Add call stack info during compile time #19067

chengduoZH · 2019-08-08T03:32:56Z

This PR's work includes:

Add call stack info during compile time.
Extract the common code of run time and compile time to operator_call_stack.

Before

Traceback (most recent call last):
  File "./train_gpu_paddle.py", line 471, in <module>
    train(n_token, cutoffs)
  File "./train_gpu_paddle.py", line 254, in train
    decr_ratio=FLAGS.decr_ratio)
  File "/ssd3/transformer-xl-paddlepaddle_tmp/optimization.py", line 151, in optimization
    _, param_grads = optimizer.minimize(loss)
  File "</usr/local/lib/python2.7/dist-packages/decorator.pyc:decorator-gen-20>", line 2, in minimize
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/wrapped_decorator.py", line 25, in __impl__
    return wrapped_func(*args, **kwargs)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/dygraph/base.py", line 86, in __impl__
    return func(*args, **kwargs)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/optimizer.py", line 594, in minimize
    no_grad_set=no_grad_set)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/optimizer.py", line 493, in backward
    no_grad_set, callbacks)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/backward.py", line 706, in append_backward
    _append_backward_vars_(root_block, fwd_op_num, grad_to_var, grad_info_map)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/backward.py", line 518, in _append_backward_vars_
    op_desc.infer_shape(block.desc)
paddle.fluid.core_avx.EnforceNotMet: Invoke operator elementwise_add_grad error.

Input(Out@GRAD) should not be null at [../paddle/fluid/operators/elementwise/elementwise_op.h:289]
PaddlePaddle Call Stacks:
0       0x7fb76db1b35ap void paddle::platform::EnforceNotMet::Init<char const*>(char const*, char const*, int) + 506
1       0x7fb76db1c065p paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int) + 165
2       0x7fb76f427ebap paddle::operators::ElementwiseOpExplicitGrad::InferShape(paddle::framework::InferShapeContext*) const + 1130
3       0x7fb76dcdf605p
4       0x7fb76dce3ed9p paddle::framework::OpDesc::InferShape(paddle::framework::BlockDesc const&) const + 985
5       0x7fb76dc00334p
6       0x7fb76db4eb14p
7             0x4c5326p PyEval_EvalFrameEx + 37958
8             0x4b9b66p PyEval_EvalCodeEx + 774
9             0x4c1f56p PyEval_EvalFrameEx + 24694
10            0x4b9b66p PyEval_EvalCodeEx + 774
11            0x4c17c6p PyEval_EvalFrameEx + 22758
12            0x4b9b66p PyEval_EvalCodeEx + 774
13            0x4c17c6p PyEval_EvalFrameEx + 22758
14            0x4b9b66p PyEval_EvalCodeEx + 774
15            0x4d57a3p
16            0x4a587ep PyObject_Call + 62
17            0x4be51ep PyEval_EvalFrameEx + 9790
18            0x4b9b66p PyEval_EvalCodeEx + 774
19            0x4d57a3p

Now:

Traceback (most recent call last):
  File "./train_gpu_paddle.py", line 471, in <module>
    train(n_token, cutoffs)
  File "./train_gpu_paddle.py", line 254, in train
    decr_ratio=FLAGS.decr_ratio)
  File "/ssd3/transformer-xl-paddlepaddle_tmp/optimization.py", line 151, in optimization
    _, param_grads = optimizer.minimize(loss)
  File "</usr/local/lib/python2.7/dist-packages/decorator.pyc:decorator-gen-20>", line 2, in minimize
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/wrapped_decorator.py", line 25, in __impl__
    return wrapped_func(*args, **kwargs)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/dygraph/base.py", line 86, in __impl__
    return func(*args, **kwargs)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/optimizer.py", line 594, in minimize
    no_grad_set=no_grad_set)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/optimizer.py", line 493, in backward
    no_grad_set, callbacks)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/backward.py", line 706, in append_backward
    _append_backward_vars_(root_block, fwd_op_num, grad_to_var, grad_info_map)
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/backward.py", line 518, in _append_backward_vars_
    op_desc.infer_shape(block.desc)
paddle.fluid.core_avx.EnforceNotMet: Invoke operator elementwise_add_grad error.

