[llvm] Allow MIN/MAX signedness flip when operands are known-negative (PR #174469)

[via llvm-commits]

================
@@ -66,6 +66,28 @@ define <16 x i8> @test_v16i8_nosignbit(<16 x i8> %a, <16 x i8> %b) {
   ret <16 x i8> %4
 }
 
+define <16 x i8> @test_v16i8_knownnegative(<16 x i8> %a, <16 x i8> %b) {
----------------
fbrv wrote:

thanks! Was not aware of this, sorry for all this forth and back, lesson learnt for future contributions

https://github.com/llvm/llvm-project/pull/174469