[pt2] add metas for max_unpool2d and max_unpool3d
#103821
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/103821
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit da9b41f: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
nkaretnikov
added
ciflow/trunk
|
Keeping the skips in |
| TORCH_CHECK( | ||
| output_size.size() == 2, | ||
| "There should be exactly two elements (width, height) in output_size, but got ", output_size.size(), " elements."); | ||
| "There should be exactly two elements (height, width) in output_size, but got ", output_size.size(), " elements."); |
There was a problem hiding this comment.
Typo fix. Wrong order of elements.
| .view_symint(size) | ||
| .gather(-1, indices_view) | ||
| .view(indices.sizes()); | ||
| .view_symint(indices.sym_sizes()); |
There was a problem hiding this comment.
For example, used in test_aot_autograd_symbolic_exhaustive_nn_functional_max_unpool1d_cpu_float32.
[ghstack-poisoned]
[ghstack-poisoned]
| from torch._ops import OpOverload | ||
| from torch._prims import _elementwise_meta, ELEMENTWISE_PRIM_TYPE_PROMOTION_KIND | ||
| from torch._prims_common import ( | ||
| corresponding_complex_dtype, |
There was a problem hiding this comment.
These alerts are needed for, e.g.:
PYTORCH_TEST_WITH_INDUCTOR=1 python -bb test/test_torch.py -v --use-pytest --import-slow-tests --import-disabled-tests -k test_nondeterministic_alert_MaxUnpool1d_cuda_float16 -v --capture=no
| ), | ||
| ) | ||
|
|
||
| self = self_.contiguous() |
There was a problem hiding this comment.
In core, this uses suggest_memory_format, so slightly wrong. But is this even necessary? Will investigate.
UPD: But it's only for the 2d variant. I'm going to ignore this for now as all tests pass.
| else: | ||
| nbatch = self.size(0) | ||
| nchannels = self.size(1) | ||
| result = self.new_empty((nbatch, nchannels, oheight, owidth)) |
There was a problem hiding this comment.
This takes a memory_format argument in core.
UPD: But it's only for the 2d variant. I'm going to ignore this for now as all tests pass.
| f"determinism just for this operation, or you can use the " | ||
| f"'warn_only=True' option, if that's acceptable for your application. " | ||
| f"You can also file an issue at https://github.com/pytorch/pytorch/issues " | ||
| f"to help us prioritize adding deterministic support for this operation.")) |
There was a problem hiding this comment.
Context::alertNotDeterministic in core.
nkaretnikov
removed
ciflow/periodic
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
5 participants
Stack from ghstack:
SymIntsupport formax_poolops #103951max_unpool2dandmax_unpool3d#103821