!49241 ResizeLinear1D 对接 ResizeD PASS，ResizeLinear1D 输入 size 支持 tuple，list

Merge pull request !49241 from haozhang/ResizeLinear1D

docs/api/api_python/ops/mindspore.ops.ResizeLinear1D.rst

@@ -11,19 +11,20 @@ mindspore.ops.ResizeLinear1D
 
     .. warning::
         实验特性，接口可能发生变化。
+        目前，昇腾平台仅支持输入 `size` 为Tuple或List的场景。
 
     参数：
         - **coordinate_transformation_mode** (str) - 指定进行坐标变换的方式，默认值是"align_corners"，还可选"half_pixel"和"asymmetric"。
 
     输入：
         - **x** (Tensor) - ResizeBilinear的输入，三维的Tensor，其shape为 :math:`(batch, channels, width)`。支持以下数据类型：float16、float32、double。
-        - **size** (Tensor) - 指定 `x` 宽的新尺寸，一维的Tensor，其shape为 :math:`(1)` ，数据类型为int64。
+        - **size** (Union[Tuple[int], List[int], Tensor[int]) - 指定 `x` 宽的新尺寸，仅含一个整数 :math:`(new\_width)` 的Tuple、List或1-D Tensor。
 
     输出：
         Tensor，调整大小后的Tensor。shape为 :math:`(batch, channels, new\_width)` 的三维Tensor，数据类型和输入是一致的。
 
     异常：
         - **TypeError** - `x` 的数据类型不支持。
-        - **TypeError** - `size` 不是int64的数据类型。
+        - **TypeError** - `size` 不是Tuple[int]、List[int]或Tensor[int]。
         - **TypeError** - `coordinate_transformation_mode` 不是string。
         - **TypeError** - `coordinate_transformation_mode` 不在支持的列表中。

mindspore/ccsrc/include/common/utils/utils.h

@@ -671,6 +671,8 @@ constexpr auto kResizeNearestNeighborV2OpName = "ResizeNearestNeighborV2";
 constexpr auto kResizeNearestNeighborV2DOpName = "ResizeNearestNeighborV2D";
 constexpr auto kReverseV2OpName = "ReverseV2";
 constexpr auto kReverseV2DOpName = "ReverseV2D";
+constexpr auto kResizeDOpName = "ResizeD";
+constexpr auto kResizeGradDOpName = "ResizeGradD";
 constexpr auto kReturnOpName = "Return";
 constexpr auto kRGBToHSVOpName = "RGBToHSV";
 constexpr auto kROIAlignGradName = "ROIAlignGrad";

mindspore/ccsrc/plugin/device/ascend/optimizer/ascend_backend_optimization.cc

@@ -42,6 +42,7 @@
 #include "plugin/device/ascend/optimizer/ir_fission/renorm_split.h"
 #include "plugin/device/ascend/optimizer/ir_fission/tensor_scatter_fission.h"
 #include "plugin/device/ascend/optimizer/ir_fission/ascend_clip_by_norm_fission.h"
+#include "plugin/device/ascend/optimizer/ir_fission/resize_linear1d_fission.h"
 #include "backend/common/pass/communication_op_fusion.h"
 #include "backend/common/pass/dropout_gen_mask_fusion.h"
 #include "backend/common/pass/dynamic_sequence_ops_adaptation.h"
@@ -389,6 +390,8 @@ void AscendBackendIRFusionOptimization(const std::shared_ptr<session::KernelGrap
   ir_fusion_pm->AddPass(std::make_shared<RenormSplit>());
   ir_fusion_pm->AddPass(std::make_shared<EraseVisitAttr>());
   ir_fusion_pm->AddPass(std::make_shared<RemoveHostKernel>());
+  ir_fusion_pm->AddPass(std::make_shared<ResizeLinear1DFission>());
+  ir_fusion_pm->AddPass(std::make_shared<ResizeLinear1DGradFission>());
   AddAscendIRFusionRulesPass(ir_fusion_pm.get());
   AddAscendIRFusionPass(ir_fusion_pm.get());
 
