Fix typos initial (#68702)

co63oc · web-flow · commit 241999bef9b7 · 2024-10-15T13:16:30.000+08:00
diff --git a/paddle/phi/kernels/autotune/auto_tune_base.h b/paddle/phi/kernels/autotune/auto_tune_base.h
@@ -285,7 +285,7 @@ MakeGatherGemmScatterTuner(ReturnType (*func)(T, T, Args...)) {
                                     Args...>::Instance(func);
 }
 
-// Define the auto_tuner inital object.
+// Define the auto_tuner initial object.
 #define DEFINE_AUTOTUNER_COMMON_OBJ(name)                                \
   template <typename T, typename ReturnType, typename... Args>           \
   class name##AutoTuner                                                  \
@@ -305,7 +305,7 @@ MakeGatherGemmScatterTuner(ReturnType (*func)(T, T, Args...)) {
     }                                                                    \
   };
 
-// Define the auto_tuner inital function.
+// Define the auto_tuner initial function.
 #define DEFINE_AUTOTUNER_FN(name)                                    \
   template <typename T, typename ReturnType, typename... Args>       \
   static name##AutoTuner<T, ReturnType, Args...>* Make##name##Tuner( \
diff --git a/paddle/phi/kernels/autotune/switch_autotune.cc b/paddle/phi/kernels/autotune/switch_autotune.cc
@@ -39,7 +39,7 @@ void AutoTuneStatus::Update() {
     return;
   }
 
-  // This fuction is called when each iter finished.
+  // This function is called when each iter finished.
   if (current_steps_id_ + 1 < start_step_id_) {
     use_autotune_ = false;
   } else if (current_steps_id_ + 1 >= start_step_id_ &&
diff --git a/paddle/phi/kernels/coalesce_tensor_kernel.cc b/paddle/phi/kernels/coalesce_tensor_kernel.cc
@@ -84,7 +84,7 @@ void GetMemSizeAndDtype(const std::vector<const DenseTensor *> &lod_tensors,
         lod_tensors[i]->initialized() ? lod_tensors[i]->data() : nullptr;
     VLOG(4) << size << " " << len;
     ss << "input(" << i << "-th tensor) dim:(" << lod_tensors[i]->dims() << ") "
-       << " addres:" << ptr << " len: " << len << ", ";
+       << " address:" << ptr << " len: " << len << ", ";
     *numel += len;
   }
   VLOG(10) << ss.str();
diff --git a/paddle/phi/kernels/cpu/auc_kernel.cc b/paddle/phi/kernels/cpu/auc_kernel.cc
@@ -163,7 +163,7 @@ void AucKernel(const Context &dev_ctx,
   auto *origin_stat_neg = dev_ctx.template Alloc<int64_t>(stat_neg_out);
   auto *auc_value = dev_ctx.template Alloc<double>(auc);
 
-  // Just for pass UT, since UT's input & output connot be set same var
+  // Just for pass UT, since UT's input & output cannot be set same var
   auto *stat_pos_in_tensor = &stat_pos;
   auto *stat_neg_in_tensor = &stat_neg;
   auto *pos_in_data = stat_pos.data<int64_t>();
diff --git a/paddle/phi/kernels/cpu/kthvalue_kernel.cc b/paddle/phi/kernels/cpu/kthvalue_kernel.cc
@@ -91,7 +91,7 @@ void KthvalueKernel(const Context& dev_ctx,
                       1,
                       common::errors::InvalidArgument(
                           "the k in the kthvalue must less equal than the "
-                          "elemenents number of the input X, but received %d .",
+                          "elements number of the input X, but received %d .",
                           k));
 
     phi::Copy<Context>(dev_ctx, x, dev_ctx.GetPlace(), false, output);
diff --git a/paddle/phi/kernels/cpu/lstsq_kernel.cc b/paddle/phi/kernels/cpu/lstsq_kernel.cc
@@ -60,7 +60,7 @@ void LstsqKernel(const Context& dev_ctx,
   int max_solu_stride = std::max(y_stride, ori_solu_stride);
   int min_solu_stride = std::min(y_stride, ori_solu_stride);
 
