-
Notifications
You must be signed in to change notification settings - Fork 21.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Compilation of <torch/extension.h> error on Windows CUDA 11.5 #69460
Comments
Adding this to the 1.11.0 milestone since this will be a blocker for the 1.11.0 release |
Grabbing for myself, as it probably involves tweaking pybind a bit |
I can reproduce it with hello world example that simply includes pybind, which looks like a CUDA compiler bug (happens during the invocation of cicc):
For debugging: hello.cpp1.ii |
Further narrow down: #include <utility>
// Base implementation for std::tuple and std::pair
template <template<typename...> class Tuple, typename... Ts> class tuple_caster {
using type = Tuple<Ts...>;
};
template <typename T1, typename T2> class type_caster
: public tuple_caster<std::pair, T1, T2> {};
__global__ void kernel() {
printf("Hello World");
}
int main(void) {
kernel <<<1, 1>>> ();
return cudaDeviceSynchronize();
} |
Need to document the regression of nvcc compiler from CUDA-11.5 on Windows |
@malfet @atalman |
My system is window 10+CUDA11.3. My problem is that when I use "torch. utils.cpp_extension.load_inline" to built a C++/CUDA extension, it shows: "ninja: build stopped: subcommand failed." |
Any workarounds? |
Looks like this issue is resolved in 11.7. I am not observing the failure anymore on #85966 |
Reenable aot tests on windows for cuda 11.7 and up Issue: #69460 seems to be mitigated in CUDA 11.7 hence re-enable this test cc @peterjc123 @mszhanyi @skyline75489 @nbcsm Pull Request resolved: #87193 Approved by: https://github.com/malfet
Reenable aot tests on windows for cuda 11.7 and up Issue: pytorch#69460 seems to be mitigated in CUDA 11.7 hence re-enable this test cc @peterjc123 @mszhanyi @skyline75489 @nbcsm Pull Request resolved: pytorch#87193 Approved by: https://github.com/malfet
Reenable aot tests on windows for cuda 11.7 and up Issue: #69460 seems to be mitigated in CUDA 11.7 hence re-enable this test cc @peterjc123 @mszhanyi @skyline75489 @nbcsm Pull Request resolved: #87193 Approved by: https://github.com/malfet
Removing from milestones, since this issue is not critical for release 1.13 and this is resolved for CUDA 11.7 |
Reenable aot tests on windows for cuda 11.7 and up Issue: pytorch#69460 seems to be mitigated in CUDA 11.7 hence re-enable this test cc @peterjc123 @mszhanyi @skyline75489 @nbcsm Pull Request resolved: pytorch#87193 Approved by: https://github.com/malfet
As CUDA-11.5 is no longer supported, just remove the check Fixes #69460
With deprecation of CUDA 11.6, we can resolve this issue, will post a PR once 11.6 is deprecated from CI |
As CUDA-11.5 is no longer supported, just remove the check Fixes #69460
We have following error when compiling CUDA 11.5 on windows
C:\actions-runner_work\pytorch\pytorch\build\win_tmp\build\torch\include\pybind11\cast.h(1429): error: too few arguments for temp
late template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]
"
(1507): here
C:\actions-runner_work\pytorch\pytorch\build\win_tmp\build\torch\include\pybind11\cast.h(1503): error: too few arguments for temp
late template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]
"
(1507): here
Complete failure log:
https://github.com/pytorch/pytorch/runs/4408796098?check_suite_focus=true
This looks like the same issue as this one:
facebookresearch/pytorch3d#843
Here is the workaround for this issue:
facebookresearch/pytorch3d@cb170ac
cc @ezyang @gchanan @zou3519 @peterjc123 @mszhanyi @skyline75489 @nbcsm @brianjo @mruberry @ngimel @bdhirsh @jbschlosser @malfet @seemethere @pytorch/pytorch-dev-infra
The text was updated successfully, but these errors were encountered: