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

[Krzysztof Drewniak]

================
@@ -241,6 +242,9 @@ func.func @conversion_broadcast(%in: vector<4xf8E5M2>, %scale: f8E8M0FNU) -> vec
 
 // -----
 
+// CHECK-GFX1100-LABEL: @conversion_scalar
+// CHECK-GFX1100: arith.scaling_extf
----------------
krzysz00 wrote:

(If there were an easy, low-dependency way to query the CPU features - that is, if we substantially refactored LLVM - these target checks could be checks on the same flags LLVM uses to test for the ultimate presence of the intrinsics. But that isn't, so chipset checks it is.)

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