Support for 128-bit Floating-Point Math Library

The Kunpeng scenario supports up to 64-bit floating-point computation for hardware. The precision of floating-point computation greater than 64 bits depends on the precision of the software floating-point library. By default, the existing quadruple-precision floating-point math library in the environment is used. If necessary, you can run the --rtlib=compiler-rt command to use the quadruple-precision floating-point math library in the compiler-rt library.