Fix cahce cache (#60935)

co63oc · web-flow · commit d23f0f6ce5ef · 2024-01-19T10:23:39.000+08:00
diff --git a/paddle/fluid/eager/amp_auto_cast.h b/paddle/fluid/eager/amp_auto_cast.h
@@ -28,7 +28,7 @@ static inline bool NeedCast(const paddle::Tensor& tensor,
       paddle::platform::is_cuda_pinned_place(place) ||
       paddle::platform::is_xpu_place(place) ||
       paddle::platform::is_custom_place(place)) {
-    // CudaPinndePlace is added for varbase created by dataloader
+    // CudaPinnedPlace is added for varbase created by dataloader
     if ((data_type == phi::DataType::FLOAT32 ||
          data_type == phi::DataType::FLOAT16 ||
          data_type == phi::DataType::BFLOAT16) &&
diff --git a/paddle/fluid/eager/autograd_meta.h b/paddle/fluid/eager/autograd_meta.h
@@ -45,7 +45,7 @@ using AbstractAutogradMeta = paddle::AbstractAutogradMeta;
  *
  * 4. overrided_stop_gradient_
  * This member is used to finish some auto-prune related work, which indicate
- * user set stop_gradient should overrided the result indicated by framework.
+ * user set stop_gradient should override the result indicated by framework.
  * All non-parameter tensor's stop_gradient properties should be true. We will
  * pass stop_gradient when we find one who need it.
  *
diff --git a/paddle/fluid/eager/backward.cc b/paddle/fluid/eager/backward.cc
@@ -84,19 +84,19 @@ void EnforceGradNodeHasInput(GradNodeBase* node) {
 }
 
 void DuplicateCheck(const std::vector<paddle::Tensor>& inputs, bool is_input) {
-  std::unordered_set<AutogradMeta*> visisted_ins;
+  std::unordered_set<AutogradMeta*> visited_ins;
   std::string msg = is_input ? "inputs" : "outputs";
   for (auto const& in : inputs) {
     AutogradMeta* auto_grad_meta = EagerUtils::unsafe_autograd_meta(in);
     PADDLE_ENFORCE_EQ(
-        visisted_ins.count(auto_grad_meta),
+        visited_ins.count(auto_grad_meta),
         0,
         paddle::platform::errors::AlreadyExists(
             "%s contain duplicate tensor %s, please check %s carefully.",
             msg,
             in.name(),
             msg));
-    visisted_ins.insert(auto_grad_meta);
+    visited_ins.insert(auto_grad_meta);
   }
 }
 
diff --git a/paddle/fluid/eager/eager_amp_auto_cast.h b/paddle/fluid/eager/eager_amp_auto_cast.h
@@ -29,8 +29,8 @@ static inline bool NeedCast(const paddle::Tensor& tensor,
       paddle::platform::is_xpu_place(place) ||
       paddle::platform::is_custom_place(place) ||
       paddle::platform::is_cpu_place(place)) {
-    // CudaPinndePlace is added for varbase created by dataloader
-    // Cpu place is for differnt place tensor, when input1 is cpu and input2 is
+    // CudaPinnedPlace is added for varbase created by dataloader
+    // Cpu place is for different place tensor, when input1 is cpu and input2 is
     // gpu
     if ((data_type == phi::DataType::FLOAT32 ||
          data_type == phi::DataType::FLOAT16 ||
diff --git a/paddle/fluid/eager/eager_layout_auto_tune.h b/paddle/fluid/eager/eager_layout_auto_tune.h
@@ -135,7 +135,7 @@ inline std::shared_ptr<EagerLayoutTransformer> EagerLayoutAutotune(
         paddle::imperative::LayoutAutoTune::Instance().SetDefaultLayout(
             phi::DataLayout::NCHW);
       } else {
-        VLOG(4) << "DisableLayoutAutoTune accoding to Conv op"
+        VLOG(4) << "DisableLayoutAutoTune according to Conv op"
                 << " dtype : " << data_type << " format : " << (*attr);
         egr::Controller::Instance().DisableLayoutAutoTune();
         return transposer;
@@ -209,15 +209,15 @@ inline std::shared_ptr<EagerLayoutTransformer> EagerLayoutAutotune<int, int>(
     int* start_axis,
     int* stop_axis) {
   if (DesiredLayout() == phi::DataLayout::UNDEFINED) {
-    VLOG(4) << "Optimze Layout was not started" << op_name;
+    VLOG(4) << "Optimize Layout was not started" << op_name;
     return std::make_shared<EagerLayoutTransformer>(
         op_name, tensors_vector, tensors_vector[0][0].layout());
   }
 
