[llvm-branch-commits] [llvm] [LoongArch] Split 256-bit build_vector to avoid using LASX element insertion (PR #154918)

[via llvm-branch-commits]

================
@@ -1911,9 +1911,8 @@ lowerVECTOR_SHUFFLE_XVREPLVEI(const SDLoc &DL, ArrayRef<int> Mask, MVT VT,
   if (fitsRegularPattern<int>(Begin, 1, End - HalfSize, SplatIndex, 0) &&
       fitsRegularPattern<int>(Begin + HalfSize, 1, End, SplatIndex + HalfSize,
                               0)) {
-    APInt Imm(64, SplatIndex);
     return DAG.getNode(LoongArchISD::VREPLVEI, DL, VT, V1,
-                       DAG.getConstant(Imm, DL, Subtarget.getGRLenVT()));
+                       DAG.getConstant(SplatIndex, DL, Subtarget.getGRLenVT()));
----------------
zhaoqi5 wrote:

Tests for la32 passed.

https://github.com/llvm/llvm-project/pull/154918