uni20::async::TbbCudaSchedulerOptions Struct Reference

Arena configuration for the unified host/CUDA TBB scheduler. More...

#include <tbb_cuda_scheduler.hpp>

Public Attributes
int	cuda_max_concurrency_per_device = 1
	Maximum worker participation in each CUDA device arena.
std::optional< int >	default_cuda_device {}
	Device used to route CUDA tasks that have not established affinity.
int	host_max_concurrency = oneapi::tbb::task_arena::automatic
	Maximum participation in the host arena.

Detailed Description

Arena configuration for the unified host/CUDA TBB scheduler.

Member Data Documentation

cuda_max_concurrency_per_device

int uni20::async::TbbCudaSchedulerOptions::cuda_max_concurrency_per_device = 1

Maximum worker participation in each CUDA device arena.

CUDA submission defaults to one worker per device; callers may opt into concurrent submission independently of stream and provider capacity.

default_cuda_device

std::optional<int> uni20::async::TbbCudaSchedulerOptions::default_cuda_device {}

Device used to route CUDA tasks that have not established affinity.

The first enrolled device is used when this is empty.

host_max_concurrency

int uni20::async::TbbCudaSchedulerOptions::host_max_concurrency = oneapi::tbb::task_arena::automatic

Maximum participation in the host arena.

---

The documentation for this struct was generated from the following file:

• tbb_cuda_scheduler.hpp