Skip to content

[pytorch/cuda] apply 16-bit mask to the index for device guard registry #45485

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.

Sign up for GitHub

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

Closed
wants to merge 1 commit into from
Closed

[pytorch/cuda] apply 16-bit mask to the index for device guard registry #45485

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions c10/core/impl/DeviceGuardImplInterface.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -209,7 +209,18 @@ class C10_API DeviceGuardImplRegistrar {
static ::c10::impl::DeviceGuardImplRegistrar C10_ANONYMOUS_VARIABLE(g_##DeviceType)(::c10::DeviceType::DevType, new DeviceGuardImpl());

inline const DeviceGuardImplInterface* getDeviceGuardImpl(DeviceType type) {
#if defined(__CUDACC__)
// Two adjacent int16_t fields DeviceType and DeviceIndex has field access
// miscompiled on NVCC. To workaround this issue, we apply a mask to the
// DeviceType. First check if the DeviceType is 16-bit.
// FB employees can see
// https://fb.workplace.com/groups/llvm.gcc/permalink/4053565044692080/
// for more details
static_assert(sizeof(DeviceType) == 2, "DeviceType is not 16-bit");
auto p = device_guard_impl_registry[static_cast<size_t>(type) & 0xFFFF].load();
#else
auto p = device_guard_impl_registry[static_cast<size_t>(type)].load();
#endif
// This seems to be the first place where you make use of a device
// when you pass devices to factory functions. Give a nicer error
// message in this case.
Expand Down