Fixing cudax::execution CUDA stream scheduler #6175

ericniebler · 2025-10-10T00:28:08Z

this pr ports cudax's stream scheduler to the new sender algorithm customization framework.

fixes #5564

copy-pr-bot · 2025-10-10T00:28:12Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

ericniebler · 2025-10-10T00:28:29Z

/ok to test d4ca1d7

ericniebler · 2025-10-14T00:02:06Z

/ok to test 4393729

ericniebler · 2025-10-14T04:20:25Z

/ok to test 0b25974

ericniebler · 2025-10-14T15:22:58Z

/ok to test 0b25974

ericniebler · 2025-10-14T15:39:20Z

/ok to test 4ed1940

ericniebler · 2025-10-14T18:38:31Z

/ok to test 95bf8fb

ericniebler · 2025-10-14T23:00:07Z

/ok to test 0879990

cudax/include/cuda/experimental/__execution/stream/sequence.cuh

cudax/include/cuda/experimental/__execution/stream/starts_on.cuh

cudax/include/cuda/experimental/__container/async_buffer.cuh

mhoemmen

I have reviewed this and made comments about changes which I don't quite understand.

I am approving this per @ericniebler 's request so that it can run through the nightlies.

cudax/test/execution/test_on.cu

mhoemmen · 2025-10-16T03:28:50Z

cudax/test/execution/test_stream_context.cu

 //
 //===----------------------------------------------------------------------===//

+// BUGBUG BUGBUG


What's the bug?

the bug is that i forgot to remove this comment. :-P

mhoemmen · 2025-10-16T03:30:43Z

cudax/include/cuda/experimental/__execution/bulk.cuh

      execution::set_stopped(static_cast<_Rcvr&&>(__state_->__rcvr_));
    }

-    [[nodiscard]] _CCCL_NODEBUG_API constexpr auto get_env() const noexcept -> __fwd_env_t<env_of_t<_Rcvr>>


I do not understand the consequences of this macro change.

It changes the functions' attributes so that debuggers won't step over them. Otherwise it has no effect.

@ericniebler You mean, it stops inlining. We do need to restore this at some point : - )

i would like to keep these changes to the function attributes. i learned many things while stepping through the experimental execution code in cuda-gdb. it turns out i rarely want "nodebug" apis. i very occasionally want "trivial" (force-inline nodebug) apis, but most apis should just be _CCCL_API.

mhoemmen · 2025-10-16T03:31:56Z

cudax/include/cuda/experimental/__execution/continues_on.cuh

  }

-  _CCCL_NO_UNIQUE_ADDRESS continues_on_t __tag_;
+  /*_CCCL_NO_UNIQUE_ADDRESS*/ continues_on_t __tag_;


Eric explained that tests were failing without removing this macro. I'm treating this removal as a temporary phenomenon. There are other ways to get this effect, e.g., the "compressed pair" pattern that the reference implementation of mdspan classically used.

i need to track down the compiler bug, file it, and then selectively disable _CCCL_NO_UNIQUE_ADDRESS acrosss all of CCCL wherever the bug can potentially manifest. then i can start using _CCCL_NO_UNIQUE_ADDRESS again.

@ericniebler The other option is to use a different programming technique that guarantees that empty members occupy zero bytes. A classic one is a "compressed pair" or "compressed tuple" for storing the members, that does not store empty members.

that's true. it will be important when this code is no longer experimental to have all the space optimizations before we lock in the ABI.

I want to note that I am on the verge of completely dropping any use of _CCCL_NO_UNIQUE_ADDRESS

Its just a complete trainwreck in the waiting

@miscco wrote:

Its just a complete trainwreck in the waiting

100% agree; as a language feature it means well, but in practice for us it's nothing but trouble.

mhoemmen · 2025-10-16T03:32:08Z

cudax/include/cuda/experimental/__execution/cpos.cuh

 {
  template <__disposition _OtherDisposition>
-  _CCCL_NODEBUG_API constexpr auto operator==(__completion_tag<_OtherDisposition>) const noexcept -> bool
+  _CCCL_TRIVIAL_API constexpr auto operator==(__completion_tag<_OtherDisposition>) const noexcept -> bool


I do not understand the consequences of this change.

a "trivial" api is one that is force-inline-ed and annotated so that debuggers step over them. execution::set_stopped(move(rcvr)) just calls move(rcvr).set_stopped() for example. it is never interesting to step into execution::set_stopped; you'd rather step directly into move(rcvr).set_stopped().

these annotations make debugging easier and shorten stack traces to just the good parts.

@ericniebler wrote:

these annotations make debugging easier and shorten stack traces to just the good parts.

We're gonna put these macros back like they were after the nightlies run, right?

that isn't my intention. these changes are intentional. they improve the debugability of the code.

