Failed to build A kernel built on other implementatioins but not on NVIDIA

Dear all, I am getting this after I tried to clBuildProgram a program

param = 0x000000000020e380 "Error: Cannot yet select: 0xb30b9d0: i32,ch = AtomicLoadAdd 0xb346f50:1, 0xbd75440, 0xb346f50<Volatile LDST4[%def_localTotaldistributionCount]> [ORD=1438] [ID=15]

I don’t see this problem on other implementations.

Any idea?

Likely some problem with the SDK. Keep in mind the SDK is newish so its likely to have a few kinks in it.