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

[Kunwar Grover]

================
@@ -702,7 +702,7 @@ void mlir::arith::populateArithToAMDGPUConversionPatterns(
   if (allowPackedF16Rtz)
     patterns.add<TruncfToFloat16RewritePattern>(patterns.getContext(), benefit);
 
-  if (chipset >= kGfx950) {
+  if (chipset == kGfx950) {
----------------
Groverkss wrote:

I would hoist this out to variable "supportsPackedExtTrunc"

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