Re: [Tinycc-devel] ARM codegen issue, possibly fixed already

[Harald van Dijk]

On 14/07/2019 21:31, grischka wrote:
> Harald van Dijk wrote:
> > As that commit is an optimization, not a codegen fix, I would like to
> > ask whether the original issue is still present and just requires a more
> > complex test case to expose, or whether it has accidentally been fixed
> > by this commit.
>
> No, I think this bug must exist since the very early days of
> the project, on all 32-bit platforms.  This might fix it:

That's interesting: I had not seen it on i386, so I figured it would be 
ARM-specific.

> --- a/tccgen.c
> +++ b/tccgen.c
> @@ -1990,6 +1990,7 @@ static void gen_opl(int op)
>           vtop[-1] = vtop[-2];
>           vtop[-2] = tmp;
>           /* stack: L1 L2 H1 H2 */
> +        save_regs(4);
>           /* compare high */
>           op1 = op;
>           /* when values are equal, we need to compare low words. since

That looks good for the reduced test case; looking at the generated 
code, the changes make sense. I'll do some more testing over the course 
of the next week, thanks.

> > Thanks and keep up the good work!
> >
> > Cheers,
> > Harald van Dijk