[CUDA] Drop CUDA 10 support #89582
[CUDA] Drop CUDA 10 support #89582
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/89582
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 FailuresAs of commit 39a85f4: FLAKY - The following jobs failed but were likely due to flakiness present on master:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
There's some legacy code wrapped with the pytorch/caffe2/utils/math_gpu.cu Line 725 in 41c3b41 |
aten/src/ATen/Dispatch.h
Outdated
| }) | ||
| #endif | ||
|
|
||
| // Workaround for C10_UNUSED because CUDA 10.2 and below fails to handle unused |
aten/src/ATen/Dispatch.h
Outdated
| #if defined(__CUDACC__) && CUDA_VERSION < 11000 | ||
| #define C10_UNUSED_DISPATCH_CUDA_WORKAROUND | ||
| #else | ||
| #define C10_UNUSED_DISPATCH_CUDA_WORKAROUND C10_UNUSED |
There was a problem hiding this comment.
probably all C10_UNUSED_DISPATCH... can be replaced by C10_UNUSED now?
There was a problem hiding this comment.
Nit: can we just replace C10_UNUSED_DISPATCH_CUDA_WORKAROUND with C10_UNUSED everywhere?
| template <typename T> | ||
| C10_HOST_DEVICE constexpr thrust::complex<T> | ||
| cuda101bug_cast_c10_complex_to_thrust_complex(const c10::complex<T>& x) { | ||
| #if defined(CUDA_VERSION) && (CUDA_VERSION < 10020) |
There was a problem hiding this comment.
similarly here, cuda101bug... uses should just be replaced with static_cast, as the comment suggests
|
Hmm, what macro should be used instead of |
|
CC @jithunnair-amd who might know more about the last issue |
|
I've left the version guards in the |
eqy
added
the
ciflow/periodic
aten/src/ATen/Dispatch.h
Outdated
| #if defined(__CUDACC__) && CUDA_VERSION < 11000 | ||
| #define C10_UNUSED_DISPATCH_CUDA_WORKAROUND | ||
| #else | ||
| #define C10_UNUSED_DISPATCH_CUDA_WORKAROUND C10_UNUSED |
There was a problem hiding this comment.
Nit: can we just replace C10_UNUSED_DISPATCH_CUDA_WORKAROUND with C10_UNUSED everywhere?
|
@pytorchmergebot merge -f "ROCM failure appears unrelated" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Follow-up of #89582 to drop flags like `CUDA11OrLater` in tests. Note that in some places it appears that `TEST_WITH_ROCM` is _implicitly_ guarded against via the `CUDA11OrLater` version check, based on my best-guess of how `torch.version.cuda` would behave in ROCM builds, so I've added `not TEST_WITH_ROCM` in cases where ROCM wasn't previously explicitly allowed. CC @ptrblck @malfet @ngimel Pull Request resolved: #92605 Approved by: https://github.com/ngimel
5 participants
CC @ptrblck @ngimel @malfet