Python Callstacks:
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/framework.py", line 1776, in append_op
    attrs=kwargs.get("attrs", None))
  File "/ssd3/zcd_Paddle/build_fast/python/paddle/fluid/layers/math_op_patch.py", line 149, in __impl__
    attrs={'axis': axis})
  File "/ssd3/transformer-xl-paddlepaddle_tmp/model.py", line 371, in gather_nd
    index_join = raw_index * tensor.shape[1] + col_index
  File "/ssd3/transformer-xl-paddlepaddle_tmp/model.py", line 403, in _gather_logprob
    return gather_nd(logprob, idx)
  File "/ssd3/transformer-xl-paddlepaddle_tmp/model.py", line 494, in mask_adaptive_logsoftmax
    _gather_logprob(tail_logprob, cur_target))
  File "/ssd3/transformer-xl-paddlepaddle_tmp/model.py", line 558, in transformer
    proj_same_dim=proj_same_dim)
  File "/ssd3/transformer-xl-paddlepaddle_tmp/transformer_xl.py", line 200, in _build_model
    name='encoder'
  File "./train_gpu_paddle.py", line 84, in model_fn
    loss, new_mems = transofrmer_xl._build_model()
  File "./train_gpu_paddle.py", line 99, in single_core_graph
    is_training=is_training)
  File "./train_gpu_paddle.py", line 128, in create_model
    mems=mems)
  File "./train_gpu_paddle.py", line 238, in train
    batch_size=bsz_per_trainner, n_token=n_token, cutoffs=cutoffs)
  File "./train_gpu_paddle.py", line 471, in <module>
    train(n_token, cutoffs)
C++ Callstacks:
Input(Out@GRAD) should not be null at [../paddle/fluid/operators/elementwise/elementwise_op.h:289]
PaddlePaddle Call Stacks:
0       0x7fb76db1b35ap void paddle::platform::EnforceNotMet::Init<char const*>(char const*, char const*, int) + 506
1       0x7fb76db1c065p paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int) + 165
2       0x7fb76f427ebap paddle::operators::ElementwiseOpExplicitGrad::InferShape(paddle::framework::InferShapeContext*) const + 1130
3       0x7fb76dcdf605p
4       0x7fb76dce3ed9p paddle::framework::OpDesc::InferShape(paddle::framework::BlockDesc const&) const + 985
5       0x7fb76dc00334p
6       0x7fb76db4eb14p
7             0x4c5326p PyEval_EvalFrameEx + 37958
8             0x4b9b66p PyEval_EvalCodeEx + 774
9             0x4c1f56p PyEval_EvalFrameEx + 24694
10            0x4b9b66p PyEval_EvalCodeEx + 774
11            0x4c17c6p PyEval_EvalFrameEx + 22758
12            0x4b9b66p PyEval_EvalCodeEx + 774

test=develop

chenwhql

just my thinking, maybe the Traceback info and C++ callstacks info is useless for most users, because they just use our platform, they will not modify our source code, they just need to know where their own mistakes are. The error message is too long now.

chenwhql · 2019-08-08T10:45:10Z

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

+
+
+class TestCompileTimeException(OpTest):
+    def test_run_time_exception(self):


maybe test_compile_time_exception is better?

test=develop

chengduoZH · 2019-08-08T12:42:21Z

just my thinking, maybe the Traceback info and C++ callstacks info is useless for most users, because they just use our platform, they will not modify our source code, they just need to know where their own mistakes are. The error message is too long now.

C++ Callstack info is really long, but some error info of C++ Callstack is useful for users and developers.

chenwhql

LGTM

Add call stack info during runtime and compile time

f5590b1

test=develop

chengduoZH force-pushed the add_error_info_during_compile branch from 11f2c87 to f5590b1 Compare August 8, 2019 03:36

chengduoZH requested a review from Xreki August 8, 2019 03:41

Rename operator_call_stack

9a8997f

test=develop

chengduoZH force-pushed the add_error_info_during_compile branch from 10d56aa to 9a8997f Compare August 8, 2019 06:05

luotao1 added the Error Message label Aug 8, 2019

Add unit test

efab284

test=develop

chengduoZH requested a review from chenwhql August 8, 2019 09:04

chenwhql reviewed Aug 8, 2019

View reviewed changes

follow comment

142d30c

test=develop

chenwhql approved these changes Aug 8, 2019

View reviewed changes

chengduoZH merged commit 21440b4 into PaddlePaddle:develop Aug 8, 2019

chengduoZH mentioned this pull request Aug 20, 2019

[Cherry Pick] Add error info during compile #19300

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add call stack info during compile time #19067

Add call stack info during compile time #19067

chengduoZH commented Aug 8, 2019 •

edited

Loading

chenwhql left a comment

chenwhql Aug 8, 2019

chengduoZH Aug 8, 2019

chengduoZH commented Aug 8, 2019

chenwhql left a comment



		class TestCompileTimeException(OpTest):
		def test_run_time_exception(self):

Add call stack info during compile time #19067

Add call stack info during compile time #19067

Conversation

chengduoZH commented Aug 8, 2019 • edited Loading

chenwhql left a comment

Choose a reason for hiding this comment

chenwhql Aug 8, 2019

Choose a reason for hiding this comment

chengduoZH Aug 8, 2019

Choose a reason for hiding this comment

chengduoZH commented Aug 8, 2019

chenwhql left a comment

Choose a reason for hiding this comment

chengduoZH commented Aug 8, 2019 •

edited

Loading