@@ -485,6 +488,8 @@ void RunOpAscendBackendIRFusionOptimization(const std::shared_ptr<session::Kerne
   ir_fusion_pm->AddPass(std::make_shared<EraseVisitAttr>());
   ir_fusion_pm->AddPass(std::make_shared<RemoveHostKernel>());
   ir_fusion_pm->AddPass(std::make_shared<PackFission>());
+  ir_fusion_pm->AddPass(std::make_shared<ResizeLinear1DFission>());
+  ir_fusion_pm->AddPass(std::make_shared<ResizeLinear1DGradFission>());
   const auto &pass_creators =
     opt::Factory<PatternProcessPass>::Instance().GetPassCreatorsByType(kPassType::kIRFusionFissionPass);
   for (const auto &pass_creator : pass_creators) {

mindspore/ccsrc/plugin/device/ascend/optimizer/ir_fission/resize_linear1d_fission.cc

@@ -0,0 +1,196 @@
+/**
+ * Copyright 2023 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "plugin/device/ascend/optimizer/ir_fission/resize_linear1d_fission.h"
+#include <memory>
+#include <vector>
+#include "include/common/utils/anfalgo.h"
+#include "plugin/device/ascend/optimizer/ascend_helper.h"
+#include "backend/common/session/anf_runtime_algorithm.h"
+#include "backend/common/optimizer/helper.h"
+#include "include/common/utils/utils.h"
+
+namespace mindspore {
+namespace opt {
+namespace {
+constexpr size_t kResizeLinear1DInputNum = 2;
+constexpr int64_t kExpandDim = -1;
+constexpr int64_t kSqueezeDim = 2;
+
+AnfNodePtr AddExpandDimsNode(const FuncGraphPtr &func_graph, const AnfNodePtr &input_node,
+                             const PatternProcessPass &pass) {
+  MS_EXCEPTION_IF_NULL(func_graph);
+  MS_EXCEPTION_IF_NULL(input_node);
+
+  // Add ExpandDims Node
+  std::vector<AnfNodePtr> expand_dims_inputs = {
+    NewValueNode(std::make_shared<Primitive>(prim::kPrimExpandDims->name())), input_node};
+  auto expand_dims = pass.NewCNode(expand_dims_inputs, func_graph);
+
+  // Set ExpandDims OutShape and Type
+  auto dtype = common::AnfAlgo::GetOutputInferDataType(input_node, 0);
+  auto expand_shape = common::AnfAlgo::GetOutputInferShape(input_node, 0);
+  (void)expand_shape.insert(expand_shape.end() + kExpandDim, 1);
+  (void)common::AnfAlgo::SetOutputInferTypeAndShape({dtype}, {expand_shape}, expand_dims.get());
+
+  // Set ExpandDims Attr
+  common::AnfAlgo::SetNodeAttr(kAttrAxis, MakeValue(kExpandDim), expand_dims);
+  common::AnfAlgo::SetNodeAttr("is_backend_insert", MakeValue(true), expand_dims);
+
+  return expand_dims;
+}
+}  // namespace
+
+const BaseRef ResizeLinear1DFission::DefinePattern() const {
+  VarPtr Xs = std::make_shared<SeqVar>();
+  auto resize_linear1d_prim = std::make_shared<Primitive>(prim::kPrimResizeLinear1D->name());
+  return VectorRef({resize_linear1d_prim, Xs});
+}
+
+const AnfNodePtr ResizeLinear1DFission::Process(const FuncGraphPtr &func_graph, const AnfNodePtr &node,
+                                                const EquivPtr &) const {
+  MS_EXCEPTION_IF_NULL(func_graph);
+  MS_EXCEPTION_IF_NULL(node);
+  auto resize_linear1d = node->cast<CNodePtr>();
+  MS_EXCEPTION_IF_NULL(resize_linear1d);
+  auto kernel_graph = func_graph->cast<KernelGraphPtr>();
+  MS_EXCEPTION_IF_NULL(kernel_graph);
+  if (common::AnfAlgo::IsDynamicShape(node)) {
+    return nullptr;
+  }
+
+  if (resize_linear1d->size() != kResizeLinear1DInputNum + 1) {
+    MS_LOG(INFO) << "The node " << resize_linear1d->DebugString() << " is not equal to " << kResizeLinear1DInputNum
+                 << "inputs";
+    return nullptr;
+  }
+
+  if (!common::AnfAlgo::HasNodeAttr("coordinate_transformation_mode", resize_linear1d)) {
+    MS_LOG(EXCEPTION) << "ResizeLinear1D need to set coordinate_transformation_mode attribute.";
+  }
+
+  const auto ori_inputs = resize_linear1d->inputs();
+
+  // Add ExpandDims Node
+  auto expand_dims = AddExpandDimsNode(func_graph, ori_inputs[kDim1], *this);
+
+  // Get ResizeD Node
+  std::vector<AnfNodePtr> resize_d_inputs = {NewValueNode(std::make_shared<Primitive>(kResizeDOpName)), expand_dims};
+  auto resize_d = func_graph->NewCNode(resize_d_inputs);
+  MS_EXCEPTION_IF_NULL(resize_d);
+
+  // Set ResizeD OutShape and Type
+  auto out_type = common::AnfAlgo::GetOutputInferDataType(node, 0);
+  auto out_shape = common::AnfAlgo::GetOutputInferShape(node, 0);
+  (void)out_shape.insert(out_shape.end() + kExpandDim, 1);
+  (void)common::AnfAlgo::SetOutputInferTypeAndShape({out_type}, {out_shape}, resize_d.get());
+
+  // Set ResizeD Attr
+  std::vector<int64_t> size_value = {out_shape[kIndex3]};
