[Mlir-commits] [mlir] [mlir][ArithToAMDGPU] limit scaling truncf/extf support to gfx950 (PR #155431)

[Matt Arsenault]

================
@@ -241,6 +242,9 @@ func.func @conversion_broadcast(%in: vector<4xf8E5M2>, %scale: f8E8M0FNU) -> vec
 
 // -----
 
+// CHECK-GFX1100-LABEL: @conversion_scalar
+// CHECK-GFX1100: arith.scaling_extf
----------------
arsenm wrote:

I don't understand the operations involved here. Why does this need target knowledge, to emit something that isn't a target operation? Why isn't there target legalization? 

https://github.com/llvm/llvm-project/pull/155431