diff --git a/paddle/pten/core/CMakeLists.txt b/paddle/pten/core/CMakeLists.txt
index 7c8ace2bc7ef4..2c134b8c6d2e2 100644
--- a/paddle/pten/core/CMakeLists.txt
+++ b/paddle/pten/core/CMakeLists.txt
@@ -18,6 +18,7 @@ cc_library(pten_device_context SRCS device_context.cc DEPS tensor_base )
 
 
 cc_library(meta_tensor SRCS meta_tensor.cc DEPS tensor_base tensor_meta dense_tensor)
+cc_library(infermeta_utils SRCS infermeta_utils.cc DEPS meta_tensor)
 
 cc_test(unroll_array_ops_test SRCS unroll_array_ops_test.cc)
 cc_library(ddim SRCS ddim.cc DEPS eigen3 boost enforce)
diff --git a/paddle/pten/core/infermeta_utils.cc b/paddle/pten/core/infermeta_utils.cc
new file mode 100644
index 0000000000000..9f0037d18edf6
--- /dev/null
+++ b/paddle/pten/core/infermeta_utils.cc
@@ -0,0 +1,73 @@
+/* Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+
+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 "paddle/pten/core/infermeta_utils.h"
+
+namespace pten {
+
+void InferMetaContext::SetMetaConfig(MetaConfig config) {
+  config_ = std::move(config);
+}
+
+void InferMetaContext::EmplaceBackInput(
+    std::shared_ptr<pten::MetaTensor> input) {
+  int index = inputs_.size();
+  inputs_.emplace_back(std::move(input));
+  input_range_.emplace_back(std::pair<int, int>(index, index + 1));
+}
+void InferMetaContext::EmplaceBackOutput(
+    std::shared_ptr<pten::MetaTensor> output) {
+  int index = outputs_.size();
+  outputs_.emplace_back(std::move(output));
+  output_range_.emplace_back(std::pair<int, int>(index, index + 1));
+}
+void InferMetaContext::EmplaceBackAttr(paddle::any attr) {
+  attrs_.emplace_back(std::move(attr));
+}
+
+void InferMetaContext::EmplaceBackInputs(
+    paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> inputs) {
+  int index = inputs_.size();
+  input_range_.emplace_back(std::pair<int, int>(index, index + inputs.size()));
+  inputs_.insert(inputs_.end(),
+                 std::make_move_iterator(inputs.begin()),
+                 std::make_move_iterator(inputs.end()));
+}
+void InferMetaContext::EmplaceBackOutputs(
+    paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> outputs) {
+  int index = outputs_.size();
+  output_range_.emplace_back(
+      std::pair<int, int>(index, index + outputs.size()));
+  outputs_.insert(outputs_.end(),
+                  std::make_move_iterator(outputs.begin()),
+                  std::make_move_iterator(outputs.end()));
+}
+
+const std::pair<int, int>& InferMetaContext::InputRangeAt(size_t idx) const {
+  return input_range_.at(idx);
+}
+const std::pair<int, int>& InferMetaContext::OutputRangeAt(size_t idx) const {
+  return output_range_.at(idx);
+}
+
+const MetaConfig& InferMetaContext::GetMetaConfig() const { return config_; }
+
+const MetaTensor& InferMetaContext::InputAt(size_t idx) const {
+  return *inputs_.at(idx);
+}
+MetaTensor* InferMetaContext::MutableOutputAt(size_t idx) {
+  return outputs_.at(idx).get();
+}
+
+}  // namespace pten
diff --git a/paddle/pten/core/infermeta_utils.h b/paddle/pten/core/infermeta_utils.h
new file mode 100644
index 0000000000000..c6812dee92b6a
--- /dev/null
+++ b/paddle/pten/core/infermeta_utils.h
@@ -0,0 +1,170 @@
+/* Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+
+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. */
+
+#pragma once
+
+#include <string>
+#include <utility>
+
+#include "paddle/pten/core/meta_tensor.h"
+#include "paddle/utils/small_vector.h"
+
+namespace pten {
+
+// TODO(chenweihang): add other flags if needed
+struct MetaConfig {
+  bool is_runtime{true};
+
+  MetaConfig() = default;
+
+  // supporting implicit construction is easier to use
+  MetaConfig(bool is_runtime) : is_runtime(is_runtime) {}  // NOLINT
+};
+
+class InferMetaContext {
+ public:
+  InferMetaContext() = default;
+  explicit InferMetaContext(MetaConfig config) : config_(config) {}
+
+  void SetMetaConfig(MetaConfig config);
+  void EmplaceBackInput(std::shared_ptr<pten::MetaTensor> input);
+  void EmplaceBackOutput(std::shared_ptr<pten::MetaTensor> output);
+  void EmplaceBackAttr(paddle::any attr);
+
+  void EmplaceBackInputs(
+      paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> inputs);
+  void EmplaceBackOutputs(
+      paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> outputs);
+
+  const std::pair<int, int>& InputRangeAt(size_t idx) const;
+  const std::pair<int, int>& OutputRangeAt(size_t idx) const;
+
+  const MetaConfig& GetMetaConfig() const;
+  const MetaTensor& InputAt(size_t idx) const;
+  MetaTensor* MutableOutputAt(size_t idx);
+
+  template <typename AttrType>
+  AttrType AttrAt(size_t idx) {
+    try {
+      return paddle::any_cast<AttrType>(attrs_.at(idx));
+    } catch (paddle::bad_any_cast&) {