-  // lapack is a column-major storge, transpose make the input to
+  // lapack is a column-major storage, transpose make the input to
   // have a continuous memory layout
   int info = 0;
   int m = static_cast<int>(x_dims[dim_size - 2]);
@@ -204,7 +204,7 @@ void LstsqKernel(const Context& dev_ctx,
     rwork_data = dev_ctx.template Alloc<ValueType>(rwork);
   }
 
-  // "iwork" workspace array is relavant only for "gelsd" driver
+  // "iwork" workspace array is relevant only for "gelsd" driver
   DenseTensor* iwork = new DenseTensor();
   int* iwork_data = nullptr;
   if (driver == LapackDriverType::Gelsd) {
diff --git a/paddle/phi/kernels/cpu/rnn_kernel.cc b/paddle/phi/kernels/cpu/rnn_kernel.cc
@@ -217,7 +217,7 @@ struct Layer {
                   const std::string& mode,
                   bool is_test,
                   DenseTensor* cache_input) {
-    // crate the temp input for the X * W_ih^T + Bias_ih
+    // create the temp input for the X * W_ih^T + Bias_ih
     const int& hidden_size = weight.dims()[0];  // NOLINT
     cache_input->Resize(
         common::make_ddim({input.dims()[0], input.dims()[1], hidden_size}));
@@ -272,7 +272,7 @@ struct Layer {
                    DenseTensor* last_c,
                    const DenseTensor& mask_tensor,
                    const std::string& mode) {
-    // in the output, if mask flag is 0, we will retun the zero data
+    // in the output, if mask flag is 0, we will return the zero data
     auto& place = *dev_ctx.eigen_device();
     auto out = EigenMatrix<T>::Reshape(*output, output->dims().size() - 1);
     auto mask = EigenMatrix<T>::From(
diff --git a/paddle/phi/kernels/dist_grad_kernel.cc b/paddle/phi/kernels/dist_grad_kernel.cc
@@ -65,8 +65,8 @@ void DistGradKernel(const Context& dev_ctx,
       dev_ctx, t, out, out_grad, p, -1, 1e-12, false, true, &x_grad_tmp);
 
