Dispatcher tutorial #1072
Dispatcher tutorial #1072
Conversation
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
|
Deploy preview for pytorch-tutorials-preview ready! Built with commit 2a96e0d https://deploy-preview-1072--pytorch-tutorials-preview.netlify.app |
advanced_source/dispatcher.rst
Outdated
| :end-before: END myadd_cpu | ||
|
|
||
| We'd like to register this function as an implementation of ``myops::myadd``, but we | ||
| don't want to register it as a catch-all kernel to be run in all cases; we |
There was a problem hiding this comment.
I don't know if later text relies on the notion of catch-all being introduced here, but for the immediate explanation you could probably just omit it - "we'd like to register this function as an implementation... but we only want it to be run..."
advanced_source/dispatcher.rst
Outdated
| defined in the ``TORCH_LIBRARY`` block). You can have as many | ||
| ``TORCH_LIBRARY_IMPL`` blocks for a namespace as you like; so for example, | ||
| if we also have a CUDA implementation ``myadd_cuda``, we can register it | ||
| with: |
There was a problem hiding this comment.
Might this paragraph leave you with the impression that you need a separate TORCH_LIBRARY_IMPL block for each m.impl() call?
advanced_source/dispatcher.rst
Outdated
|
|
||
| These registrations can be split across files or even across library boundaries; so | ||
| for example, you could have these two ``TORCH_LIBRARY_IMPL`` blocks compiled | ||
| into a separate ``myops_cpu`` and ``myops_cuda`` dynamic library. |
There was a problem hiding this comment.
"into separate ... dynamic libraries", I think
| core operators in PyTorch? This is how XLA support for PyTorch is | ||
| implemented: the ``torch_xla`` library contains a ``TORCH_LIBRARY_IMPL`` | ||
| that provides implementations for all basic operators on the XLA dispatch | ||
| key. |
advanced_source/dispatcher.rst
Outdated
| 2. Call the dispatch function ``myadd`` to call back into the dispatcher. | ||
|
|
||
| Without (1), your calls will infinite loop (and stack overflow), because | ||
| ``myadd`` will send you back to the autograd implementation! With (1), |
There was a problem hiding this comment.
I'm not sure how to integrate the information smoothly, but it might clarify to note that we get back here because this function will be registered to Autograd (per what immediately follows ofc)
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
* Dispatcher tutorial Signed-off-by: Edward Z. Yang <ezyang@fb.com> * typofix Signed-off-by: Edward Z. Yang <ezyang@fb.com> * morefix Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Signed-off-by: Edward Z. Yang ezyang@fb.com