Eric Niebler

`mr::resource` concept should require `const` allocate and `noexcept` deallocate

3

# `const` allocate memory resources have `allocate` and `deallocate` functions, and no observers of any mutable state. that makes them (logically) immutable. `allocate` and `deallocate` do not change any observable...

[FEA]: optimize the case of storing a `resource_ref` in an `any_resource`

1

### Is this a duplicate? - [x] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cccl/issues) for this request and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md) ### Area...

[BUG]: inconsistent use of macros to suppress doxygen document generation

1

### Is this a duplicate? - [x] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cccl/issues) for this bug and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md) ### Type...

fixes #414

i notice that when i am editing a CUDA C++ source file, doxygen-style comment blocks like `///` and `//!` are not respected by Rewrap. Before:  After:  if i...

add a minimally functional execution context for CUDA streams

10

## Description this PR gives ustdex a minimally functional CUDA stream scheduler and customizes the `sync_wait` algorithm for stream execution. NB: i've based this PR on branch NVIDIA:pull-request/4539 so that...

Add domain support and make all algorithms customizable with domain-based dispatch

10

## Description this PR makes the sender algorithms of ustdex eagerly customizable based on the domain of sender argument(s). NB: i've based this PR on branch NVIDIA:pull-request/4538 so that only...

fix some portability issues with the cudax async tests

4

## Description while working on a different set of changes, i found i had to tweak the tests of ustdex for clang-cuda to accept them. these are those minor changes....

make `run_loop` lock-free and usable from device code

8

## Description the current implementation of `run_loop` uses `std::mutex` and `std::condition_variable`. these types have no analogue in `cuda::std::`. as a result, `run_loop` has been limited to host-only, and hence so...

use `forward` and friends from `std::` to leverage compiler optimizations

12

the latest releases of most modern compilers treat functions like `std::move` and `std::forward` as intrinsics, leading to better codegen at -O0 and a better debugging experience. it also makes builds...