[llvm] [VPlan] First step towards VPlan cost modeling (LegacyCM in CostCtx) (PR #92555)

[via llvm-commits]

================
@@ -1378,8 +1424,6 @@ class VPWidenCastRecipe : public VPRecipeWithIRFlags {
         ResultTy(ResultTy) {
     assert(UI.getOpcode() == Opcode &&
            "opcode of underlying cast doesn't match");
-    assert(UI.getType() == ResultTy &&
----------------
ayalz wrote:

If the original Trunc is used as Underlying instead of the original Ext, then its result type may match that of this recipe, but CM's cost may be less accurate?

https://github.com/llvm/llvm-project/pull/92555