Occur fatal error 1002 when compiling tensorflow 1.3 with vs2015 DEBUG in windows10

[HannH]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow):
  no custom code(original code cloned from github )
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
  Windows 10
• TensorFlow installed from (source or binary):
  source(cmake from tensorflow/contribe/cmake)
• TensorFlow version (use command below):
  r1.3
• Visual studio version:
  vs2015
• CUDA/cuDNN version:
  no GPU/CUDA 8.0+cuDNN5.1
• GPU model and memory:
  NVIDIA Titan X(12GB)
• Exact command to reproduce:
  tf_cc.vcxproj -> A:\C++\tensorflow-1.3.0\Source_GPU\tf_cc.dir\Debug\tf_cc.lib
  65>a:\c++\tensorflow-1.3.0\source_gpu\external\eigen_archive\eigen\src\core\products\generalblockpanelkernel.h(1977): fatal error C1002:

Describe the problem

Hello, I get error 1002 when compiling tensorflow r1.3 with vs2015 in tf_core_kernels , use DEBUG mode without GPU, which occured during compilation of GPU version too. I belive I have enough memory for compilation. However, I compiled successfully through RELEASE mode without error.

Same issue in tensorflow-r1.2.

[reedwm]

Can you post the entire error output that occurs?

[snnn]

I believe debug build is not supported, especially for gpu build. You may try to replace /DEBUG:Full with /DEBUG:Fastlink in your compiler flags. And you may also need to split tf_core_kernerls into multiple static libraries. And you need a debug build of python,cuda,... If I were you, I'll give up

[HannH]

@reedwm the error output have just 2 sentences, which I have written above. I would post a part of build log for you
1> count_extremely_random_stats_op.cc
1> finished_nodes_op.cc
1> grow_tree_op.cc
1> reinterpret_string_to_float_op.cc
1> sample_inputs_op.cc
1> scatter_add_ndim_op.cc
1> tree_predictions_op.cc
1> tree_utils.cc
1> update_fertile_slots_op.cc
1> hard_routing_function_op.cc
1> k_feature_gradient_op.cc
1> k_feature_routing_function_op.cc
1> routing_function_op.cc
1> routing_gradient_op.cc
1> stochastic_hard_routing_function_op.cc
1> stochastic_hard_routing_gradient_op.cc
1> unpack_path_op.cc
1> utils.cc
1> skip_gram_kernels.cc
1> skip_gram_ops.cc
1> cross_replica_ops.cc
1> infeed_ops.cc
1> outfeed_ops.cc
1> replication_ops.cc
1> tpu_configuration_ops.cc
1> tpu_sendrecv_ops.cc
1>a:\c++\tensorflow-1.3.0\source_gpu\external\eigen_archive\eigen\src\core\products\generalblockpanelkernel.h(1989): fatal error C1002: 在第 2 遍中编译器的堆空间不足(compiler is out of heap space in pass 2)
1>cl : 命令行 error D8040: 创建子进程或与子进程通讯时出错(error creating or communicating with child process)
========== 生成: 成功(success) 0 个，失败 (failure)1 个，最新 0 个，跳过 0 个 ==========
sorry for using the chinese version, I have translated the output in english

[HannH]

@snnn I agree with you, there are some problem during compiling in Debug mode now. So I post the error for later development of tensorflow.
It seems like there have some problem in MinSizeRel and RelWithDebInfo mode, because i cannot compile tf in these modes either.

[snnn]

Did you use the 32 bits cl.exe or 64 bits cl.exe?

set PreferredToolArchitecture=x64

Run this command before open tensorflow.sln

Or, add "-T host=x64" to cmake command line args.

[HannH]

@snnn Thank you for your suggestion! Because compile is time-consuming, I would tell you the result later.
Anthor question: I found it is necessary to compile a tensorflow project statically in vs, which caused the result bloat. For example, project 'tf_label_image_example' is 164 MB in release mode, but there only one file named 'main.cc' from 'examples\label_image' in this project. Is there any way to compile it dynamiclly to squeeze the file size?

[reedwm]

/CC @mrry

[HannH]

@snnn OK, your suggestion works, thank you!

[HannH]

I found it is easy to build tensorflow project dynamically in vs 2015 for tf-r1.3. Close the issue.

[yuyijie1995]

@HannH I have the same problem , which solution is useful to you ? Is the "set PreferredToolArchitecture=x64"? I try to add "-T host=x64" to cmake command line args. but another error happened . Can I have your qq or wechat number?