Reduce overhead when Future invokes callbacks inline #57638
Closed
Conversation
This file contains hidden or 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
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit 4bab8db (more details on the Dr. CI page):
🕵️ 1 new failure recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
This was referenced May 5, 2021
lw
added a commit
that referenced
this pull request
May 5, 2021
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
ghstack-source-id: 128188696
Pull Request resolved: #57638
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
lw
added a commit
that referenced
this pull request
May 6, 2021
Pull Request resolved: #57638 In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. ghstack-source-id: 128297741 Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
This was referenced May 7, 2021
Closed
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
This was referenced May 17, 2021
mrshenli
approved these changes
May 17, 2021
It's a static assert, hence an incorrect signature will cause a compile error. I don't know how we can test that? |
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
This was referenced May 20, 2021
In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. In other words, if the compiler were to inline this new version of `addCallback` it would obtain the _exact_ same code as that explicit fastpath.
Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
[ghstack-poisoned]
dgl-intel
pushed a commit
to dgl-intel/pytorch
that referenced
this pull request
May 21, 2021
Pull Request resolved: pytorch#57638 In RPC there are a few instances of "fastpaths" which do `if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }`. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (because `addCallback` invokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around a `std::function`. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweaking `addCallback` (and similar methods) so they can handle raw lambdas, and so that they do _not_ wrap them into `std::function`s if they are invoked inline. ghstack-source-id: 129567067 Differential Revision: [D28222808](https://our.internmc.facebook.com/intern/diff/D28222808/)
|
This pull request has been merged in 1d7cf4b. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Stack from ghstack:
In RPC there are a few instances of "fastpaths" which do
if (fut.isCompleted()) { do_sth(); } else { fut.addCallback(do_sth); }. I intend to get rid of them, for reasons I'll clarify later but which in a nutshell have to do with CUDA correctness and readability. Note that dropping the fastpath introduces no change in behavior (becauseaddCallbackinvokes the callback inline anyways), thus the only perf concern comes from the fact that the fastpath avoids constructing and passing around astd::function. I don't think this is a significant performance hit. Regardless, this PR preemptively addresses this concern, by tweakingaddCallback(and similar methods) so they can handle raw lambdas, and so that they do not wrap them intostd::functions if they are invoked inline. In other words, if the compiler were to inline this new version ofaddCallbackit would obtain the exact same code as that explicit fastpath.Differential Revision: D28222808