Skip to content
This repository has been archived by the owner on Mar 21, 2024. It is now read-only.

Documentation changes for wait_parity/try_wait_parity #190

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions docs/extended_api/synchronization_primitives/barrier.md
Original file line number Diff line number Diff line change
Expand Up @@ -21,8 +21,9 @@ It has the same interface and semantics as [`cuda::std::barrier`], with the

## Barrier Operations

| [`cuda::barrier::init`] | Initialize a `cuda::barrier`. `(friend function)` |
| [`cuda::device::barrier_native_handle`] | Get the native handle to a `cuda::barrier`. `(function template)` |
| [`cuda::barrier::init`] | Initialize a `cuda::barrier`. `(friend function)` |
| [`cuda::device::barrier_native_handle`] | Get the native handle to a `cuda::barrier`. `(function template)` |
| [`cuda::barrier::wait_parity/try_wait_parity`] | Wait on or check the parity of the barrier. |

## NVCC `__shared__` Initialization Warnings

Expand Down Expand Up @@ -98,6 +99,7 @@ __global__ void example_kernel() {

[`cuda::barrier::init`]: ./barrier/init.md
[`cuda::device::barrier_native_handle`]: ./barrier/barrier_native_handle.md
[`cuda::barrier::wait_parity/try_wait_parity`]: ./barrier/wait_parity.md

[`cuda::std::barrier`]: https://en.cppreference.com/w/cpp/thread/barrier

Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
---
grand_parent: Extended API
parent: Barriers
---

# `cuda::barrier::wait_parity` and `cuda::barrier::try_wait_parity`

Defined in header `<cuda/std/barrier>`:

```cuda
__host__ __device__ void cuda::std::barrier::wait_parity(bool phase);
__host__ __device__ bool cuda::std::barrier::try_wait_parity(bool phase);
```

`barrier::wait_parity` stalls execution while the barrier is not at the specified parity.
Copy link
Collaborator

@gonzalobg gonzalobg Mar 7, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

atomic::wait(value) "waits until the value is no longer value".

Here, wait_parity(value) "waits until the value becomes value" which is the opposite.

Is this intended? Seems subtly different.

@wmaxey ? @ogiroux ?

Copy link
Contributor

@davedsth davedsth Mar 7, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should swap it to match atomic wait. Also make sense as we are waiting for phase to change

`barrier::try_wait_parity` returns true if the parity of the barrier matches the given parity.

## Return Value

`barrier::try_wait_parity` returns a boolean representing whether the barrier is at the given phase

<!-- TODO: Create an example when trunk is live on godbolt
## Example

```cuda
#include <cuda/barrier>

__global__ void example_kernel(cuda::barrier<cuda::thread_scope_block>& bar) {
bar.wait_parity(false);
}
```
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be good to have an example that shows how to use these APIs to accomplish something minimally useful.

It might not be clear to the user what these APIs are for, what the parity of the barrier is initially, how to get the parity of a phase, etc.

-->

[See it on Godbolt](https://godbolt.org/z/dr4798Y76){: .btn }


[`cuda::thread_scope`]: ./thread_scopes.md

[thread.barrier.class paragraph 12]: https://eel.is/c++draft/thread.barrier.class#12

[coalesced threads]: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#coalesced-group-cg

[`concurrentManagedAccess` property]: https://docs.nvidia.com/cuda/cuda-runtime-api/structcudaDeviceProp.html#structcudaDeviceProp_116f9619ccc85e93bc456b8c69c80e78b
[`hostNativeAtomicSupported` property]: https://docs.nvidia.com/cuda/cuda-runtime-api/structcudaDeviceProp.html#structcudaDeviceProp_1ef82fd7d1d0413c7d6f33287e5b6306f