+  auto x_shape = common::AnfAlgo::GetOutputInferShape(ori_inputs[kDim1], 0);
+  float scale = static_cast<float>(size_value[kDim0]) / static_cast<float>(x_shape[kDim2]);
+  std::vector<float> scales = {scale};
+  common::AnfAlgo::SetNodeAttr("sizes", MakeValue(size_value), resize_d);
+  common::AnfAlgo::SetNodeAttr("scales", MakeValue(scales), resize_d);
+  common::AnfAlgo::SetNodeAttr("mode", MakeValue("linear"), resize_d);
+  common::AnfAlgo::CopyNodeAttr("coordinate_transformation_mode", resize_linear1d, resize_d);
+
+  // Get Squeeze Node
+  std::vector<AnfNodePtr> squeeze_inputs = {NewValueNode(std::make_shared<Primitive>(prim::kPrimSqueeze->name())),
+                                            resize_d};
+  auto squeeze = func_graph->NewCNode(squeeze_inputs);
+  MS_EXCEPTION_IF_NULL(squeeze);
+
+  // Set Squeeze Attr
+  std::vector<int64_t> axis = {kSqueezeDim};
+  common::AnfAlgo::SetNodeAttr("axis", MakeValue(axis), squeeze);
+
+  // Set abstract and scope
+  squeeze->set_abstract(resize_linear1d->abstract());
+  squeeze->set_scope(resize_linear1d->scope());
+
+  return squeeze;
+}
+
+const BaseRef ResizeLinear1DGradFission::DefinePattern() const {
+  VarPtr Xs = std::make_shared<SeqVar>();
+  auto resize_linear1d_grad_prim = std::make_shared<Primitive>(prim::kPrimResizeLinear1DGrad->name());
+  return VectorRef({resize_linear1d_grad_prim, Xs});
+}
+
+const AnfNodePtr ResizeLinear1DGradFission::Process(const FuncGraphPtr &func_graph, const AnfNodePtr &node,
+                                                    const EquivPtr &) const {
+  MS_EXCEPTION_IF_NULL(func_graph);
+  MS_EXCEPTION_IF_NULL(node);
+  auto resize_linear1d_grad = node->cast<CNodePtr>();
+  MS_EXCEPTION_IF_NULL(resize_linear1d_grad);
+  auto kernel_graph = func_graph->cast<KernelGraphPtr>();
+  MS_EXCEPTION_IF_NULL(kernel_graph);
+
+  if (resize_linear1d_grad->size() != kResizeLinear1DInputNum + 1) {
+    MS_LOG(INFO) << "The node " << resize_linear1d_grad->DebugString() << " is not equal to " << kResizeLinear1DInputNum
+                 << "inputs";
+    return nullptr;
+  }
+
+  if (!common::AnfAlgo::HasNodeAttr("coordinate_transformation_mode", resize_linear1d_grad)) {
+    MS_LOG(EXCEPTION) << "ResizeLinear1DGrad need to set coordinate_transformation_mode attribute.";
+  }
+
+  const auto ori_inputs = resize_linear1d_grad->inputs();
+
+  // Add ExpandDims Node
+  auto expand_dims = AddExpandDimsNode(func_graph, ori_inputs[kDim1], *this);
+
+  // Get ResizeGradD Node
+  std::vector<AnfNodePtr> resize_grad_d_inputs = {NewValueNode(std::make_shared<Primitive>(kResizeGradDOpName)),
+                                                  expand_dims};
+  auto resize_grad_d = func_graph->NewCNode(resize_grad_d_inputs);
+  MS_EXCEPTION_IF_NULL(resize_grad_d);
+
+  // Set ResizeGradD OutShape and Type
+  auto out_type = common::AnfAlgo::GetOutputInferDataType(node, 0);
+  auto out_shape = common::AnfAlgo::GetOutputInferShape(node, 0);
+  (void)out_shape.insert(out_shape.end() + kExpandDim, 1);
+  (void)common::AnfAlgo::SetOutputInferTypeAndShape({out_type}, {out_shape}, resize_grad_d.get());
+
+  // Set ResizeGradD Attr
+  auto origin_shape = common::AnfAlgo::GetPrevNodeOutputInferShape(node, kIndex1);
+  auto x_shape = common::AnfAlgo::GetOutputInferShape(ori_inputs[kDim1], 0);
+  float scale = static_cast<float>(x_shape[kDim2]) / static_cast<float>(origin_shape[kDim2]);
+  std::vector<float> scales = {scale};
+  common::AnfAlgo::SetNodeAttr("original_size", MakeValue(origin_shape), resize_grad_d);
+  common::AnfAlgo::SetNodeAttr("scales", MakeValue(scales), resize_grad_d);
+  common::AnfAlgo::SetNodeAttr("mode", MakeValue("linear"), resize_grad_d);
+  common::AnfAlgo::CopyNodeAttr("coordinate_transformation_mode", resize_linear1d_grad, resize_grad_d);
+
+  // Get Squeeze Node
+  std::vector<AnfNodePtr> squeeze_inputs = {NewValueNode(std::make_shared<Primitive>(prim::kPrimSqueeze->name())),
+                                            resize_grad_d};
+  auto squeeze = func_graph->NewCNode(squeeze_inputs);
+  MS_EXCEPTION_IF_NULL(squeeze);
+
+  // Set Squeeze Attr
+  std::vector<int64_t> axis = {kSqueezeDim};
+  common::AnfAlgo::SetNodeAttr("axis", MakeValue(axis), squeeze);
+
+  // Set abstract and scope
+  squeeze->set_abstract(resize_linear1d_grad->abstract());
+  squeeze->set_scope(resize_linear1d_grad->scope());
+
+  return squeeze;
+}
+}  // namespace opt
+}  // namespace mindspore