-  bool no_tranpose = tensors_vector[0][0].layout() == DesiredLayout();
+  bool no_transpose = tensors_vector[0][0].layout() == DesiredLayout();
   bool is_valid = ((*start_axis) == 1 && (*stop_axis) == 3);
   if (op_name == "flatten" || op_name == "flatten_contiguous_range") {
-    if (no_tranpose && is_valid) {
+    if (no_transpose && is_valid) {
       return std::make_shared<EagerFlattenOpTransformer>(op_name);
     }
   }
@@ -232,7 +232,7 @@ EagerLayoutAutotune<paddle::experimental::Scalar>(
                                kSlotSmallVectorSize>& tensors_vector,
     paddle::experimental::Scalar* axis) {
   if (DesiredLayout() == phi::DataLayout::UNDEFINED) {
-    VLOG(4) << "Optimze Layout was not started" << op_name;
+    VLOG(4) << "Optimize Layout was not started" << op_name;
     return std::make_shared<EagerLayoutTransformer>(
         op_name, tensors_vector, tensors_vector[0][0].layout());
   }
diff --git a/paddle/fluid/eager/eager_layout_transformer.h b/paddle/fluid/eager/eager_layout_transformer.h
@@ -129,7 +129,7 @@ class EagerLayoutTransformer {
     dim_size_ = in.shape().size();
     bool need_trans =
         !(final_layout_ == Layout::UNDEFINED || final_layout_ == in.layout());
-    // This is for Agnostic op when layout is differnet
+    // This is for Agnostic op when layout is different
     if (need_trans) {
       auto out_tensor = EagerTraceTransposeOp(final_layout_, in);
       phi::DenseTensorUtils::GetMutableMeta(
diff --git a/paddle/fluid/eager/eager_tensor.h b/paddle/fluid/eager/eager_tensor.h
@@ -199,7 +199,7 @@ inline bool IsVariableCompatTensor(const paddle::Tensor& tensor) {
  * **/
 class EagerVariable final {
  public:
-  /* Default constructor and name constructor should only be used for contruct
+  /* Default constructor and name constructor should only be used for construct
    * output and in fluid*/
   EagerVariable() = default;
 
@@ -293,7 +293,7 @@ class EagerVariable final {
         true,
         paddle::platform::errors::Fatal(
             "Tensor %s does not hold phi::SelectedRows or phi::DenseTensor. "
-            "Or it holds empty impl, this should not happend since we should "
+            "Or it holds empty impl, this should not happened since we should "
             "treat all kinds of tensor as what they are.",
             tensor.name()));
     *framework_tensor = *tensor_dense;
@@ -314,7 +314,7 @@ class EagerVariable final {
     PADDLE_ENFORCE_NOT_NULL(compat_tensor,
                             paddle::platform::errors::Fatal(
                                 "Tensor %s holds empty impl, this should not "
-                                "happend since we should "
+                                "happened since we should "
                                 "treat all kinds of tensor as what they are.",
                                 tensor.name()));
     *framework_holder = compat_tensor->Get<VarType>();
diff --git a/paddle/fluid/eager/grad_node_info.h b/paddle/fluid/eager/grad_node_info.h
@@ -293,14 +293,14 @@ class GradNodeBase {
 
   std::map<int64_t, std::tuple<size_t, size_t, std::shared_ptr<TensorHook>>>
   GetGradientHookFuntions() {
-    VLOG(7) << "GetGradientHookFuntions ";
+    VLOG(7) << "GetGradientHookFunctions ";
     return gradient_hooks_;
   }
 
   void SetGradientHookFuntions(
       std::map<int64_t, std::tuple<size_t, size_t, std::shared_ptr<TensorHook>>>
           hooks) {
-    VLOG(7) << "SetGradientHookFuntions ";
+    VLOG(7) << "SetGradientHookFunctions ";
     gradient_hooks_ = hooks;
   }
 
