[llvm] [LoongArch] Optimize inserting extracted element for v4i64/v8i32 (PR #152629)

[via llvm-commits]

================
@@ -1640,6 +1640,28 @@ defm : PairInsertExtractPatV8<v8f32, f32>;
 defm : PairInsertExtractPatV4<v4i64, GRLenVT>;
 defm : PairInsertExtractPatV4<v4f64, f64>;
 
+foreach imm = 0...7 in {
----------------
tangaac wrote:

It's specially for extracting element from 0 index.
As you can see in tests.

https://github.com/llvm/llvm-project/pull/152629