mindspore/ccsrc/plugin/device/ascend/optimizer/ir_fission/resize_linear1d_fission.h

@@ -0,0 +1,46 @@
+/**
+ * Copyright 2023 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_IR_FUSION_RESIZE_LINEAR1D_FISSION_H_
+#define MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_IR_FUSION_RESIZE_LINEAR1D_FISSION_H_
+
+#include "ir/anf.h"
+#include "backend/common/optimizer/pattern_engine.h"
+#include "backend/common/optimizer/helper.h"
+#include "backend/common/optimizer/optimizer.h"
+
+namespace mindspore {
+namespace opt {
+class ResizeLinear1DFission : public PatternProcessPass {
+ public:
+  explicit ResizeLinear1DFission(bool multigraph = true) : PatternProcessPass("resize_linear1d_fission", multigraph) {}
+  ~ResizeLinear1DFission() override = default;
+  const BaseRef DefinePattern() const override;
+  const AnfNodePtr Process(const FuncGraphPtr &, const AnfNodePtr &, const EquivPtr &) const override;
+};
+
+class ResizeLinear1DGradFission : public PatternProcessPass {
+ public:
+  explicit ResizeLinear1DGradFission(bool multigraph = true)
+      : PatternProcessPass("resize_linear1d_grad_fission", multigraph) {}
+  ~ResizeLinear1DGradFission() override = default;
+  const BaseRef DefinePattern() const override;
+  const AnfNodePtr Process(const FuncGraphPtr &, const AnfNodePtr &, const EquivPtr &) const override;
+};
+}  // namespace opt
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_IR_FUSION_RESIZE_LINEAR1D_FISSION_H_