diff --git a/paddle/fluid/eager/to_static/run_program_op_node.h b/paddle/fluid/eager/to_static/run_program_op_node.h
@@ -553,7 +553,7 @@ inline void PirRunProgramAPI(
     // program_id, global_inner_scope, false, skip_eager_delete_vars);
   } else {
     paddle::platform::RecordEvent record_event(
-        "get_interpretercore_cahce",
+        "get_interpretercore_cache",
         paddle::platform::TracerEventType::UserDefined,
         1);
     VLOG(2) << "Get interpretercore cache by program:" << program_id;
@@ -710,7 +710,7 @@ inline void RunProgramAPI(
         "create_new_interpretercore",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "No interpretercore cahce, so create a new interpretercore "
+    VLOG(2) << "No interpretercore cache, so create a new interpretercore "
                "for program: "
             << program_id;
     // Step 1. share input_vars & parameters into scope
@@ -781,10 +781,10 @@ inline void RunProgramAPI(
     VLOG(2) << "Get skip GC vars size is: " << skip_eager_delete_vars.size();
   } else {
     paddle::platform::RecordEvent record_event(
-        "get_interpretercore_cahce",
+        "get_interpretercore_cache",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "Get interpretercore cahce by program:" << program_id;
+    VLOG(2) << "Get interpretercore cache by program:" << program_id;
     // Step 1. get cache interpretercore
     auto &cached_value =
         interpretercore_info_cache.GetMutable(program_id,
@@ -889,7 +889,7 @@ inline void RunProgramGradAPI(
         "create_new_interpretercore",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "No interpretercore cahce, so create a new interpretercore"
+    VLOG(2) << "No interpretercore cache, so create a new interpretercore"
                "for program: "
             << program_id;
     details::ShareTensorsIntoScope(out_grad, global_inner_scope);
@@ -968,10 +968,10 @@ inline void RunProgramGradAPI(
     VLOG(2) << "Get skip GC vars size is: " << skip_eager_delete_vars.size();
   } else {
     paddle::platform::RecordEvent record_event(
-        "get_interpretercore_cahce",
+        "get_interpretercore_cache",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "Get interpretercore cahce by program:" << program_id;
+    VLOG(2) << "Get interpretercore cache by program:" << program_id;
     auto &cached_value =
         interpretercore_info_cache.GetMutable(program_id,
                                               global_inner_scope,
@@ -1095,7 +1095,7 @@ inline void PirRunProgramGradAPI(
         "create_new_interpretercore",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "No interpretercore cahce, so create a new interpretercore";
+    VLOG(2) << "No interpretercore cache, so create a new interpretercore";
     // Step 1. share input_vars & parameters into scope
     auto passed_kernel_program =
         paddle::framework::ApplyIrPass(backward_program, place);
@@ -1152,10 +1152,10 @@ inline void PirRunProgramGradAPI(
     details::print_collection(skip_eager_delete_vars);
   } else {
     paddle::platform::RecordEvent record_event(
-        "get_interpretercore_cahce",
+        "get_interpretercore_cache",
         paddle::platform::TracerEventType::UserDefined,
         1);
-    VLOG(2) << "Get interpretercore cahce by program:" << program_id;
+    VLOG(2) << "Get interpretercore cache by program:" << program_id;
     auto &cached_value =
         interpretercore_info_cache.GetMutable(program_id,
                                               global_inner_scope,
diff --git a/paddle/fluid/eager/utils.cc b/paddle/fluid/eager/utils.cc
@@ -133,7 +133,7 @@ std::vector<AutogradMeta*> EagerUtils::autograd_meta(
   std::vector<AutogradMeta*> ret;
   ret.reserve(targets->size());
 
-  // for autograd_meta we can tolerent it has nullptr.
+  // for autograd_meta we can tolerate it has nullptr.
   for (auto& target : *targets) {
     auto* p_autograd_meta = autograd_meta(&target);
     ret.emplace_back(p_autograd_meta);
@@ -146,7 +146,7 @@ std::vector<AutogradMeta*> EagerUtils::autograd_meta(
   std::vector<AutogradMeta*> ret;
   ret.reserve(targets->size());
 
-  // for autograd_meta we can tolerent it has nullptr.
+  // for autograd_meta we can tolerate it has nullptr.
   for (auto& target : *targets) {
     auto* p_autograd_meta = autograd_meta(target);
     ret.emplace_back(p_autograd_meta);
@@ -506,7 +506,7 @@ std::shared_ptr<egr::GradNodeBase> EagerUtils::GetGradAccumulationNode(
         PADDLE_THROW(paddle::platform::errors::Fatal(
             "GetGradAccumulationNode should only be called on leaf tensor, but "
             "target tensor: %s has GradNode which is not a "
-            "GradNodeAccumulation, and this should not happend unless target "
+            "GradNodeAccumulation, and this should not happened unless target "
             "tensor is modified by some ops and calling set history for it.",
             tensor.name()));
       }

Original file line number	Diff line number	Diff line change
`@@ -45,7 +45,7 @@ using AbstractAutogradMeta = paddle::AbstractAutogradMeta;`
`45`	`45`	`*`
`46`	`46`	`* 4. overrided_stop_gradient_`
`47`	`47`	`* This member is used to finish some auto-prune related work, which indicate`
`48`		`- * user set stop_gradient should overrided the result indicated by framework.`
	`48`	`+ * user set stop_gradient should override the result indicated by framework.`
`49`	`49`	`* All non-parameter tensor's stop_gradient properties should be true. We will`
`50`	`50`	`* pass stop_gradient when we find one who need it.`
`51`	`51`	`*`
Original file line number	Diff line number	Diff line change
`@@ -293,14 +293,14 @@ class GradNodeBase {`
`293`	`293`
`294`	`294`	`std::map<int64_t, std::tuple<size_t, size_t, std::shared_ptr<TensorHook>>>`
`295`	`295`	`GetGradientHookFuntions() {`
`296`		`- VLOG(7) << "GetGradientHookFuntions ";`
	`296`	`+ VLOG(7) << "GetGradientHookFunctions ";`
`297`	`297`	`return gradient_hooks_;`
`298`	`298`	`}`
`299`	`299`
`300`	`300`	`void SetGradientHookFuntions(`
`301`	`301`	`std::map<int64_t, std::tuple<size_t, size_t, std::shared_ptr<TensorHook>>>`
`302`	`302`	`hooks) {`
`303`		`- VLOG(7) << "SetGradientHookFuntions ";`
	`303`	`+ VLOG(7) << "SetGradientHookFunctions ";`
`304`	`304`	`gradient_hooks_ = hooks;`
`305`	`305`	`}`
`306`	`306`