Re:Lock-free multithreading with atomic operations
by dragontamer on Tuesday July 30, 2019 @10:47AM
(#20565277)
Ah, you're right. They call them "barriers" instead of "fences", so I didn't pick up on that paragraph.
Still, C11 and C++ atomics aren't widely implemented yet in the GPU world. OpenCL 1.2 doesn't have them (and OpenCL 2.0 is not commonly implemented), CUDA doesn't have them yet, AMD ROCm doesn't have them yet.
And anyone with older compilers will probably be working with fence intrinsics, instead of "innate" barriers per atomic instruction.
Still, C11 and C++ atomics aren't widely implemented yet in the GPU world. OpenCL 1.2 doesn't have them (and OpenCL 2.0 is not commonly implemented), CUDA doesn't have them yet, AMD ROCm doesn't have them yet.
And anyone with older compilers will probably be working with fence intrinsics, instead of "innate" barriers per atomic instruction.