mindspore/core/ops/resize_linear_1d.cc

@@ -20,6 +20,7 @@
 #include <algorithm>
 
 #include "abstract/ops/primitive_infer_map.h"
+#include "ops/op_utils.h"
 #include "utils/check_convert_utils.h"
 #include "abstract/abstract_value.h"
 #include "abstract/dshape.h"
@@ -50,89 +51,68 @@ const int64_t kInputShape0Dim = 3;
 const int64_t kInputShape1Dim = 1;
 abstract::ShapePtr ResizeLinear1DInferShape(const PrimitivePtr &primitive,
                                             const std::vector<AbstractBasePtr> &input_args) {
-  auto input_x_arg = input_args[kInputIndex0];
+  MS_EXCEPTION_IF_NULL(primitive);
-  auto size_arg = input_args[kInputIndex1];
+  auto prim_name = primitive->name();
-  if (!input_x_arg->isa<abstract::AbstractTensor>()) {
-    MS_EXCEPTION(TypeError) << "Images only support tensor!";
-  }
-  if (!size_arg->isa<abstract::AbstractTensor>()) {
-    MS_EXCEPTION(TypeError) << "Size only support tensor!";
-  }
-  auto input_x_shape = input_x_arg->cast<abstract::AbstractTensorPtr>();
-  MS_EXCEPTION_IF_NULL(input_x_shape);
-  auto input_x_shape_value_ptr = input_x_shape->BuildValue();
-  MS_EXCEPTION_IF_NULL(input_x_shape_value_ptr);
-  auto input_x_type = input_x_arg->BuildType();
-  MS_EXCEPTION_IF_NULL(input_x_type);
-  auto input_x_type_id = input_x_type->cast<TensorTypePtr>();
-  MS_EXCEPTION_IF_NULL(input_x_type_id);
-  auto input_x_type_element = input_x_type_id->element();
-  MS_EXCEPTION_IF_NULL(input_x_type_element);
-  auto size_shape = size_arg->cast<abstract::AbstractTensorPtr>();
-  MS_EXCEPTION_IF_NULL(size_shape);
-  auto size_shape_value_ptr = size_shape->BuildValue();
-  MS_EXCEPTION_IF_NULL(size_shape_value_ptr);
-  auto size_shape_tensor = size_shape_value_ptr->cast<tensor::TensorPtr>();
-  auto size_type = size_arg->BuildType();
-  MS_EXCEPTION_IF_NULL(size_type);
-  auto size_type_id = size_type->cast<TensorTypePtr>();
-  MS_EXCEPTION_IF_NULL(size_type_id);
-  auto size_type_element = size_type_id->element();
-  MS_EXCEPTION_IF_NULL(size_type_element);
-  auto shape0_ptr = std::make_shared<abstract::Shape>(
-    CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_x_arg->BuildShape())[kShape]);
-  auto shape1_ptr =
-    std::make_shared<abstract::Shape>(CheckAndConvertUtils::ConvertShapePtrToShapeMap(size_arg->BuildShape())[kShape]);
-  auto shape0_v = shape0_ptr->shape();
-  auto shape1_v = shape1_ptr->shape();
-  // support dynamic shape
-  if (IsDynamicRank(shape0_v) || IsDynamicRank(shape1_v)) {
-    return std::make_shared<abstract::Shape>(ShapeVector({abstract::Shape::kShapeRankAny}));
-  }
-  if (shape0_v.size() < kInputShape0Dim) {
-    MS_EXCEPTION(ValueError) << "For 'ResizeLinear1D', the rank of images tensor must be greater than 3. But got "
-                             << shape0_v.size();
-  }
-  if (shape1_v.size() != kInputShape1Dim) {
-    MS_EXCEPTION(ValueError) << "For 'ResizeLinear1D', the size tensor must be a 1-D tensor. But got "
-                             << shape1_v.size() << "-D";
-  }
-  if (size_arg->isa<abstract::AbstractTensor>() && size_arg->BuildValue()->isa<tensor::Tensor>()) {
-    int64_t out_width = 0;
-    if (size_shape_tensor->data_type() == kNumberTypeInt32) {
-      auto size_shape_ptr = reinterpret_cast<int32_t *>(size_shape_tensor->data_c());
-      out_width = static_cast<int64_t>(size_shape_ptr[kInputIndex0]);
-    } else if (size_shape_tensor->data_type() == kNumberTypeInt64) {
-      auto size_shape_ptr = reinterpret_cast<int64_t *>(size_shape_tensor->data_c());
-      out_width = size_shape_ptr[kInputIndex0];
-    }
-    if (out_width <= 0) {
-      MS_EXCEPTION(ValueError) << "The size must be positive , but got " << out_width;
-    }
 
