[Mlir-commits] [mlir] [mlir][AMDGPU] Enable emulating vector buffer_atomic_fadd for bf16 on gfx942 (PR #129029)

[Krzysztof Drewniak]

================
@@ -189,6 +189,14 @@ void mlir::amdgpu::populateAmdgpuEmulateAtomicsPatterns(
     } else {
       target.addIllegalOp<RawBufferAtomicFmaxOp>();
     }
+    // gfx950 has bf16 atomics
+    if (chipset < Chipset(9, 5, 0)) {
----------------
krzysz00 wrote:

1. Yeah, especially since every downstream inventing their own thing here's a bit awkward
2. The keywords here are DLTI - Data Layout and Target Information - which is getting a decent amount of setup but doesn't seem to see much use
3. @rengolin 's a relevant person
4. In all honesty ... patches welcome on this

https://github.com/llvm/llvm-project/pull/129029