   if (x_grad) {
-    // do reduce, the implemetation of cpu SumKernel has bug, it changes
-    // the dims of output iternally, so we Resize x/y_grad twice.
+    // do reduce, the implementation of cpu SumKernel has bug, it changes
+    // the dims of output internally, so we Resize x/y_grad twice.
     auto res_x = GetReduceDims(x_grad_tmp.dims(), x.dims());
     if (!std::get<0>(res_x).empty()) {
       x_grad->Resize(common::make_ddim(std::get<1>(res_x)));
diff --git a/paddle/phi/kernels/fused_attention_kernel.h b/paddle/phi/kernels/fused_attention_kernel.h
@@ -42,8 +42,8 @@ namespace phi {
  * @param num_heads           The number head for multi_head_attention.
  * @param transpose_qkv_wb    The qkv_w shape is (h, 3h), do transpose to it.
  * @param pre_layer_norm      if true, the attention op uses pre_layer_norm
- *                            architecure, else, uses post_layer_norm
- * architecuture. [default false].
+ *                            architecture, else, uses post_layer_norm
+ * architecture. [default false].
  * @param epsilon             Constant for numerical stability [default 1e-5].
  * @param attn_dropout_rate   Probability of setting units to zero.
  * @param is_test             (bool, default false) Set to true for inference
diff --git a/paddle/phi/kernels/fusion/gpu/block_attn.h b/paddle/phi/kernels/fusion/gpu/block_attn.h
@@ -1689,7 +1689,7 @@ void blha(const phi::GPUContext &dev_ctx,
       mask_broadcast_num_heads = false;
     } else {
       PADDLE_THROW(errors::InvalidArgument(
-          "Unknow dimension for attn_mask, the q_num_head(2nd) "
+          "Unknown dimension for attn_mask, the q_num_head(2nd) "
           "dimension is invalid, it should be 1 or q_num_head(%d), "
           "but got %d",
           q_num_head,
diff --git a/paddle/phi/kernels/fusion/gpu/block_multi_head_attention_kernel.cu b/paddle/phi/kernels/fusion/gpu/block_multi_head_attention_kernel.cu
@@ -544,7 +544,7 @@ void DispatchWithDtype(
       //            unpadding_v.numel(),
       //            "unpadding_v",
       //            unpadding_v.numel());
-      // Reshape fmha_buf to 3-D because FlashAttnUnpaddedKernel requries
+      // Reshape fmha_buf to 3-D because FlashAttnUnpaddedKernel requires
       // q,k,v,out all in 3-D [token_num, q_num_head, dim_head].
       auto fmha_shape = fmha_buf.dims();
       fmha_buf.Resize({token_num, q_num_head, dim_head});
diff --git a/paddle/phi/kernels/fusion/xpu/multi_encoder_xpu_kernel.cc b/paddle/phi/kernels/fusion/xpu/multi_encoder_xpu_kernel.cc
@@ -141,8 +141,8 @@ void MultiEncoderXPUKernel(
       fc_weight_data_XPUTypeFP16.push_back(
           reinterpret_cast<const XPUTypeFP16*>(fc_weight[i]->data()));
     } else {
-      // Int8 weight also convert to int16_t* for temperary storage.
-      // The kenerl dytpe of int8 is choosen by quant_type in
+      // Int8 weight also convert to int16_t* for temporary storage.
+      // The kernel dtype of int8 is chosen by quant_type in
       // xpu::transformer_encoder
       fc_weight_data_int16_t.push_back(
           reinterpret_cast<const int16_t*>(fc_weight[i]->data()));
@@ -222,7 +222,7 @@ void MultiEncoderXPUKernel(
     if (!enable_int8 && local_quant) {
       TRANSFORMER_ENCODER_KERNEL_IMPL(XPUTypeFP16, XPUTypeFP16, float)
     } else {
-      // The kenerl dytpe of int8 is choosen by quant_type in
+      // The kernel dtype of int8 is chosen by quant_type in
       // xpu::transformer_encoder This template args, int16_t, is only for skip
       // quant fc
       TRANSFORMER_ENCODER_KERNEL_IMPL(XPUTypeFP16, int16_t, int16_t)
diff --git a/paddle/phi/kernels/is_empty_kernel.cc b/paddle/phi/kernels/is_empty_kernel.cc
@@ -24,7 +24,7 @@ void IsEmptyKernel(const Context& dev_ctx,
                    const DenseTensor& x,
                    DenseTensor* out) {
   // Note: is_empty is always executed on CPU and the output data should
-  // always be allocated for CPUPlace. We reigister CUDA kernel for this op to
+  // always be allocated for CPUPlace. We register CUDA kernel for this op to
   // avoid the unnecessary data transform.
   bool* out_data = dev_ctx.template HostAlloc<bool>(out);
   out_data[0] = common::product(x.dims()) == 0;
diff --git a/paddle/phi/kernels/memcpy_kernel.cc b/paddle/phi/kernels/memcpy_kernel.cc
@@ -72,7 +72,7 @@ void MemcpyD2HKernel(const Context& dev_ctx,
 
     default:
       PADDLE_THROW(errors::InvalidArgument(
-          "Arugment 'dst_place_type' only support 0-1, but got: %d",
+          "Argument 'dst_place_type' only support 0-1, but got: %d",
           dst_place_type));
       break;
   }
diff --git a/paddle/phi/kernels/primitive/compute_primitives.h b/paddle/phi/kernels/primitive/compute_primitives.h
@@ -471,8 +471,8 @@ __device__ __forceinline__ void ElementwiseConstant(OutT* out, OpFunc compute) {
 }
 
