-
Notifications
You must be signed in to change notification settings - Fork 281
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ch4/ofi: refactor MPIDI_OFI_request_t #6895
Open
hzhou
wants to merge
26
commits into
pmodels:main
Choose a base branch
from
hzhou:2402_ofi_req
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
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
MPIR_gpu_req is a union type for either a MPL_gpu_request or a MPIR_Typerep_req, thus it is not just for gpu. Potentially this type can be extended to include other internal async task handles. Thus we rename it to MPIR_async_req. We also establish the convention of naming the variable async_req.
Add an inline wrapper for testing MPIR_async_req. Modify the order of header inclusion due to the dependency on typerep_pre.h.
Refactor the async copy in receive events using MPIR_async facilities.
Refactor the async copy before sending a chunk.
Both gpu_send_task_queue and gpu_recv_task_queue have been ported to async things.
hzhou
force-pushed
the
2402_ofi_req
branch
2 times, most recently
from
February 6, 2024 23:31
ed2c544
to
0cf1638
Compare
Pipeline send allocates chunk buffers then spawns async copy. The allocation may run out of genq buffers, thus it is disigned as async tasks. The send copy are triggered upon completion of buffer alloc, thus it is renamed into spawn_send_copy and turned into internal static function. This removes MPIDI_OFI_global.gpu_send_queue.
Pipeline recv allocates chunk buffers and then post fi_trecv. The allocation may run out of genq buffers and we also control the number of outstanding recvs, thus it is designed as async tasks. The async recv copy are triggered in recv event when data arrives. This removes MPIDI_OFI_global.gpu_recv_queue. All ofi-layer progress routines for gpu pipelining are now removed.
Consolidate the gpu pipeline code. MPIDI_OFI_gpu_pipeline_request is now an internal struct in ofi_gpu_pipeline.c, rename to struct chunk_req. MPIDI_OFI_gpu_pipeline_recv_copy is now an internal function, rename to start_recv_copy.
Move all gpu pipeline specific code into ofi_gpu_pipeline.c. Make a new function MPIDI_OFI_gpu_pipeline_recv that fills rreq with persistent pipeline_info data. Rename the original MPIDI_OFI_gpu_pipeline_recv into static function start_recv_chunk.
Make the code cleaner to separate the pipeline_info type into a union of send and recv.
Don't mix the usage of cc_ptr, use separate and explicit counters to track the progress and completion of chunks.
Follow a similar approach as nonblocking collectives, internal pipeline chunks use separate tag space (MPIDI_OFI_GPU_PIPELINE_SEND) and incrementing tags to avoid mismatch with regular messages.
hzhou
force-pushed
the
2402_ofi_req
branch
3 times, most recently
from
February 7, 2024 21:16
951bf49
to
127a607
Compare
hzhou
force-pushed
the
2402_ofi_req
branch
2 times, most recently
from
February 8, 2024 16:54
b7f89a7
to
4318961
Compare
Separate the recv tasks between the initial header and chunks since the paths clearly separates them. Use a single async item for all chunk recvs rather than unnecessarily enqueuing individual chunks since we can track the chunks in the state.
It is needed to compile under noinline configuration.
Move these utility functions to ofi_impl.h since they are simple and non-specific. It also simplifies figuring out which file to include especially for .c files.
Separate non-intercepting paths with distinct union members to make the code more explicit.
The paths of pipeline code is not intercepting with other paths, thus pipeline_info can be part of the same union.
Rather than sharing the pack_buffer field in pack_send, which obfuscates the code, use a separate union member for the huge send path.
It is part of the recv path, thus need be in the recv struct. The recv paths may switch paths depending on the actual protocols used by sender, thus some of the fields from different paths need live in the same struct and use the NULL sentinel to tell whether certain path is in effect. Currently this include remote_info for huge_send protocol, and pack_buffer for pack_recv. It's possible to have huge_recv+pack_recv.
The am emulated inject path does not intercept with any native paths, thus it should be part of the big union.
The util.iov field is used by fi_trecvmsg with FI_CLAIM flag or the huge recv path for threshold checking and fi_read.
The fast path of MPIDI_OFI_EVENT_SEND and MPIDI_OFI_EVENT_RECV has already been handled in MPIDI_OFI_dispatch_optimized. Thus it is not critical to mark likely/unlikely branches in MPIDI_OFI_dispatch_function. Just use a big switch for simplicity.
There is no way to recover a pipelined message in the native ofi path if the receiver didn't expect it. We could always save the necessary information in all recv paths, but I am not sure we are willing to accept the overhead, which harms the small message latencies. Assert failure for now. The proper pipeline implementation need to happen as active messages or even at the MPIR-layer.
Remove a redundant assignment of *request.
test:mpich/ch4/ofi |
test:mpich/ch4/ofi |
raffenet
approved these changes
Feb 20, 2024
This request object changes look good. I think I had a few comments on the previous PR that I need to revisit before this can go in. |
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.
Pull Request Description
Having multiple non-intercepting code paths to use the same structure fields in request makes code difficult to read.
This PR separates the fields into distinct union members with more intuitive names in order to make the code more readable.
Before this PR the fields are defined as:
It is quite confusing on which paths are using which fields and which paths are intercepting. After this PR:
The union fields are not intercepting and clearly identifies the code paths.
[skip warnings]
Author Checklist
Particularly focus on why, not what. Reference background, issues, test failures, xfail entries, etc.
Commits are self-contained and do not do two things at once.
Commit message is of the form:
module: short description
Commit message explains what's in the commit.
Whitespace checker. Warnings test. Additional tests via comments.
For non-Argonne authors, check contribution agreement.
If necessary, request an explicit comment from your companies PR approval manager.