[MPS][BE] Delete unused lerp functors #152443
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/152443
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 32 PendingAs of commit da6a11c with merge base d35e900 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot merge -f "Lint + MPS are green" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
@pytorchmergebot revert -m 'failing MPS test: test/test_optim.py::TestOptimRenewedMPS::test_can_load_from_to_named_state_dict_is_named_optim0_False_is_named_optim1_False_Adafactor_mps_float32' -c nosignal test_optim.py::TestOptimRenewedMPS::test_can_load_from_to_named_state_dict_is_named_optim0_False_is_named_optim1_False_Adafactor_mps_float32 GH job link HUD commit link |
|
cc @malfet |
|
@pytorchbot successfully started a revert job. Check the current status here. |
|
@malfet your PR has been successfully reverted. |
This reverts commit 0a2d320. Reverted #152443 on behalf of https://github.com/wdvr due to failing MPS test: test/test_optim.py::TestOptimRenewedMPS::test_can_load_from_to_named_state_dict_is_named_optim0_False_is_named_optim1_False_Adafactor_mps_float32 ([comment](#152443 (comment)))
|
Starting merge as part of PR stack under #152466 |
|
@pytorchbot merge -f "Lint + MPS + trunk are green" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Which multiplies two arguments for either scalar or complex data types This allows one to get rid of bunch of complex specialization in BinaryOps Pull Request resolved: #152466 Approved by: https://github.com/dcci ghstack dependencies: #152443
As well as `.add`/`.sub` with complex alpha Before this change `python3 -c "import torch;print(torch.rand(10, device='mps', dtype=torch.complex64).add(torch.rand(10, device='mps', dtype=torch.complex64), alpha=.5j))"` used to fail with ``` RuntimeError: value cannot be converted to type double without overflow ``` Pull Request resolved: #152479 Approved by: https://github.com/dcci ghstack dependencies: #152443, #152466
First of all, by extending `c10::metal::cast_to` to work correctly with complex dtypes, by introducing two more specializations: one that casts complex to scalar, and another that casts scalar to complex (as default metal typecast will turn `float x` into `float2(x, x)`) Add ComplexHalf and ComplexFloat enum values to `c10::metal::ScalarTypes` and handle them in `val_at_offs(ptr, offs, type)` Pull Request resolved: #152504 Approved by: https://github.com/dcci ghstack dependencies: #152443, #152466, #152479
Which was almost a complete copy-n-paste from exec_binary_kernel anyway Just add `Scalar` as an optional argument and figure out kernel name during the invocation rather than in executor Pull Request resolved: #152485 Approved by: https://github.com/Skylion007 ghstack dependencies: #152443, #152466, #152479, #152504
As typecasting harness shoudl take care of all permutations Fix bug in `exec_binary_kernel` where it was not properly downcasting CPU double/complexDouble scalars to floats Fixes #152582 Pull Request resolved: #152510 Approved by: https://github.com/Skylion007, https://github.com/dcci, https://github.com/cyyever ghstack dependencies: #152443, #152466, #152479, #152504, #152485
Stack from ghstack (oldest at bottom):
exec_binary_alpha_kernel#152485c10::metal::mul#152466For
lerp.Scalar_outweight (aka alpha) is not an optional argument, so no point in having those specializations.But move
alpha=1.0ahead of dispatching to Metal shaders, as plain copy of tensor should still be fasterpytorch/aten/src/ATen/native/mps/operations/BinaryOps.mm
Lines 285 to 290 in a1a4fee