+      PADDLE_THROW(paddle::platform::errors::InvalidArgument(
+          "Attribute cast error in InferMeta Context."));
+    }
+  }
+
+ private:
+  MetaConfig config_;
+
+  // NOTE(chenweihang): Because the MetaTensor is a base class, and MetaTensor
+  // objects are all created in each round, so we have to use smart pointer
+  // here, maybe we can implemented a new InferMetaContext and a series utils
+  // specifically for fluid to avoid using shared_ptr
+  paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> inputs_;
+  paddle::SmallVector<std::shared_ptr<pten::MetaTensor>> outputs_;
+  paddle::SmallVector<paddle::any> attrs_;
+
+  paddle::SmallVector<std::pair<int, int>> input_range_;
+  paddle::SmallVector<std::pair<int, int>> output_range_;
+};
+
+#define PT_INFER_META(...) \
+  ::pten::InferMetaFnImpl<decltype(&__VA_ARGS__), &__VA_ARGS__>::Call
+
+#define PT_SPECIALIZE_InferMetaFnCallHelper_FOR_ATTRIBUTE(attr_type)           \
+  template <typename... Tail>                                                  \
+  struct InferMetaFnCallHelper<attr_type, Tail...> {                           \
+    template <int in_idx, int attr_idx, int out_idx, typename... PreviousArgs> \
+    static void Call(InferMetaContext* ctx, PreviousArgs&... pargs) {          \
+      static_assert(out_idx == 0,                                              \
+                    "InferMeta's Attributes should appear before Outputs.");   \
+      attr_type arg = ctx->AttrAt<attr_type>(attr_idx);                        \
+      InferMetaFnCallHelper<                                                   \
+          Tail...>::template Call<in_idx, attr_idx + 1, out_idx>(pargs...,     \
+                                                                 arg);         \
+    }                                                                          \
+  }
+
+template <typename T>
+struct InferMetaTypeTag {};
+
+template <typename Fn, Fn fn>
+struct InferMetaFnImpl;
+
+template <typename Return, typename... Args, Return (*infer_meta_fn)(Args...)>
+struct InferMetaFnImpl<Return (*)(Args...), infer_meta_fn> {
+  static void Call(InferMetaContext* ctx) {
+    InferMetaFnCallHelper<Args...,
+                          InferMetaTypeTag<int>>::template Call<0, 0, 0>(ctx);
+  }
+
+ private:
+  template <typename... RemainingArgs>
+  struct InferMetaFnCallHelper;
+
+  template <typename... Tail>
+  struct InferMetaFnCallHelper<const MetaTensor&, Tail...> {
+    template <int in_idx, int attr_idx, int out_idx, typename... PreviousArgs>
+    static void Call(InferMetaContext* ctx, PreviousArgs&... pargs) {
+      static_assert(attr_idx == 0,
+                    "InferMeta's Input should appear before Attributes.");
+      static_assert(out_idx == 0,
+                    "InferMeta's Input should appear before Outputs.");
+      const std::pair<int, int> range = ctx->InputRangeAt(in_idx);
+      const MetaTensor& arg = ctx->InputAt(range.first);
+      InferMetaFnCallHelper<
+          Tail...>::template Call<in_idx + 1, attr_idx, out_idx>(ctx,
+                                                                 pargs...,
+                                                                 arg);
+    }
+  };
+
+  // TODO(chenweihang): support vector<MetaTensor> input later
+
+  template <typename... Tail>
+  struct InferMetaFnCallHelper<MetaTensor*, Tail...> {
+    template <int in_idx, int attr_idx, int out_idx, typename... PreviousArgs>
+    static void Call(InferMetaContext* ctx, PreviousArgs&... pargs) {
+      const std::pair<int, int> range = ctx->OutputRangeAt(out_idx);
+      MetaTensor* arg = ctx->MutableOutputAt(range.first);
+      InferMetaFnCallHelper<
+          Tail...>::template Call<in_idx, attr_idx, out_idx + 1>(ctx,
+                                                                 pargs...,
+                                                                 arg);
+    }
+  };
+
+  // TODO(chenweihang): support vector<MetaTensor> output later
+
+  template <typename... Tail>
+  struct InferMetaFnCallHelper<MetaConfig, Tail...> {
+    template <int in_idx, int attr_idx, int out_idx, typename... PreviousArgs>
+    static void Call(InferMetaContext* ctx, PreviousArgs&... pargs) {
+      const MetaConfig& arg = ctx->GetMetaConfig();
+      InferMetaFnCallHelper<Tail...>::template Call<in_idx, attr_idx, out_idx>(
+          ctx, pargs..., arg);
+    }
+  };
+
+  /* End case */
+  template <typename T>
+  struct InferMetaFnCallHelper<InferMetaTypeTag<T>> {
+    template <int in_idx, int attr_idx, int out_idx>
+    static void Call(InferMetaContext* ctx, Args&... args) {
+      return infer_meta_fn(args...);
+    }
+  };
+};
+
+}  // namespace pten