[MPS] Fix the ChannelsLast memory format in cat_out_mps() (#91786)
- Fixed the memory leak with the `malloc()`
- Introduced shortened data type strings (optional) to avoid getting extra long cached graph string keys with ops such as cat_out()
- Fixed data type issues in Monterey
- Removed the unused `use_scalar_value` argument from `getTensorsStringKey()`
- Clean up and refactoring
Fixes #89353
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91786
Approved by: https://github.com/kulinseth
diff --git a/test/test_mps.py b/test/test_mps.py
index bb10422..0525a0c 100644
--- a/test/test_mps.py
+++ b/test/test_mps.py
@@ -8873,7 +8873,7 @@
# does not support float64 Tensors.
# A few ops are currently broken on their reference inputs, but not their sample inputs. These should
# get patched up and this workaround removed.
- broken_on_ref_inputs = op.name in ['cat', 'clamp', 'where']
+ broken_on_ref_inputs = op.name in ['clamp', 'where']
inputs = op.reference_inputs(device, dtype) if not broken_on_ref_inputs else op.sample_inputs(device, dtype)
for sample_input in inputs:
self.compare_with_reference(op, op.ref, sample_input)
