Skip to content

Conversation

@ljk53
Copy link
Contributor

@ljk53 ljk53 commented Oct 3, 2019

Summary:
Pull Request resolved: #27274

This is yet another fix to address #26764.

PR #26908 toggles NonVariableTypeMode in ATen dispatcher, which is where
USE_STATIC_DISPATCH takes place thus it's most logically sound place to do
such tweaks.

However, we observed nontrivial perf regression due to this fix. Turns out
the numel() tensor method gets called in several for-loops thus incurs ~7M
thread_local updates in a single forward call:

7173330 numel
    558 size
    416 q_scale
    302 _empty_affine_quantized
    288 contiguous
    257 q_zero_point
    216 qscheme
    173 empty
    110 set_
    105 as_strided
    104 permute
...

As numel() is not called from a single place so a natural workaround is to
update function_wrapper.py so that it only adds the guard on gen_namespace_function()
case and ignore the gen_tensor_method() case. But some tensor methods are actually
being called from JIT side directly (e.g. "aten::eq_" -> "(self).eq_") so the
only "band aid" left on the table is to insert guard on JIT->aten path as originally
did on #26868 - this is a simplified version of it as it doesn't hurt to extend the
NonVariableMode scope a little bit to also cover stack drop/pack calls.

On Android we only expose JIT API so we don't need worry about TensorMethods being
called directly. On iOS we don't provide a wrapper yet but we can mention this caveat
in the doc. Hopefully by the time it's widely used we can finish Variable/Tensor
unification and remove all these hacks.

Test Plan:

Differential Revision: D17732489

Pulled By: ljk53

fbshipit-source-id: c14ca66aebc6b6f17ad6efac7ca47f9487c98de5

@pytorchbot pytorchbot added the oncall: jit Add this issue/PR to JIT oncall triage queue label Oct 3, 2019
@ljk53 ljk53 requested a review from soumith October 3, 2019 21:00
@ljk53 ljk53 mentioned this pull request Oct 3, 2019
…pytorch#27274)

Summary:
Pull Request resolved: pytorch#27274

This is yet another fix to address pytorch#26764.

PR pytorch#26908 toggles NonVariableTypeMode in ATen dispatcher, which is where
USE_STATIC_DISPATCH takes place thus it's most logically sound place to do
such tweaks.

However, we observed nontrivial perf regression due to this fix. Turns out
the numel() tensor method gets called in several for-loops thus incurs ~7M
thread_local updates in a single forward call:
```
7173330 numel
    558 size
    416 q_scale
    302 _empty_affine_quantized
    288 contiguous
    257 q_zero_point
    216 qscheme
    173 empty
    110 set_
    105 as_strided
    104 permute
...
```

As numel() is not called from a single place so a natural workaround is to
update function_wrapper.py so that it only adds the guard on gen_namespace_function()
case and ignore the gen_tensor_method() case. But some tensor methods are actually
being called from JIT side directly (e.g. "aten::eq_" -> "(self).eq_") so the
only "band aid" left on the table is to insert guard on JIT->aten path as originally
did on pytorch#26868 - this is a simplified version of it as it doesn't hurt to extend the
NonVariableMode scope a little bit to also cover stack drop/pack calls.

On Android we only expose JIT API so we don't need worry about TensorMethods being
called directly. On iOS we don't provide a wrapper yet but we can mention this caveat
in the doc. Hopefully by the time it's widely used we can finish Variable/Tensor
unification and remove all these hacks.

Test Plan:
- Verified it runs quantized/fp32 MobileNetV2 models;
- Verified it fixes the perf regression (revert pytorch#26908 separately);

Differential Revision: D17732489

Pulled By: ljk53

fbshipit-source-id: c14ca66aebc6b6f17ad6efac7ca47f9487c98de5
@soumith soumith merged commit ccf3a6d into pytorch:v1.3.0 Oct 7, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

oncall: jit Add this issue/PR to JIT oncall triage queue

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants