Yes, that assertion is not good, but I actually think it should be
completely removed, because ARM64 is always dual-number and it would
never fail.

Anyway, that's not where I noticed an issue, I was getting a failure
here: lj_state.c:172. This is the example I was running:

do
  -- IR_HLOAD: nil & pri
  local m = { x = nil, y = true }
  for i=1,100 do
    if not m.x then
      m.y = not m.y
    end
  end
  assert(m.y)
end

On Thu, 14 Jul 2016 09:29:53 +0100
Charles Baylis <charles.baylis@linaro.org> wrote:

> On 13 July 2016 at 14:07, Stefan Pejic <stefan.pejic@rt-rk.com> wrote:
> > Hello,
> >
> > I ran into an issue with asm_intcomp. Is anyone working on this at
> > the moment?
>
> I'm not looking at that.
>
> > Also, I noticed that enabling assertions in src/Makefile causes some
> > tests failures. And that's happening only after the latest merges.
>
> I think the assert in asm_sload:
>     lua_assert(!irt_isint(t) || (ir->op2 &
> (IRSLOAD_CONVERT|IRSLOAD_FRAME))); should be
>     lua_assert(LJ_DUALNUM || !irt_isint(t) || (ir->op2 &
> (IRSLOAD_CONVERT|IRSLOAD_FRAME)));
>
> which would match the definition in lj_asm_x86.h. Does that fix the
> problem you see?
>
> Charles
_______________________________________________
luajit mailing list
luajit@lists.linaro.org
https://lists.linaro.org/mailman/listinfo/luajit