Sageattention2 fails to build on torch nightly 2.7.0 (8th-26th-27th-28th feb), because ambiguous symbol from torch dynamo (Windows)

[Panchovix]

Hi there, thanks for the great work.

I did update torch to latest nightly of 8th Feb, but got

FAILED: G:/Stable difussion/stable-diffusion-webui-reForge/SageAttention/build/temp.win-amd64-cpython-312/Release/csrc/qattn/qk_int_sv_f16_cuda_sm80.obj

....

G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(962): error C2872: 'std': ambiguous symbol
C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.42.34433/include\valarray(20): note: could be 'std'
G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(962): note: or       'std'
G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(962): note: the template instantiation context (the oldest one first) is
G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(1007): note: see reference to class template instantiation 'torch::dynamo::autograd::IValuePacker<__int64>' being compiled
G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(945): note: while compiling class template member function 'c10::TypePtr torch::dynamo::autograd::IValuePacker<__int64>::packed_type(void)'
G:/Stable difussion/stable-diffusion-webui-reForge/venv/Lib/site-packages/torch/include\torch/csrc/dynamo/compiled_autograd.h(1007): note: see the first reference to 'torch::dynamo::autograd::IValuePacker<__int64>::packed_type' in 'torch::dynamo::autograd::IValuePacker<unsigned __int64>::packed_type'
  File "C:\Program Files\Python312\Lib\subprocess.py", line 571, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['ninja', '-v']' returned non-zero exit status 1.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "G:\Stable difussion\stable-diffusion-webui-reForge\SageAttention\setup.py", line 202, in <module>
    setup(
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\__init__.py", line 104, in setup
    return distutils.core.setup(**attrs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\core.py", line 184, in setup
    return run_commands(dist)
           ^^^^^^^^^^^^^^^^^^
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\core.py", line 200, in run_commands
    dist.run_commands()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\dist.py", line 969, in run_commands
    self.run_command(cmd)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\dist.py", line 967, in run_command
    super().run_command(command)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\dist.py", line 988, in run_command
    cmd_obj.run()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\wheel\_bdist_wheel.py", line 387, in run
    self.run_command("build")
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\cmd.py", line 316, in run_command
    self.distribution.run_command(command)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\dist.py", line 967, in run_command
    super().run_command(command)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\dist.py", line 988, in run_command
    cmd_obj.run()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\command\build.py", line 132, in run
    self.run_command(cmd_name)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\cmd.py", line 316, in run_command
    self.distribution.run_command(command)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\dist.py", line 967, in run_command
    super().run_command(command)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\dist.py", line 988, in run_command
    cmd_obj.run()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\command\build_ext.py", line 91, in run
    _build_ext.run(self)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\command\build_ext.py", line 359, in run
    self.build_extensions()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\torch\utils\cpp_extension.py", line 921, in build_extensions
    build_ext.build_extensions(self)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\command\build_ext.py", line 479, in build_extensions
    self._build_extensions_serial()
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\command\build_ext.py", line 505, in _build_extensions_serial
    self.build_extension(ext)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\command\build_ext.py", line 252, in build_extension
    _build_ext.build_extension(self, ext)
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\setuptools\_distutils\command\build_ext.py", line 560, in build_extension
    objects = self.compiler.compile(
              ^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\torch\utils\cpp_extension.py", line 893, in win_wrap_ninja_compile
    _write_ninja_file_and_compile_objects(
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\torch\utils\cpp_extension.py", line 1899, in _write_ninja_file_and_compile_objects
    _run_ninja_build(
  File "G:\Stable difussion\stable-diffusion-webui-reForge\venv\Lib\site-packages\torch\utils\cpp_extension.py", line 2256, in _run_ninja_build
    raise RuntimeError(message) from e
RuntimeError: Error compiling objects for extension

Mostly to catch it now.

Using nightly of 1 week ago (2025-02-01) works fine. Probably some related dynamo changes affected this.

Thanks in advance!

[jason-huang03]

can it work on ealier stable release of pytorch like torch 2.5.1?

[Panchovix]

It works fine on torch 2.5.1 or 2.6.0.

But for the future stable 2.7.0, not sure how It will behave. Could also report it on the pytorch repo.

[jason-huang03]

@Panchovix It's weird. People have successfully build SageAttention with torch 2.7 on RTX5090. What GPU do you use?

[Panchovix]

Tried today again (after getting a RTX 5090) and still fails on windows nightly.

Using torch 2.6.0 from https://huggingface.co/w-e-w/torch-2.6.0-cu128.nv/tree/main works.

[aoleg]

Fails to build on 2.7 nightly as well. Same error, "error C2872: 'std': ambiguous symbol"

[maxious]

pytorch/pytorch#144707 is probably the problematic change.

Commenting out these lines (Around line 1097 on my nightly checkout) in your torch in python site-packages:

//    } else if constexpr (::std::is_same_v<T, ::std::string>) {
//      return at::StringType::get();

Did allow sageattention to compile and run.

Obviously if you had some code that did want to do gradients over strings/text you'd need to fix that properly 😂 (and upstream pytorch should fix this, it's nothing to do with sageattention specifically)

[Panchovix]

pytorch/pytorch#144707 is probably the problematic change.

Commenting out these lines (Around line 1097 on my nightly checkout) in your torch in python site-packages:

//    } else if constexpr (::std::is_same_v<T, ::std::string>) {
//      return at::StringType::get();

Did allow sageattention to compile and run.

Obviously if you had some code that did want to do gradients over strings/text you'd need to fix that properly 😂 (and upstream pytorch should fix this, it's nothing to do with sageattention specifically)

Many thanks, it was that! Commented on the PR

[green-s]

I got this same error when I tried to build a few days back but tried again with the latest nightly with $env:CL='/permissive-' set prior and it built successfully. Not sure if it was that or just that the error has been fixed since I last tried because I didn't try to build prior but I found mention of that flag fixing the same error elsewhere.