Fix elemenewist_add_op elementwise_add_op [fluid_ops] (#74245)

Author: co63oc

paddle/fluid/pybind/compiled_program.cc

@@ -387,7 +387,7 @@ void BindCompiledProgram(pybind11::module &m) {  // NOLINT
             self.fuse_gemm_epilogue_ = b;
           },
           R"DOC((bool, optional): fuse_gemm_epilogue indicate whether
-                to fuse matmul_op, elemenewist_add_op and activation_op,
+                to fuse matmul_op, elementwise_add_op and activation_op,
                 it may make the execution faster. Default is False.
 
                 Examples:
@@ -410,7 +410,7 @@ void BindCompiledProgram(pybind11::module &m) {  // NOLINT
             PADDLE_ENFORCE_NE(self.IsFinalized(),
                               true,
                               common::errors::PreconditionNotMet(
-                                  "BuildStrategy has been finlaized, cannot be "
+                                  "BuildStrategy has been finalized, cannot be "
                                   "configured again."));
             self.fuse_dot_product_attention_ = b;
           },

paddle/phi/backends/device_ext.h

@@ -395,7 +395,7 @@ struct C_DeviceInterface {
                               size_t size);
 
   /**
-   * @brief Asynchonrize memory copy from host to device
+   * @brief Asynchronize memory copy from host to device
    *
    * @param[C_Device]   device     Core fill it with a physical id
    * @param[C_Stream]   stream
@@ -410,7 +410,7 @@ struct C_DeviceInterface {
                                     size_t size);
 
   /**
-   * @brief Asynchonrize memory copy from device to host
+   * @brief Asynchronize memory copy from device to host
    *
    * @param[C_Device]   device     Core fill it with a physical id
    * @param[C_Stream]   stream
@@ -425,7 +425,7 @@ struct C_DeviceInterface {
                                     size_t size);
 
   /**
-   * @brief Asynchonrize memory copy from device to device
+   * @brief Asynchronize memory copy from device to device
    *
    * @param[C_Device]   device     Core fill it with a physical id
    * @param[C_Stream]   stream
@@ -440,7 +440,7 @@ struct C_DeviceInterface {
                                     size_t size);
 
   /**
-   * @brief Peer asynchonrize memory copy from host to device
+   * @brief Peer asynchronize memory copy from host to device
    *
    * @param[C_Device]   device     Core fill it with a physical id
    * @param[C_Stream]   stream

paddle/phi/backends/onednn/axpy_handler.h

@@ -53,7 +53,7 @@ class OneDNNAXPYHandler {
   // Private implementation idiom to hide dependency on oneDNN headers.
   class Impl;
   // We need custom deleter, since the compiler is unable to parameterize
-  // an allocator's default deleter due to incomple type.
+  // an allocator's default deleter due to incomplete type.
   std::unique_ptr<Impl, void (*)(Impl*)> pimpl_;
 };
 }  // namespace funcs

paddle/phi/backends/onednn/matmul_utils.h

@@ -59,7 +59,7 @@ class MatmulOneDNNHandler : public OneDNNHandlerNoCachingT<XT, dnnl::matmul> {
     std::vector<int64_t> x_strides(x_dims.size() - 3, 1);
     std::vector<int64_t> y_strides(x_dims.size() - 3, 1);
     std::vector<int64_t> out_strides(x_dims.size() - 3, 1);
-    std::vector<int64_t> out_ddims(x_dims.size() - 3, 1);
+    std::vector<int64_t> out_dims(x_dims.size() - 3, 1);
 
     x_strides.reserve(x_dims.size());
     y_strides.reserve(x_dims.size());
@@ -78,20 +78,20 @@ class MatmulOneDNNHandler : public OneDNNHandlerNoCachingT<XT, dnnl::matmul> {
     }
 
     out_strides.insert(out_strides.end(), {M * N, N, 1});
-    out_ddims.insert(out_ddims.end(),
-                     {std::max(x_dims[MB_idx], y_dims[MB_idx]), M, N});
+    out_dims.insert(out_dims.end(),
+                    {std::max(x_dims[MB_idx], y_dims[MB_idx]), M, N});
 
     for (int i = x_dims.size() - 4; i >= 0; --i) {
-      out_ddims[i] = std::max(x_dims[i], y_dims[i]);
+      out_dims[i] = std::max(x_dims[i], y_dims[i]);
       x_strides[i] = x_dims[i + 1] * x_strides[i + 1];
       y_strides[i] = y_dims[i + 1] * y_strides[i + 1];
 
