cl 命令行 无法打开 calibration_statistics_collector_average_min_max.obj.params about tensorflow HOT 10 OPEN

same issue cl : Command line error D8022 : cannot open 'bazel-out/x64_windows-opt/bin/tensorflow/compiler/mlir/quantization/tensorflow/calibrator/_objs/calibration_statistics_collector_average_min_max/calibration_statistics_collector_average_min_max.obj.params'
TF Version 2.15.1

from tensorflow.

Seems that the latest version of TF doesn't support MSVC 2019/MSVC2022 compiler any more, please use CLANG to Build.

from tensorflow.

@Venkat6871 do you have any update on this issue? I had turned on Developer mode, it is not access right problem. I checked the BUILD file under calibrator folder, no problem of this file. and calibration_statistics_collector_average_min_max.obj.params file had existed, why cl.exe cannot open this file. it is format issue or not?
BR

from tensorflow.

Hi @ShifengJin ,
Sorry for the delay, Could you verify that the version of the C compiler used by Bazel are compatible with TensorFlow. Sometimes, mismatched compiler versions can lead to build errors. And if you are using an older version of the C++ compiler, consider upgrading to a newer version that may have better memory management. Please delete any existing build files and directories to ensure a fresh start.

Thank you!

from tensorflow.

Microsoft (R) C/C++ Optimizing Compiler Version 19.39.33523 for x64
Copyright (C) Microsoft Corporation. All rights reserved.
it should be latest version of MSVC

from tensorflow.

MSVC\14.39.33519

from tensorflow.

I'm also facing exact error and it is not actually bind to this particular calibration_statistics_collector_average_min_max.obj.params file, but it occurs for some other *.params file as well.

Faced this while trying to build the following target,
bazel build --config=opt --cxxopts="-std=c++17" //tensorflow:tensorflow_cc.dll

from tensorflow.

that's a problem why we cannot use c++17 to build, i used win-clang, it can be build successfully. actually, this file is existed on the bazel_out folder already, if you run cl.exe manually, this file can be opened. I suspected it related with build setup. but we had followed the instruction in release note. I also met another problem of smallvector like below:
tensorflow/dtensor/mlir/expansions/save_restore_spmd_expander.cc(812): error C2976: 'llvm::SmallVector': too few template arguments. I checked the code, find that for TFV2.15.0 they use llvm17.0.6. (I want to use the latest llvm with --override_repository=llvm-raw= failed also) and in save_restore_spmd_expander.cc:812
StatusOr<llvm::SmallVector> GetLayoutsFromAssignVariableOps
I am mad with TF c++ windows build, I gave up windows build , and works on linux environment now.

from tensorflow.

@Venkat6871 please don't wait the his response, we met the same issue, I can provided the info you wanted

from tensorflow.

As @weijianghn said, I do not get any errors when compiling in LLVM/clang 17.0.6. Getting it only in MSVC. But I'm facing other error because of missing symbols while trying to link the tensorflow.dll. The problem is I could solve missing symbols by simply adding TF_EXPORT in the respective source and recompile, but the missing symbol is from a dynamically generated proto file C++ header stub.

from tensorflow.