mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
e41a0b33ec
Previously if a wrapper tensor subclass is fakified, the inner tensors would end up having the same device as the outer tensor. This PR makes it so that inner and outer tensors can have different devices. See OffloadTensor PR https://github.com/pytorch/pytorch/pull/141840/files#diff-3bc0cf540b694f4ec0a3749f78b047456657a53a5657e495ffb68e5970c5fdaaR1955 for an application. A simpler test has been added in this PR. This is technically bc-breaking because now the callback passed to MetaConverter needs to accept an extra argument, but no one external should be using this anyway? Pull Request resolved: https://github.com/pytorch/pytorch/pull/141839 Approved by: https://github.com/bdhirsh ghstack dependencies: #141166
Note [TH abstraction violation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ TH/THC provide some hpp headers, which are proper C++ headers rather than C headers. These headers serve double duty as *internal implementation detail* headers, whose contents should largely not be used by external clients. Ideally, we would not install these headers at all; instead, you should use public functions (in headers like `THTensor.h`, NOT `THTensor.hpp`) to manipulate these structs. However, there are a few places in torch/csrc where we violate this abstraction. They are marked with a pointer to this note. Each of those sites will have to be refactored when we refactor the guts of THTensor and related structures.