take some notes about sparse API (#45720)

paddle/fluid/pybind/eager_functions.cc

@@ -523,8 +523,8 @@ static PyObject* eager_api_sparse_coo_tensor(PyObject* self,
       std::dynamic_pointer_cast<phi::DenseTensor>(non_zero_indices.impl());
   auto dense_elements =
       std::dynamic_pointer_cast<phi::DenseTensor>(non_zero_elements.impl());
-  // TODO(zhangkaihuo): After create SparseTensor, call coalesced() to sort and
-  // merge duplicate indices
+  // TODO(zhangkaihuo): After creating SparseCooTensor, call coalesced() to sort
+  // and merge duplicate indices
   std::shared_ptr<phi::SparseCooTensor> coo_tensor =
       std::make_shared<phi::SparseCooTensor>(
           *dense_indices, *dense_elements, phi::make_ddim(dense_shape));
@@ -537,7 +537,7 @@ static PyObject* eager_api_sparse_coo_tensor(PyObject* self,
   autograd_meta->SetStopGradient(static_cast<bool>(stop_gradient));
   if (!autograd_meta->GetMutableGradNode()) {
     VLOG(3) << "Tensor(" << name
-            << ") have not GradNode, add GradNodeAccumulation for it.";
+            << ") doesn't have GradNode, add GradNodeAccumulation to it.";
     autograd_meta->SetGradNode(
         std::make_shared<egr::GradNodeAccumulation>(autograd_meta));
   }

paddle/fluid/pybind/eager_utils.cc

@@ -92,7 +92,7 @@ int TensorDtype2NumpyDtype(phi::DataType dtype) {
 }
 
 bool PyObject_CheckLongOrConvertToLong(PyObject** obj) {
-  if ((PyLong_Check(*obj) && !PyBool_Check(*obj))) {
+  if (PyLong_Check(*obj) && !PyBool_Check(*obj)) {
     return true;
   }
 
@@ -129,7 +129,7 @@ bool PyObject_CheckStr(PyObject* obj) { return PyUnicode_Check(obj); }
 bool CastPyArg2AttrBoolean(PyObject* obj, ssize_t arg_pos) {
   if (obj == Py_None) {
     return false;  // To be compatible with QA integration testing. Some
-                   // test case pass in None.
+                   // test cases pass in None.
   } else if (obj == Py_True) {
     return true;
   } else if (obj == Py_False) {
@@ -305,7 +305,7 @@ std::vector<int> CastPyArg2VectorOfInt(PyObject* obj, size_t arg_pos) {
     Py_ssize_t len = PyList_Size(obj);
     PyObject* item = nullptr;
     for (Py_ssize_t i = 0; i < len; i++) {
-      item = PyList_GetItem(obj, i);
+      item = PyList_GET_ITEM(obj, i);
       if (PyObject_CheckLongOrConvertToLong(&item)) {
         result.emplace_back(static_cast<int>(PyLong_AsLong(item)));
       } else {
@@ -321,13 +321,13 @@ std::vector<int> CastPyArg2VectorOfInt(PyObject* obj, size_t arg_pos) {
     Py_ssize_t len = PyTuple_Size(obj);
     PyObject* item = nullptr;
     for (Py_ssize_t i = 0; i < len; i++) {
-      item = PyTuple_GetItem(obj, i);
+      item = PyTuple_GET_ITEM(obj, i);
       if (PyObject_CheckLongOrConvertToLong(&item)) {
         result.emplace_back(static_cast<int>(PyLong_AsLong(item)));
       } else {
         PADDLE_THROW(platform::errors::InvalidArgument(
             "argument (position %d) must be "
-            "list of bool, but got %s at pos %d",
+            "list of int, but got %s at pos %d",
             arg_pos + 1,
             reinterpret_cast<PyTypeObject*>(item->ob_type)->tp_name,
             i));

paddle/phi/api/include/tensor.h

@@ -574,7 +574,7 @@ class PADDLE_API Tensor final {
    * unified to Tensor, but Tensor itself is heterogeneous.
    *
    * Tensor can generally be represented by void* and size_t, place.
-   * This is suitable for most scenarios including CPU, GPU, HIP, CPU, etc.,
+   * This is suitable for most scenarios including CPU, GPU, HIP, NPU, etc.,
    * but there are a few cases where this definition cannot be described,
    * such as the Tensor representation in third-party lib such as Metal,
    * OpenCL, etc., as well as some special Tensor implementations, including

paddle/phi/core/dense_tensor.h

@@ -29,7 +29,7 @@ namespace phi {
 
 class DenseTensorUtils;
 
-/// \brief The Dense tensor store values in a contiguous sequential block
+/// \brief The Dense tensor stores values in a contiguous sequential block
 /// of memory where all values are represented. Tensors or multi-dimensional
 /// arrays are used in math operators.
 /// During the entire life cycle of a DenseTensor, its device type and key

paddle/phi/core/tensor_base.h

@@ -55,12 +55,12 @@ class TensorBase {
   virtual bool valid() const = 0;
 
   /// \brief Test whether the storage is allocated.
-  /// return Whether the storage is allocated.
+  /// \return Whether the storage is allocated.
   virtual bool initialized() const = 0;
 
   // TODO(Aurelius84): This interface is under intermediate state now.
   // We will remove DataType argument in the future. Please DO NOT
-  // rely on Datatype to much when design and implement other feature.
+  // rely on Datatype too much when designing and implementing other features.
 
   /// \brief Allocate memory with requested size from allocator.
   /// \return The mutable data pointer value of type T.
@@ -70,7 +70,7 @@ class TensorBase {
 
   /// \brief Return the type information of the derived class to support
   /// safely downcast in non-rtti environment.
-  /// return The type information of the derived class.
+  /// \return The type information of the derived class.
   TypeInfo<TensorBase> type_info() const { return type_info_; }
 
  private: