[llvm] [RISCV] Handle codegen for Big Endian (PR #172668)

[Djordje Todorovic via llvm-commits]

================
@@ -22571,16 +22598,32 @@ static MachineBasicBlock *emitBuildPairF64Pseudo(MachineInstr &MI,
       MF.getMachineMemOperand(MPI, MachineMemOperand::MOStore, 4, Align(8));
   MachineMemOperand *MMOHi = MF.getMachineMemOperand(
       MPI.getWithOffset(4), MachineMemOperand::MOStore, 4, Align(8));
-  BuildMI(*BB, MI, DL, TII.get(RISCV::SW))
-      .addReg(LoReg, getKillRegState(MI.getOperand(1).isKill()))
-      .addFrameIndex(FI)
-      .addImm(0)
-      .addMemOperand(MMOLo);
-  BuildMI(*BB, MI, DL, TII.get(RISCV::SW))
-      .addReg(HiReg, getKillRegState(MI.getOperand(2).isKill()))
-      .addFrameIndex(FI)
-      .addImm(4)
-      .addMemOperand(MMOHi);
+
+  // For big-endian, store the high part at offset 0 and the low part at
+  // offset 4.
+  if (!Subtarget.isLittleEndian()) {
----------------
djtodoro wrote:

Yes, thanks for the suggestion.

https://github.com/llvm/llvm-project/pull/172668