[llvm] [LV] Don't predicate divs with invariant divisor when folding tail (PR #98904)

[Florian Hahn via llvm-commits]

================
@@ -254,117 +227,9 @@ define void @udiv_urem_feeding_gep(i64 %x, ptr %dst, i64 %N) {
 ; CHECK-NEXT:  [[ENTRY:.*]]:
 ; CHECK-NEXT:    [[MUL_1_I:%.*]] = mul i64 [[X]], [[X]]
 ; CHECK-NEXT:    [[MUL_2_I:%.*]] = mul i64 [[MUL_1_I]], [[X]]
-; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N]], 1
-; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
----------------
fhahn wrote:

It looks this is down to a large difference in the costs of `udiv <4 x i64>` vs 4 * `udiv i64` + scalarization overhead estimate in LV (80 vs 7!). This looks like an X86 cost-model issue independent of the patch

https://github.com/llvm/llvm-project/pull/98904