Ah, got it. Thanks for explaining!

ericniebler · 2025-10-16T04:09:08Z

cudax/test/execution/test_stream_context.cu

+// BUGBUG BUGBUG
+


Suggested change

// BUGBUG BUGBUG

github-actions · 2025-10-16T04:42:44Z

🥳 CI Workflow Results

🟩 Finished in 20m 20s: Pass: 100%/42 | Total: 2h 47m | Max: 10m 32s | Hits: 99%/21364

See results here.

github-project-automation bot added this to CCCL Oct 10, 2025

github-project-automation bot moved this to Todo in CCCL Oct 10, 2025

cccl-authenticator-app bot moved this from Todo to In Progress in CCCL Oct 10, 2025

This comment has been minimized.

Sign in to view

ericniebler force-pushed the fixing-the-stream-scheduler branch 6 times, most recently from 4393729 to 04afd01 Compare October 14, 2025 00:01

This comment has been minimized.

Sign in to view

ericniebler force-pushed the fixing-the-stream-scheduler branch from 04afd01 to 0b25974 Compare October 14, 2025 04:19

This comment has been minimized.

Sign in to view

ericniebler added 4 commits October 14, 2025 18:37

get the CUDA stream scheduler in cudax::execution working again

10257dc

add missing #include

2754a37

use concepts portability macros

5f05d4c

disable unavoidable execution space warnings

95bf8fb

ericniebler force-pushed the fixing-the-stream-scheduler branch from b21f145 to 95bf8fb Compare October 14, 2025 18:38

This comment has been minimized.

Sign in to view

portability fixes and warning suppression

0879990

This comment has been minimized.

Sign in to view

ericniebler marked this pull request as ready for review October 15, 2025 01:50

ericniebler requested review from a team as code owners October 15, 2025 01:50

ericniebler requested a review from pciolkosz October 15, 2025 01:50

cccl-authenticator-app bot moved this from In Progress to In Review in CCCL Oct 15, 2025

ericniebler linked an issue Oct 15, 2025 that may be closed by this pull request

[BUG]: CUDA stream scheduler in cudax execution library is broken #5564

Closed

1 task

This comment has been minimized.

Sign in to view

ericniebler changed the title ~~[WIP] Fixing cudax::execution CUDA stream scheduler~~ Fixing cudax::execution CUDA stream scheduler Oct 15, 2025

miscco reviewed Oct 15, 2025

View reviewed changes

cudax/include/cuda/experimental/__execution/stream/sequence.cuh Show resolved Hide resolved

cudax/include/cuda/experimental/__execution/stream/starts_on.cuh Outdated Show resolved Hide resolved

pciolkosz reviewed Oct 15, 2025

View reviewed changes

cudax/include/cuda/experimental/__container/async_buffer.cuh Outdated Show resolved Hide resolved

spit and polish

982f48e

ericniebler requested a review from a team as a code owner October 15, 2025 23:24

ericniebler requested a review from alliepiper October 15, 2025 23:24

fix bugs in bulk for the stream scheduler

a6c97a8

ericniebler force-pushed the fixing-the-stream-scheduler branch from 570ab90 to a6c97a8 Compare October 15, 2025 23:30

This comment has been minimized.

Sign in to view

mhoemmen approved these changes Oct 16, 2025

View reviewed changes

uncomment execution::on tests

af438e6

ericniebler commented Oct 16, 2025

View reviewed changes

This comment has been minimized.

Sign in to view

remove stray BUGBUG comment

fab2af9

ericniebler enabled auto-merge (squash) October 16, 2025 05:06

ericniebler disabled auto-merge October 16, 2025 05:06

ericniebler enabled auto-merge (squash) October 16, 2025 05:07

pciolkosz approved these changes Oct 16, 2025

View reviewed changes

ericniebler merged commit 907a153 into NVIDIA:main Oct 16, 2025
53 checks passed

github-project-automation bot moved this from In Review to Done in CCCL Oct 16, 2025

ericniebler deleted the fixing-the-stream-scheduler branch October 16, 2025 15:55

Fixing cudax::execution CUDA stream scheduler #6175

Fixing cudax::execution CUDA stream scheduler #6175

Uh oh!

Conversation

ericniebler commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

copy-pr-bot bot commented Oct 10, 2025

Uh oh!

ericniebler commented Oct 10, 2025

Uh oh!

This comment has been minimized.

ericniebler commented Oct 14, 2025

Uh oh!

This comment has been minimized.

ericniebler commented Oct 14, 2025

Uh oh!

This comment has been minimized.

ericniebler commented Oct 14, 2025

Uh oh!

ericniebler commented Oct 14, 2025

Uh oh!

This comment has been minimized.

ericniebler commented Oct 14, 2025

Uh oh!

This comment has been minimized.

ericniebler commented Oct 14, 2025

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

mhoemmen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

github-actions bot commented Oct 16, 2025

🥳 CI Workflow Results

🟩 Finished in 20m 20s: Pass: 100%/42 | Total: 2h 47m | Max: 10m 32s | Hits: 99%/21364

Uh oh!

ericniebler commented Oct 10, 2025 •

edited

Loading