dependencies/libcudacxx/docs/extended_api/synchronization_primitives/atomic/fetch_max.md

metadata

grand_parent: Synchronization Primitives
parent: cuda::atomic

cuda::atomic::fetch_max

Defined in header <cuda/atomic>:

template <typename T, cuda::thread_scope Scope>
__host__ __device__
T cuda::atomic<T, Scope>::fetch_max(T const& val,
                                    cuda::std::memory_order order
                                      = cuda::std::memory_order_seq_cst);

Atomically find the maximum of the value stored in the cuda::atomic and val. The maximum is found using cuda::std::max.

Example

#include <cuda/atomic>

__global__ void example_kernel() {
  cuda::atomic<int> a(0);
  auto x = a.fetch_max(1); // Operates as if unsigned.
  auto y = a.load();
  assert(x == 1 && y == 0);
}

See it on Godbolt{: .btn }