-    std::vector<int64_t> output_shape = shape0_v;
+  const int64_t shape0_dim = 3;
-    output_shape.pop_back();
+  std::vector<int64_t> output_shape(shape0_dim, abstract::Shape::kShapeDimAny);
-    output_shape.push_back(out_width);
+
-    return std::make_shared<abstract::Shape>(output_shape);
+  auto shape0 = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[0]->BuildShape())[kShape];
-  } else {
+  if (!IsDynamicRank(shape0)) {
-    ShapeVector shape_out = shape0_v;
+    (void)CheckAndConvertUtils::CheckInteger("images' rank", SizeToLong(shape0.size()), kEqual, shape0_dim, prim_name);
-    shape_out.pop_back();
+    output_shape[kInputIndex0] = shape0[kInputIndex0];
-    shape_out.push_back(abstract::Shape::kShapeDimAny);
+    output_shape[kInputIndex1] = shape0[kInputIndex1];
-    return std::make_shared<abstract::Shape>(shape_out);
   }
+
+  auto value_ptr = input_args[kInputIndex1]->BuildValue();
+  MS_EXCEPTION_IF_NULL(value_ptr);
+
+  if (!IsValueKnown(value_ptr)) {
+    return std::make_shared<abstract::Shape>(output_shape);
+  }
+
+  auto size_type = input_args[kInputIndex1]->BuildType();
+  std::vector<int64_t> size_value{};
+  if (size_type->isa<TensorType>()) {
+    const int64_t kDimOne = 1;
+    auto size_shape = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[kInputIndex1]->BuildShape())[kShape];
+    (void)CheckAndConvertUtils::CheckInteger("rank of size's shape", SizeToLong(size_shape.size()), kEqual, kDimOne,
+                                             prim_name);
+    size_value = CheckAndConvertUtils::CheckTensorIntValue("size", value_ptr, prim_name);
+  } else if (IsIdentidityOrSubclass(size_type, kTuple) || IsIdentidityOrSubclass(size_type, kList)) {
+    size_value = CheckAndConvertUtils::CheckIntOrTupleInt("size", value_ptr, prim_name);
+  } else {
+    MS_EXCEPTION(TypeError) << "For primitive[" << prim_name << "], the `size` "
+                            << " must be a tuple、list or tensor with all Int elements, but got "
+                            << value_ptr->type_name() << ".";
+  }
+
+  const int64_t size_num = 1;
+  (void)CheckAndConvertUtils::CheckInteger("size", SizeToLong(size_value.size()), kEqual, size_num, prim_name);
+  const int64_t kNumZero = 0;
+  for (size_t i = 0; i < size_value.size(); ++i) {
+    CheckAndConvertUtils::CheckInteger("size", size_value[i], kGreaterThan, kNumZero, prim_name);
+  }
+
+  output_shape[kInputIndex2] = size_value[kInputIndex0];
+
+  return std::make_shared<abstract::Shape>(output_shape);
 }
+
 TypePtr ResizeLinear1DInferType(const PrimitivePtr &primitive, const std::vector<AbstractBasePtr> &input_args) {
   if (std::any_of(input_args.begin(), input_args.end(), [](const AbstractBasePtr arg) { return arg == nullptr; })) {
     MS_LOG(EXCEPTION) << "For 'ResizeLinear1D', input args contain nullptr.";
   }
   auto prim_name = primitive->name();
-  auto input_x_arg = input_args[kInputIndex0];
+  auto x_type = input_args[kInputIndex0]->BuildType();
-  auto size_arg = input_args[kInputIndex1];
+  auto size_type = input_args[kInputIndex1]->BuildType();
   const std::set<TypePtr> valid0_types = {kFloat16, kFloat32, kFloat64};
-  const std::set<TypePtr> valid1_types = {kInt64, kInt32};
+  (void)CheckAndConvertUtils::CheckTensorTypeValid("images", x_type, valid0_types, prim_name);
-  (void)CheckAndConvertUtils::CheckTensorTypeValid("images", input_x_arg->BuildType(), valid0_types, prim_name);
+  if (size_type->isa<TensorType>()) {
-  (void)CheckAndConvertUtils::CheckTensorTypeValid("size", size_arg->BuildType(), valid1_types, prim_name);
+    const std::set<TypePtr> valid1_types = {kInt32, kInt64};
-  return input_x_arg->BuildType();
+    (void)CheckAndConvertUtils::CheckTensorTypeValid("size", size_type, valid1_types, prim_name);
+  }
+  return x_type;
 }
 }  // namespace
 

