instead of clEnqueueWirteBuffer
         else if (dst->needMemoryMap() && !dst->is_subtensor())
         {
-          if (!src->has_padding() && !dst->has_padding() && src->layout() == dst->layout())
+          if (!src->has_padding() && !dst->has_padding() && permute_type == ir::PermuteType::COPY)
           {
             // This is more effective than multi-threading
             src->access([&](backend::ITensor &) { dst->enqueueWriteBuffer(src->buffer(), false); });
@@ -281,7 +281,7 @@ void PermuteLayer::run()
           }
         }
         else if (src->needMemoryMap() && !src->is_subtensor() && !src->has_padding() &&
-                 !dst->has_padding() && src->layout() == dst->layout())
+                 !dst->has_padding() && permute_type == ir::PermuteType::COPY)
         {
           // This is more effective than multi-threading
           assert(!dst->needMemoryMap());
diff --git a/runtime/onert/core/src/exec/EdgeTensor.h b/runtime/onert/core/src/exec/EdgeTensor.h
index 8df79c3890d..5ea9c4ac47b 100644
--- a/runtime/onert/core/src/exec/EdgeTensor.h
+++ b/runtime/onert/core/src/exec/EdgeTensor.h
@@ -36,7 +36,7 @@ class EdgeTensor : public backend::IPortableTensor
   ~EdgeTensor() = default;
 
   uint8_t *buffer() const override { return _buffer.get(); }
-  ir::Layout layout() const override { return _layout; }
+  ir::Layout layout() const { return _layout; }
   void set_dynamic() override { _info.setDynamic(); }
   bool applyShape(const ir::Shape &new_shape) override;
   void setShape(const ir::Shape &new_shape) override { _info.shape(new_shape); }
diff --git a/runtime/onert/core/src/exec/IPermuteFunction.cc b/runtime/onert/core/src/exec/IPermuteFunction.cc
index 95198373c16..6775c17d1a6 100644
--- a/runtime/onert/core/src/exec/IPermuteFunction.cc
+++ b/runtime/onert/core/src/exec/IPermuteFunction.cc
@@ -36,9 +36,6 @@ using namespace onert;
 inline nnfw::cker::Shape getShape(const backend::ITensor *tensor)
 {
   const ir::Shape shape = tensor->getShape();
-
-  assert(tensor->layout() == ir::Layout::NHWC);
-
   auto rank = shape.rank();
   nnfw::cker::Shape ret(rank);
   auto data = ret.DimsData();
diff --git a/runtime/onert/core/src/exec/IPermuteFunction.h b/runtime/onert/core/src/exec/IPermuteFunction.h
index 517d0dc6bee..604e0c72d8b 100644
--- a/runtime/onert/core/src/exec/IPermuteFunction.h
+++ b/runtime/onert/core/src/exec/IPermuteFunction.h
@@ -93,7 +93,7 @@ class IPermuteFunction : public IFunction
       // Now there is no case where both src and dst have cl buffer.
       assert(!src->needMemoryMap());
 
-      if (!src->has_padding() && !dst->has_padding() && src->layout() == dst->layout())
+      if (!src->has_padding() && !dst->has_padding() && permute_type == ir::PermuteType::COPY)
       {
         src->access([&](backend::ITensor &) { dst->enqueueWriteBuffer(src->buffer(), false); });
       }
@@ -110,7 +110,7 @@ class IPermuteFunction : public IFunction
       }
     }
     else if (src->needMemoryMap() && !src->is_subtensor() && !src->has_padding() &&
-             !dst->has_padding() && src->layout() == dst->layout())
+             !dst->has_padding() && permute_type == ir::PermuteType::COPY)
     {
       assert(!dst->needMemoryMap());
       dst->access([&](backend::ITensor &) { src->enqueueReadBuffer(dst->buffer(), true); });
diff --git a/runtime/onert/core/src/exec/IPermuteFunction.test.cc b/runtime/onert/core/src/exec/IPermuteFunction.test.cc
index fd5f73f7f3e..586e2305708 100644
--- a/runtime/onert/core/src/exec/IPermuteFunction.test.cc
+++ b/runtime/onert/core/src/exec/IPermuteFunction.test.cc
@@ -71,7 +71,7 @@ class MockUpTensor : public ITensor
 
   uint8_t *buffer() const override { return _data; }
 
-  ir::Layout layout() const override { return _layout; }
+  ir::Layout layout() const { return _layout; }
   ir::DataType data_type() const override { return _type_info.type(); }
   float data_scale() const override { return _type_info.scale(); }
   int32_t data_zero_point() const override { return _type_info.zero_point(); }
diff --git a/runtime/onert/core/src/exec/feature/MockTensor.test.h b/runtime/onert/core/src/exec/feature/MockTensor.test.h
index 1d2d375e2ae..bdddad99aae 100644
--- a/runtime/onert/core/src/exec/feature/MockTensor.test.h
+++ b/runtime/onert/core/src/exec/feature/MockTensor.test.h
@@ -47,7 +47,7 @@ template <typename T> class MockTensor : public onert::backend::ITensor
 
 public: // DUMMY methods
   size_t total_size() const override { return 0; }
-  onert::ir::Layout layout() const override { return _layout; }
+  onert::ir::Layout layout() const { return _layout; }
   onert::ir::DataType data_type() const override { return onert::ir::DataType::UINT8; }
   float data_scale() const override { return 0; }
   int32_t data_zero_point() const override { return 0; }
diff --git a/runtime/onert/core/src/exec/feature/nchw/Reader.h b/runtime/onert/core/src/exec/feature/nchw/Reader.h
index e1a963cbdc6..0519b995718 100644
--- a/runtime/onert/core/src/exec/feature/nchw/Reader.h
+++ b/runtime/onert/core/src/exec/feature/nchw/Reader.h
@@ -52,8 +52,6 @@ template <typename T> class Reader : public feature::Reader<T>
   Reader(backend::ITensor *tensor)
     : _ptr{tensor->buffer() + tensor->calcOffset({0, 0, 0, 0})}, _len{tensor->total_size()}
   {
-    assert(tensor->layout() == ir::Layout::NCHW);
-
     const auto start_offset = tensor->calcOffset({0, 0, 0, 0});
     auto shape = tensor->getShape();
     _strides.W = shape.dim(3) == 1 ? 0 : tensor->calcOffset({0, 0, 0, 1}) - start_offset;
diff --git a/runtime/onert/core/src/exec/feature/nhwc/Reader.h b/runtime/onert/core/src/exec/feature/nhwc/Reader.h
index 3e3c431bfa5..eec0675553a 100644
--- a/runtime/onert/core/src/exec/feature/nhwc/Reader.h
+++ b/runtime/onert/core/src/exec/feature/nhwc/Reader.h
@@ -53,8 +53,6 @@ template <typename T> class Reader : public feature::Reader<T>
   Reader(const backend::ITensor *tensor)
     : _ptr{tensor->buffer() + tensor->calcOffset({0, 0, 0, 0})}, _len{tensor->total_size()}
   {
-    assert(tensor->layout() == ir::Layout::NHWC);
-
     const auto start_offset = tensor->calcOffset({0, 0, 0, 0});
     auto shape = tensor->getShape();
     _strides.C = shape.dim(3) == 1 ? 0 : tensor->calcOffset({0, 0, 0, 1}) - start_offset;