-      out_strides[i] = out_ddims[i + 1] * out_strides[i + 1];
+      out_strides[i] = out_dims[i + 1] * out_strides[i + 1];
     }
 
     auto x_md = memory::desc(x_dims, OneDNNGetDataType<XT>(), x_strides);
     auto y_md = memory::desc(y_dims, OneDNNGetDataType<YT>(), y_strides);
-    auto out_md = memory::desc(out_ddims, OneDNNGetDataType<OT>(), out_strides);
+    auto out_md = memory::desc(out_dims, OneDNNGetDataType<OT>(), out_strides);
 
     this->AcquireForwardPrimitiveDescriptor(x_md, y_md, out_md);
   }

paddle/phi/core/dense_tensor.cc

@@ -230,8 +230,7 @@ void DenseTensor::set_meta(const DenseTensorMeta& meta) {
    1. Designed behaviour: DenseTensor constructed with its underlying storage_
    initialized
    2. Legacy behaviour(fluid): DenseTensor constructed using default
-   constructor, where
-                               storage_ won't be initialized until the first
+   constructor, where storage_ won't be initialized until the first
    call to mutable_data(place)
    */
 void DenseTensor::ResizeAndAllocate(const DDim& dims) {

paddle/phi/core/framework/var_type_helper.h

@@ -281,8 +281,7 @@ extern inline paddle::framework::proto::VarType::Type ToRealType(
     default:
       PADDLE_THROW(common::errors::Unimplemented(
           "Unknown complex value data type (%s), now only support complex64 "
-          "and "
-          "complex128.",
+          "and complex128.",
           VarDataTypeToString(t)));
   }
 }

paddle/phi/kernels/funcs/fft_fill_conj.h

@@ -80,7 +80,7 @@ HOSTDEVICE inline bool is_conj_part(const int64_t dst_idx,
 
 // FFTFillConjFunctor fill the destination tensor with source tensor and
 // conjugate symmetry element of source tensor .
-// Use framework::ForRange to iterate destination element with
+// Use phi::ForRange to iterate destination element with
 // supporting different device
 template <typename C>
 struct FFTFillConjFunctor {

paddle/phi/kernels/funcs/sequence_padding.cu

@@ -164,14 +164,6 @@ class UnpaddingDenseTensorFunctor<phi::GPUContext, T> {
               pad_seq_len,
               step_width,
               layout);
-    /*
-    if (!norm_by_times && seq_num == 1UL && pad_seq_len == max_seq_len) {
-      paddle::framework::TensorCopy(pad_tensor, dev_ctx.GetPlace(), dev_ctx,
-    seq_tensor);
-      seq_tensor->Resize(seq_tensor_dims);
-      return;
-    }
-    */
 
     const int kBlockSize = 512;
 

paddle/phi/kernels/funcs/strided_slice.h

@@ -529,9 +529,9 @@ void StridedSliceGradCompute(const Context& dev_ctx,
                              const std::vector<int>& infer_flags,
                              const std::vector<int>& decrease_axis,
                              TensorArray* x_grad) {
-  // Note(weixin):Since the shape of `framework::GradVarName("Input")` of
+  // Note(weixin):Since the shape of `x_grad` of
   // StridedSliceGrad cannot be calculated by
-  // `framework::GradVarName("Output")`, the dim of "Input" is used to
+  // `out_grad`, the dim of "x" is used to
   // calculate the output shape. when set it to inplace OP, there may be
   // some problems.
   const int64_t size = x.size();

test/legacy_test/test_fold_op.py

@@ -229,7 +229,7 @@ def test_padding_shape():
                 )
 
             def test_dilations_shape():
-                # dialtions_size must be 2
+                # dilations_size must be 2
                 x = paddle.randn(shape=[2, 6, 6], dtype="float32")
                 out = fold(
                     x,
@@ -239,7 +239,7 @@ def test_dilations_shape():
                 )
 
             def test_strides_shape():
-                # strids_size must be 2
+                # strides_size must be 2
                 x = paddle.randn(shape=[2, 6, 6], dtype="float32")
                 out = fold(
                     x,