Open maleadt opened 1 year ago
See https://github.com/JuliaConcurrent/Atomix.jl/tree/main/lib/AtomixCUDA and https://github.com/JuliaGPU/CUDA.jl/pull/1790
The alternative is that we rely on LLVM atomics, and lower them to something SPIR-V/OpenCL compatible in GPUCompiler.
See https://github.com/JuliaConcurrent/Atomix.jl/tree/main/lib/AtomixCUDA and https://github.com/JuliaGPU/CUDA.jl/pull/1790