[llvm] [llvm][ARM][CodeGen] Disable MEMCPY LDM/STM inlining for Cortex v7-m (PR #106378)

[David Green via llvm-commits]

================
@@ -138,6 +138,116 @@ SDValue ARMSelectionDAGInfo::EmitSpecializedLibcall(
   return CallResult.second;
 }
 
+SDValue ARMSelectionDAGInfo::EmitMemcpyAsLdSt(
+    SelectionDAG &DAG, SDLoc dl, const ARMSubtarget &Subtarget, SDValue Chain,
+    SDValue Dst, SDValue Src, uint64_t SizeVal, bool isVolatile,
+    MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
+  // Do repeated batches of 4-byte loads and stores.
+  unsigned BytesLeft = SizeVal & 3;
+  unsigned NumMemOps = SizeVal >> 2;
+  unsigned EmittedNumMemOps = 0;
+  EVT VT = MVT::i32;
+  unsigned VTSize = 4;
+  unsigned i = 0;
+  // Emit a maximum of 4 loads in Thumb1 since we have fewer registers
+  const unsigned MaxLoads = Subtarget.isThumb1Only() ? 4 : 6;
+  SDValue TFOps[6];
+  SDValue Loads[6];
----------------
davemgreen wrote:

Use a container type that will tell us if we overflow it. I think I would just use a SmallVector and push onto it.

https://github.com/llvm/llvm-project/pull/106378