 /*
- * @brief Get ReturnsCount random data fromm compute according to state, state
- * can be curandStatePhilox4_32_10_t, hiprandStatePhilox4_32_10_t which has beed
+ * @brief Get ReturnsCount random data from compute according to state, state
+ * can be curandStatePhilox4_32_10_t, hiprandStatePhilox4_32_10_t which has been
  * initialized.
  *
  * @template paraments
@@ -580,8 +580,8 @@ __device__ __forceinline__ void Cumsum(OutT* out,
  * out: The register pointer of out, the size is 2.
  * in: The register pointer of input, the size is 2.
  * num: The num of this block
- * monotonic_type: if monotonic_type = 1 then sorted in ascending order, eles
- * sorted in escending.
+ * monotonic_type: if monotonic_type = 1 then sorted in ascending order, else
+ * sorted in descending.
  */
 #define SHARED_SIZE_LIMIT 1024
 // each thread load 2 data from global memory so SHARED_SIZE_LIMIT must
@@ -638,8 +638,8 @@ __device__ __forceinline__ void Sort(OutT* out,
  * in: The register pointer of input, the size is 2.
  * in_index: The register pointer of in_index, the size is 2.
  * num: The num of this block.
- * monotonic_type: if monotonic_type = 1 then sorted in ascending order, eles
- * sorted in escending.
+ * monotonic_type: if monotonic_type = 1 then sorted in ascending order, else
+ * sorted in descending.
  */
 template <typename InT, typename OutT, typename IndexType>
 __device__ __forceinline__ void Sort(OutT* out,
diff --git a/paddle/phi/kernels/primitive/functor_primitives.h b/paddle/phi/kernels/primitive/functor_primitives.h
@@ -234,7 +234,7 @@ struct DivFunctor<T,
   inline T initial() { return static_cast<T>(1.0f); }
 
   inline HOSTDEVICE T operator()(const T a, const T b) const {
-    // For int32/int64, need to check whether the divison is zero.
+    // For int32/int64, need to check whether the division is zero.
     PADDLE_ENFORCE_NE(b,
                       0,
                       common::errors::InvalidArgument(
diff --git a/paddle/phi/kernels/selected_rows/impl/dgc_clip_by_norm_kernel_impl.h b/paddle/phi/kernels/selected_rows/impl/dgc_clip_by_norm_kernel_impl.h
@@ -38,7 +38,7 @@ void DGCClipByNormKernel(const Context& dev_ctx,
   if (static_cast<int>(*current_step) < static_cast<int>(rampup_begin_step)) {
     VLOG(10) << "current_step:" << *current_step
              << " < rampup_begin_step:" << rampup_begin_step
-             << " so does't use dgc_clip_by_norm";
+             << " so doesn't use dgc_clip_by_norm";
     return;
   }
 
diff --git a/paddle/phi/kernels/xpu/cross_entropy_kernel.cc b/paddle/phi/kernels/xpu/cross_entropy_kernel.cc
@@ -62,7 +62,7 @@ void CrossEntropyWithSoftmaxKernel(const Context& dev_ctx,
     // transpose before softmax due to the following two reasons:
     // 1. the XPU cross_entropy APIs supports cross entropy on the last dim
     // only, so the transpose here is unavoidable for them.
-    // 2. the XPU softmax api would do the transpose internaly if axis is not
+    // 2. the XPU softmax api would do the transpose internally if axis is not
     // the last dim and we can eliminate a transpose call if we explicitly
     // transpose the inputs before the softmax calculation.
     XPUType* logits_trans = RAII_GUARD.alloc_l3_or_gm<XPUType>(len);

Original file line number	Diff line number	Diff line change
`@@ -39,7 +39,7 @@ void AutoTuneStatus::Update() {`
`39`	`39`	`return;`
`40`	`40`	`}`
`41`	`41`
`42`		`- // This fuction is called when each iter finished.`
	`42`	`+ // This function is called when each iter finished.`
`43`	`43`	`if (current_steps_id_ + 1 < start_step_id_) {`
`44`	`44`	`use_autotune_ = false;`
`45`	`45`	`} else if (current_steps_id_ + 1 >= start_step_id_ &&`
Original file line number	Diff line number	Diff line change
`@@ -84,7 +84,7 @@ void GetMemSizeAndDtype(const std::vector<const DenseTensor *> &lod_tensors,`
`84`	`84`	`lod_tensors[i]->initialized() ? lod_tensors[i]->data() : nullptr;`
`85`	`85`	`VLOG(4) << size << " " << len;`
`86`	`86`	`ss << "input(" << i << "-th tensor) dim:(" << lod_tensors[i]->dims() << ") "`
`87`		`- << " addres:" << ptr << " len: " << len << ", ";`
	`87`	`+ << " address:" << ptr << " len: " << len << ", ";`
`88`	`88`	`*numel += len;`
`89`	`89`	`}`
`90`	`90`	`VLOG(10) << ss.str();`