Merge pull request #15204 from wopeizl/debug/support

add the python callstack for debug support test=develop

paddle/fluid/framework/op_proto_maker.cc

@@ -82,6 +82,10 @@ void OpProtoAndCheckerMaker::operator()(proto::OpProto* proto,
   AddAttr<std::string>(OpNamescopeAttrName(), "Operator name with namesope.")
       .SetDefault("");
 
+  AddAttr<std::vector<std::string>>(OpCreationCallstackAttrName(),
+                                    "Callstack for Op Creatation.")
+      .SetDefault({});
+
   Validate();
 }
 

paddle/fluid/framework/op_proto_maker.h

@@ -47,6 +47,7 @@ class OpProtoAndCheckerMaker {
   static const char *OpRoleAttrName() { return "op_role"; }
   static const char *OpRoleVarAttrName() { return "op_role_var"; }
   static const char *OpNamescopeAttrName() { return "op_namescope"; }
+  static const char *OpCreationCallstackAttrName() { return "op_callstack"; }
 
   void operator()(proto::OpProto *proto, OpAttrChecker *attr_checker);
 

paddle/fluid/framework/operator.cc

@@ -16,9 +16,15 @@ limitations under the License. */
 #include <glog/logging.h>
 
 #include <algorithm>
+#include <sstream>
+#include <string>
+#include <vector>
+#include "gflags/gflags.h"
+#include "glog/logging.h"
 #include "paddle/fluid/framework/data_transform.h"
 #include "paddle/fluid/framework/executor.h"
 #include "paddle/fluid/framework/lod_tensor.h"
+#include "paddle/fluid/framework/op_proto_maker.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/shape_inference.h"
 #include "paddle/fluid/framework/transfer_scope_cache.h"
@@ -156,27 +162,55 @@ RuntimeContext::RuntimeContext(const VariableNameMap& innames,
 }
 
 void OperatorBase::Run(const Scope& scope, const platform::Place& place) {
-  VLOG(4) << place << " " << DebugStringEx(&scope);
-  if (platform::is_gpu_place(place)) {
+  try {
+    VLOG(4) << place << " " << DebugStringEx(&scope);
+    if (platform::is_gpu_place(place)) {
 #ifndef PADDLE_WITH_CUDA
-    PADDLE_THROW("Cannot run operator on place %s", place);
+      PADDLE_THROW("Cannot run operator on place %s", place);
 #else
-    auto dev_id = boost::get<platform::CUDAPlace>(place).device;
-    platform::SetDeviceId(dev_id);
+      auto dev_id = boost::get<platform::CUDAPlace>(place).device;
+      platform::SetDeviceId(dev_id);
 #endif
-  }
+    }
 
-  // The profile has a process-wide mutex, results in serious performance issue
-  // in concurrency scenerio. Here use an `if` to fix this issue.
-  // Please not remove the `if`, ask @Superjomn if there are any concern.
-  if (platform::IsProfileEnabled()) {
-    platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
-    platform::RecordEvent record_event(Type(), pool.Get(place));
-    RunImpl(scope, place);
-  } else {
-    RunImpl(scope, place);
+    // The profile has a process-wide mutex, results in serious performance
+    // issue
+    // in concurrency scenerio. Here use an `if` to fix this issue.
+    // Please not remove the `if`, ask @Superjomn if there are any concern.
+    if (platform::IsProfileEnabled()) {
+      platform::DeviceContextPool& pool =
+          platform::DeviceContextPool::Instance();
+      platform::RecordEvent record_event(Type(), pool.Get(place));
+      RunImpl(scope, place);
+    } else {
+      RunImpl(scope, place);
+    }
+
+    VLOG(3) << place << " " << DebugStringEx(&scope);
+  } catch (platform::EnforceNotMet exception) {
+    if (Attrs().count("sub_block") != 0) {
+      throw exception;
+    }
+
+    auto& callstack = Attr<std::vector<std::string>>(
+        OpProtoAndCheckerMaker::OpCreationCallstackAttrName());
+
+    if (callstack.empty()) {
+      throw exception;
+    }
+    std::ostringstream sout;
+    sout << "Invoke operator " << Type() << " error.\n";
+    sout << "Python Callstacks: \n";
+    for (auto& line : callstack) {
+      sout << line;
+    }
+    sout << "C++ Callstacks: \n";
+    sout << exception.err_str_;
+    exception.err_str_ = sout.str();
+    throw exception;
+  } catch (...) {
+    std::rethrow_exception(std::current_exception());
   }
-  VLOG(3) << place << " " << DebugStringEx(&scope);
 }
 
 bool OperatorBase::HasInputs(const std::string& name) const {

paddle/fluid/platform/enforce.h

@@ -263,6 +263,10 @@ inline void throw_on_error(T e) {
 #define __THROW_ON_ERROR_ONE_ARG(COND, ARG) \
   ::paddle::platform::throw_on_error(COND, ::paddle::string::Sprintf(ARG));
 
+#ifdef _WIN32
+#define __PADDLE_THROW_ON_ERROR(COND, ...) \
+  __THROW_ON_ERROR_ONE_ARG(COND, __VA_ARGS__)
+#else  // _WIN32
 #define __PADDLE_THROW_ON_ERROR(COND, ...)                                \
   __PADDLE_THROW_ERROR_I(                                                 \
       __VA_ARGS__, ::paddle::platform::throw_on_error(COND, __VA_ARGS__), \
@@ -274,6 +278,7 @@ inline void throw_on_error(T e) {
       ::paddle::platform::throw_on_error(COND, __VA_ARGS__),              \
       ::paddle::platform::throw_on_error(COND, __VA_ARGS__),              \
       __THROW_ON_ERROR_ONE_ARG(COND, __VA_ARGS__))
+#endif  // _WIN32
 
 #define __PADDLE_UNARY_COMPARE(COND, ...)                 \
   do {                                                    \

paddle/fluid/pybind/const_value.cc

@@ -49,6 +49,9 @@ void BindConstValue(pybind11::module* m) {
   op_proto_and_checker_maker.def(
       "kOpNameScopeAttrName",
       framework::OpProtoAndCheckerMaker::OpNamescopeAttrName);
+  op_proto_and_checker_maker.def(
+      "kOpCreationCallstackAttrName",
+      framework::OpProtoAndCheckerMaker::OpCreationCallstackAttrName);
 }
 
 }  // namespace pybind

python/paddle/fluid/framework.py

@@ -19,6 +19,7 @@ from collections import defaultdict
 import contextlib
 import os
 import re
+import traceback
 import six
 
 import numpy as np
@@ -625,6 +626,11 @@ class Operator(object):
         if type is None:
             raise ValueError(
                 "`type` to initilized an Operator can not be None.")
+        else:
+            callstack_var_name = op_maker.kOpCreationCallstackAttrName()
+            op_attrs[callstack_var_name] = list(
+                reversed(traceback.format_stack()))[1:]
+
         self.desc.set_type(type)
         proto = OpProtoHolder.instance().get_op_proto(type)
 

python/paddle/fluid/tests/unittests/test_operator_desc.py

@@ -69,7 +69,7 @@ class TestOperator(unittest.TestCase):
             set(mul_op.attr_names),
             set([
                 "x_num_col_dims", "y_num_col_dims", "op_role", "op_role_var",
-                "op_namescope"
+                "op_namescope", "op_callstack"
             ]))
         self.assertEqual(mul_op.has_attr("x_num_col_dims"), True)
         self.assertEqual(mul_op.attr_type("x_num_col_dims"), core.AttrType.INT)