mindspore/python/mindspore/nn/layer/basic.py

@@ -487,7 +487,7 @@ class Upsample(Cell):
             upsample = P.image_ops.ResizeLinear1D(
                 coordinate_transformation_mode=coordinate_transformation_mode
             )
-            return upsample(tensor, Tensor(size, dtype=mstype.int32))
+            return upsample(tensor, size)
 
         def run_bilinear(tensor, ndim, size):
             if self.scale_factor is not None:

mindspore/python/mindspore/ops/function/nn_func.py

@@ -2102,7 +2102,7 @@ def interpolate(x, size=None, scale_factor=None, mode="nearest", align_corners=N
         resize = _get_cache_prim(P.image_ops.ResizeLinear1D)(
             coordinate_transformation_mode
         )
-        return resize(x, Tensor(size, dtype=mstype.int32))
+        return resize(x, size)
 
     def run_bilinear(x, size, align_corners=None, scale_factor=None):
         resize = _get_cache_prim(P.ResizeBilinearV2)(align_corners, not align_corners)

mindspore/python/mindspore/ops/operations/image_ops.py

@@ -641,6 +641,7 @@ class ResizeLinear1D(Primitive):
 
     .. warning::
         This is an experimental feature and is subjected to change.
+        Currently, the Ascend platform only supports scenarios where the input `size` is Tuple or List.
 
     Args:
         coordinate_transformation_mode (str): Default is 'align_corners'. Describes how to transform the coordinate
@@ -649,19 +650,20 @@ class ResizeLinear1D(Primitive):
     Inputs:
         - **x** (Tensor) - A 3-D tensor which to resize, with shape [batch, channel, width]. Must be one of the
           following types: uint8, int8, int16, int32, int64, float16, float32, double.
-        - **size** (Tensor) - A 1-D int64 Tensor, describes the size of the output tensor.
+        - **size** (Union[Tuple[int], List[int], Tensor[int]]): describes the new width of `x` .
+          A tuple or list or 1-D tensor with only one int element :math:`(new\_width)`.
 
     Outputs:
         A 3-D tensor which shape is [batch, channel, new_width] with the same type as `x`.
 
     Raises:
         TypeError: If dtype of `x` is not in the support list.
-        TypeError: If `size` is not a 1-D int64_t tensor.
+        TypeError: If `size` is not in Union[Tuple[int], List[int], Tensor[int]].
         TypeError: If `coordinate_transformation_mode` is not a string.
         TypeError: If `coordinate_transformation_mode` is not in the support list.
 
     Supported Platforms:
-        ``GPU`` ``CPU``
+        ``GPU`` ``CPU`` `Ascend`
 
     Examples:
         >>> input = Tensor([[[1, 2, 3], [4, 5, 6]]], mindspore.float32)