[llvm] [AMDGPU] Add getCmpSelInstrCost override to re-enable SimplifyCFG speculation for vector types (PR #208043)

[Frederik Harwath via llvm-commits]

================
@@ -357,17 +357,20 @@ define <4 x i16> @uadd_sat_v4i16(<4 x i16> %arg0, <4 x i16> %arg1) {
 ; GFX8-NEXT:  bb:
 ; GFX8-NEXT:    [[ARG0_0:%.*]] = extractelement <4 x i16> [[ARG0:%.*]], i64 0
 ; GFX8-NEXT:    [[ARG0_1:%.*]] = extractelement <4 x i16> [[ARG0]], i64 1
+; GFX8-NEXT:    [[ARG0_2:%.*]] = extractelement <4 x i16> [[ARG0]], i64 2
----------------
frederik-h wrote:

This change is from changing Extract to true in the scalarization overhead computation.

https://github.com/llvm/llvm-project/pull/208043