Re: [Tinycc-devel] [PATCH] When handling '.' operator, cope with VT_LLOCAL

[Michael Matz]

Hi,

On Sun, 22 Feb 2015, Edmund Grimley Evans wrote:

> > > The only disadvantage that I can think of is that certain
> > > expressions which would cause GCC to warn that "value computed is
> > > not used" would not get optimised.
> >
> > I think the conversion to rvalue might not actually have to emit
> > code, in which case unused stuff would be as optimized as before (as
> > far as tcc is optimizing :)).
>
> I can't see how to do that.

Hmm, perhaps it's not possible.  I thought via manipulating the VT_foo 
flags in vtop->r.

> > > Since I probably still have a test case to hand, should I propose
> > > an alternative patch that fixes my problem by modifying gen_op?
> >
> > Yeah, I think so.
>
> So here's an alternative patch that fixes the problem for me:
>
> diff --git a/tccgen.c b/tccgen.c
> index ae07563..9dfeb2b 100644
> --- a/tccgen.c
> +++ b/tccgen.c
> @@ -1827,6 +1827,8 @@ ST_FUNC void gen_op(int op)
>             vtop->type.t = t;
>         }
>     }
> +    if (vtop->r & VT_LVAL)
> +        gv(is_float(vtop->type.t & VT_BTYPE) ? RC_FLOAT : RC_INT);
> }
>
> #ifndef TCC_TARGET_ARM
>
> It probably doesn't change the generated code significantly, though I
> should perhaps warn that apart from the cases I alluded to earlier,
> where a value computed is not used, there may also be a few cases
> where converting the value to an rvalue earlier than it is needed
> hurts the register allocation, as in something like f(a + 0, b + 0, c
> + 0, d + 0, e + 0, f + 0). I wouldn't worry about it, though.
>
> I've just noticed that the patch above doesn't have the comment I
> wrote, something like:
>
> +    // Make sure that we have converted to an rvalue:
>
> Do you like this new patch?
>
> I need to fix this problem somehow for arm64.

Now that I also have a working aarch64 tcc, do you have a testcase for the 
problem?  The testsuite itself is clean.


Ciao,
Michael.