[llvm] [AArch64][SVE] Rework VECTOR_COMPRESS lowering (PR #171162)

[Paul Walker via llvm-commits]

================
@@ -7504,7 +7491,8 @@ SDValue AArch64TargetLowering::LowerVECTOR_COMPRESS(SDValue Op,
       Vec);
 
   // compact fills with 0s, so if our passthru is all 0s, do nothing here.
-  if (HasPassthru && !ISD::isConstantSplatVectorAllZeros(Passthru.getNode())) {
+  if (!Passthru.isUndef() &&
+      !ISD::isConstantSplatVectorAllZeros(Passthru.getNode())) {
----------------
paulwalker-arm wrote:

Assuming removing the container logic works out, can this be an early exit?

https://github.com/llvm/llvm-project/pull/171162