This patch fix bug left in D124503.
We should do
sub(add(X,Z),umin(Y,Z)) --> add(X,usub.sat(Z,Y))
instead of
sub(add(X,Z),umin(Y,Z)) --> add(X,usub.sat(Y,Z)).
I forget to update sub-minmax.ll in D124503,
[[TMP1:%.*]] = call i8 @llvm.usub.sat.i8(i8 [[Y:%.*]], i8 [[Z:%.*]]) and
[[TMP1:%.*]] = call i8 @llvm.usub.sat.i8(i8 [[Z:%.*]], i8 [[Y:%.*]]) can both pass llvm-lit test.