You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by "jinhongyii (via GitHub)" <gi...@apache.org> on 2023/09/13 16:44:30 UTC
[GitHub] [tvm] jinhongyii opened a new pull request, #15742: [Unity][Disco] separate computation and communication into 2 stream
jinhongyii opened a new pull request, #15742:
URL: https://github.com/apache/tvm/pull/15742
put computation on default stream and put communication on a new stream.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] jinhongyii commented on a diff in pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "jinhongyii (via GitHub)" <gi...@apache.org>.
jinhongyii commented on code in PR #15742:
URL: https://github.com/apache/tvm/pull/15742#discussion_r1326594671
##########
src/runtime/disco/nccl/nccl.cc:
##########
@@ -91,26 +90,34 @@ void AllReduce(NDArray send, ReduceKind reduce_kind, NDArray recv) {
NCCLThreadLocalContext* ctx = NCCLThreadLocalContext::Get();
ShapeTuple shape = send.Shape();
int64_t numel = shape->Product();
+ Device device = ctx->worker->default_device;
+ DeviceAPI::Get(device)->SyncStreamFromTo(device, nullptr, ctx->stream);
Review Comment:
makes sense
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] jinhongyii commented on pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "jinhongyii (via GitHub)" <gi...@apache.org>.
jinhongyii commented on PR #15742:
URL: https://github.com/apache/tvm/pull/15742#issuecomment-1717975788
cc: @junrushao
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao merged pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "junrushao (via GitHub)" <gi...@apache.org>.
junrushao merged PR #15742:
URL: https://github.com/apache/tvm/pull/15742
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] jinhongyii commented on pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "jinhongyii (via GitHub)" <gi...@apache.org>.
jinhongyii commented on PR #15742:
URL: https://github.com/apache/tvm/pull/15742#issuecomment-1719661144
null stream (or default stream equivalently) is used for compute. The stream created in init_ccl is for communication
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] jinhongyii closed pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "jinhongyii (via GitHub)" <gi...@apache.org>.
jinhongyii closed pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
URL: https://github.com/apache/tvm/pull/15742
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao commented on pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "junrushao (via GitHub)" <gi...@apache.org>.
junrushao commented on PR #15742:
URL: https://github.com/apache/tvm/pull/15742#issuecomment-1719305142
i'm trying to figure out - which stream is used for compute and which is for communication? are we using null stream for compute?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao commented on a diff in pull request #15742: [Unity][Disco] separate computation and communication into 2 stream
Posted by "junrushao (via GitHub)" <gi...@apache.org>.
junrushao commented on code in PR #15742:
URL: https://github.com/apache/tvm/pull/15742#discussion_r1326593947
##########
src/runtime/disco/nccl/nccl.cc:
##########
@@ -91,26 +90,34 @@ void AllReduce(NDArray send, ReduceKind reduce_kind, NDArray recv) {
NCCLThreadLocalContext* ctx = NCCLThreadLocalContext::Get();
ShapeTuple shape = send.Shape();
int64_t numel = shape->Product();
+ Device device = ctx->worker->default_device;
+ DeviceAPI::Get(device)->SyncStreamFromTo(device, nullptr, ctx->stream);
Review Comment:
one minor comment: please add a field `ctx->compute_stream` and rename `ctx->stream` to `ctx->comm_stream`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org