Unify invoking JIT operands #57850
Closed
Unify invoking JIT operands #57850
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
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
lw
requested review from
H-Huang,
mingzhe09088,
mrshenli,
pritamdamania87,
rohan-varma,
wayi1 and
zhaojuanmao
as code owners
💊 CI failures summary and remediationsAs of commit 2095b41 (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:
|
facebook-github-bot
added
the
oncall: distributed
This was referenced May 7, 2021
Closed
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
This was referenced May 17, 2021
mrshenli
approved these changes
May 18, 2021
| "size ", | ||
| stack.size()); | ||
| markComplete(std::move(ScriptResp(std::move(stack.front()))).toMessage()); | ||
| auto future = runJitOperator(*scriptCall.op(), stack); |
There was a problem hiding this comment.
Will this switch thread? Do we need to carry profiling TLS to the new thread? NVM, I saw this is implemented below and won't switch thread.
Comment on lines
662
to
664
| "Return value of a builtin operator or a " | ||
| "TorchScript function should be a single IValue, got a vector of " | ||
| "size ", |
There was a problem hiding this comment.
nit: can fit in two lines.
mrshenli
reviewed
May 18, 2021
|
|
||
| // Helpers to run user-defined functions, operators and other computations. | ||
|
|
||
| c10::intrusive_ptr<JitFuture> runJitOperator( |
There was a problem hiding this comment.
If this is not used in derived classes, this does not needs to be a member function?
There was a problem hiding this comment.
Later in the stack it will be used by RequestCallbackImpl (the subclass)
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
This was referenced May 20, 2021
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/) [ghstack-poisoned]
dgl-intel
pushed a commit
to dgl-intel/pytorch
that referenced
this pull request
May 21, 2021
Pull Request resolved: pytorch#57850 What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow. This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams. ghstack-source-id: 129567058 Differential Revision: [D28253847](https://our.internmc.facebook.com/intern/diff/D28253847/)
|
This pull request has been merged in 7742815. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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:
What I envision is a modular decomposed code, with separate steps which each consume/produce Futures, and which can be chained together to obtain the desired results. One common "starting point" for these chains is the execution of a remote function (Python or JIT or otherwise). I'm thus creating a helper function for one of these, the JIT operators (by deduplicating the places where we used to run them). More will follow.
This deduplication will also help to add CUDA support to JIT RPC, since the execution of the JIT function/operators is where we need to set our custom streams.
Differential Revision: D28253847