#3920 Ascon regression test issues

[Philipp Klaus Krause]

I'm about to add new regression tests asconaead128.c.in and asconhas256.c.in (in the pqc branch first). Each has four different implementations of the underlying cryptosystem. ref looks fine (apart from being very slow to compile and needing too much memory when inlining). But the opt32_lowsize (optimized for small code size and 32-bit targets) and bi8 (bit-interleaved version optimized for 8-bit targets) fail the test. And the opt8 (optimized for 8-bit targets) version doesn't even compile.
All are fine for test-host.

[Philipp Klaus Krause]

For the compilation failure, this small code sample reproduces issue:

char *p;

inline void ABSORB(unsigned char* s, const unsigned char* d) {
  *s++ ^= *d++;
}

void ascon_update(unsigned char* m) {
  ABSORB(p, m);
}

test.c:4: error 20: Undefined identifier '__100000000'
test.c:4: error 27: Pointer required

[Philipp Klaus Krause]

This turned out to be a quite complex issue. I've worked on it all day. The fix for the compilation failure (and a related issue that was masked by it) in in [r16072]. The runtime test failures still need to be addressed.

[Philipp Klaus Krause]

I suspect that some of the remaining failures are actually upstream endianness bugs: https://github.com/ascon/ascon-c/issues/25

[Gabriele Gorla]

On the mos6502 all implementations work except opt8 and opt8_lowsize