This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
fix race when temp space is used in copy & fix instance overwrite in g2c #8867
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
https://github.com/apache/incubator-mxnet/blob/master/src/kvstore/comm.h#L187, |
ZiyueHuang
commented
Nov 29, 2017
| @@ -76,7 +76,7 @@ | |||
|
|
|||
| # construct the module | |||
| # map the ctx_group attribute to the context assignment | |||
| group2ctxs={'dev1':mx.cpu(), 'dev2':[mx.gpu(i) for i in range(num_gpus)]} | |||
| group2ctxs={'dev1':[mx.cpu()]*num_gpus, 'dev2':[mx.gpu(i) for i in range(num_gpus)]} | |||
There was a problem hiding this comment.
This change is just for better understandability.
eric-haibin-lin
suggested changes
Nov 30, 2017
src/ndarray/ndarray.cc
Outdated
| @@ -454,7 +454,8 @@ inline void CopyFromToDnsImpl(const NDArray& from, const NDArray& to, RunContext | |||
|
|
|||
| // Make a copy of an NDArray based on storage type | |||
| template<typename from_xpu, typename to_xpu> | |||
| void CopyFromToImpl(const NDArray& from, const NDArray& to, RunContext rctx) { | |||
| void CopyFromToImpl(const NDArray& from, const NDArray& to, | |||
| RunContext rctx, std::vector<Resource> requested) { | |||
There was a problem hiding this comment.
const reference for the vector?
src/ndarray/ndarray.cc
Outdated
| @@ -518,43 +515,57 @@ void CopyFromTo(const NDArray& from, const NDArray& to, int priority) { | |||
| CHECK(from.shape().ndim() != 0) | |||
| << "source operands have zero dimension shape"; | |||
| // important: callback must always capture by value | |||
| int a = from.ctx().dev_mask(); | |||
| const auto from_ctx = from.ctx(); | |||
| int a = from_ctx.dev_mask(); | |||
There was a problem hiding this comment.
Nit : const a and const b
There was a problem hiding this comment.
Please avoid using auto for simple types
| std::vector<Engine::VarHandle> mutable_vars(1, to.var()); | ||
|
|
||
| std::vector<Resource> requested; | ||
| if (a == gpu::kDevMask && from_stype != to_stype) { |
There was a problem hiding this comment.
Accordding to original codes,
- std::vector<Resource> requested;
- if (is_same<from_xpu, mshadow::gpu>::value && from_stype != to_stype) {
- requested.push_back(ResourceManager::Get()->Request(from_ctx,
- ResourceRequest(ResourceRequest::kTempSpace)));
- }
Seems that whether temp space is used is irrelevant with the context of b ?
There was a problem hiding this comment.
Oh right. No need to request temp space if cast_storage happens on CPU.
tests/python/unittest/test_module.py
Outdated
|
|
||
| check_module_ctx_group([mx.cpu(0)], {'dev1': mx.cpu(1), 'dev2': mx.cpu(2)}) | ||
| check_module_ctx_group([mx.cpu(0)], {'dev1': mx.cpu(1), 'dev2': mx.cpu(2)}, [mx.cpu(1), mx.cpu(2)]) |
There was a problem hiding this comment.
Nit: I think explicitly mentioning optional arg names (grad_ctxs) when passing optional args is a good practice since API may change in the future
eric-haibin-lin
approved these changes
Nov 30, 2017
KellenSunderland
pushed a commit
to KellenSunderland/incubator-mxnet
that referenced
this pull request
Dec 13, 2017
…g2c (apache#8867) * fix race when temp space is used in copy * fix instance overwrite in g2c * example of g2c * address comments
zhreshold
pushed a commit
to zhreshold/mxnet
that referenced
this pull request
Dec 14, 2017
…g2c (apache#8867) * fix race when temp space is used in copy * fix instance overwrite in g2c * example of g2c * address comments
rahul003
pushed a commit
to rahul003/mxnet
that referenced
this pull request
Jun 4, 2018
…g2c (apache#8867) * fix race when temp space is used in copy * fix instance overwrite in g2c * example of g2c * address comments
zheng-da
pushed a commit
to zheng-da/incubator-mxnet
that referenced
this pull request
Jun 28, 2018
…g2c (apache#8867) * fix race when temp space is used in copy * fix instance overwrite in g2c * example of g2c * address comments
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
var of temp space should be in mutable_vars in engine.
[{}] * ctx_lenis actually one dict.cc @eric-haibin-lin
Checklist
Essentials
make lint)Changes
Comments