[libc] [libcxxabi] [clang-tools-extra] [openmp] [mlir] [clang] [flang] [compiler-rt] [libcxx] [llvm] [AArch64] Combine store (trunc X to <3 x i8>) to sequence of ST1.b. (PR #78637)

[Eli Friedman via llvm-commits]

================
@@ -21471,6 +21471,57 @@ bool isHalvingTruncateOfLegalScalableType(EVT SrcVT, EVT DstVT) {
          (SrcVT == MVT::nxv2i64 && DstVT == MVT::nxv2i32);
 }
 
+// Combine store (trunc X to <3 x i8>) to sequence of ST1.b.
+static SDValue combineI8TruncStore(StoreSDNode *ST, SelectionDAG &DAG,
+                                   const AArch64Subtarget *Subtarget) {
+  SDValue Value = ST->getValue();
+  EVT ValueVT = Value.getValueType();
+
+  if (ST->isVolatile() || !Subtarget->isLittleEndian() ||
+      ST->getOriginalAlign() >= 4 || Value.getOpcode() != ISD::TRUNCATE ||
----------------
efriedma-quic wrote:

What is `ST->getOriginalAlign() >= 4` protecting against?  Increasing the known alignment of the pointer doesn't change the generated code, as far as I can tell.



https://github.com/llvm/llvm-project/pull/78637