[AArch64][compiler-rt] floating point exception is not raised when comparing (long double)NAN by < operator

[utsumi-fj]

When investigating #59924, another issue was found. This occurs when compiling with --rtlib=compiler-rt.
The "invalid" floating point exception is not raised when comparing (long double)NAN by < operator as follows.

$ clang --version
clang version 16.0.0 (https://github.com/llvm/llvm-project.git 60b98979241151f8d6ec8af9ebc4da421f4a0258)
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /work03/lang/llvm/llvm/main/bin
$ cat test.c
#include <stdio.h>
#include <math.h>
#include <fenv.h>

int main() {
#pragma STDC FENV_ACCESS ON

  long double nan128 = (long double)NAN;
  if (nan128 < 1.0)
    puts("test");

  if (fetestexcept(FE_INVALID))
    puts("FE_INVALID");

  return 0;
}
$ clang test.c -lm
$ ./a.out
FE_INVALID
$ clang test.c -lm --rtlib=compiler-rt
$ ./a.out
$

It is expected that "FE_INVALID" is printed.

[llvmbot]

@llvm/issue-subscribers-backend-aarch64