Code generation issue for loongarch64 compiler when using -O2 or higher optimization

Summary

Loongarch64 compiler generates faulty code when -O2 optimization option or higher is used.

System Information

• Operating system: Linux
• Processor architecture: Loongarch64
• Compiler version: 3.3.1

See failures for tcalpext.pp:

https://www.freepascal.org/testsuite/cgi-bin/new-testsuite2.cgi?action=4&testfilename=test%2Fcg%2Ftcalpext.pp&os=15&cpu=8&version=58&date=&submitter=&machine=&comment=&limit=50&cond=&Category=3&failedonly=1&noskipped=1&listall=1

Steps to reproduce

Just compile required files for tests/test/cg/tcalpext.pp and compile using ppcloongarch64 -O2

Example Project

What is the current bug behavior?

The code fails on function test_function_tiny_struct. which returns a 8-bit value. It seems that, with -O2, the upper part of register $a0 is not zeroed out. I do not know if this is an error in the function or in the code after which seems to assume that upper bytes of the register are zero.