[llvm] [Xtensa] Implement base CallConvention. (PR #83280)

[Andrei Safronov via llvm-commits]

================
@@ -238,6 +239,11 @@ def L32R : RI16_Inst<0x01, (outs AR:$t), (ins L32Rtarget:$label),
   let imm16 = label;
 }
 
+//extending loads
----------------
andreisfr wrote:

@s-barannikov , I prepared second commit with call lowering. Could you please take a look? We assume in the call lowering implementation that in most cases we should generate indirect call instruction and thus we need to load address constant from ConstantPool. So, I added base implementation of the ConstantPool to second commit. If it is not very complex to review both commits, I can merge them or drop second commit in other case.

https://github.com/llvm/llvm-project/pull/83280