[Bug middle-end/70140] Inefficient expansion of __builtin_mempcpy

["wilco at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70140

--- Comment #9 from Wilco <wilco at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #8)
> (In reply to Wilco from comment #7)
> > (In reply to Martin Liška from comment #6)
> > > Created attachment 41772 [details]
> > > Patch candidate
> > > 
> > > I'm going to prepare some test-cases for that. Does it look good?
> > 
> > Yes, it now inlines small constant sizes. However large and variable sized
> > copies have the wrong return value:
> > 
> > void *f1(void *p, void *q) { return __builtin_mempcpy(p, q, 256); }
> > 
> > f1:
> > 	mov	x2, 256
> > 	b	memcpy
> 
> Yep, I've noticed. It's strange for me why it's not working. I've just asked
> at GCC ML: https://gcc.gnu.org/ml/gcc/2017-07/msg00144.html

It's marked as a tailcall so anything you generate afterwards will be ignored:

(call_insn/j 13 12 14 2 (parallel [
            (set (reg:DI 0 x0)
                (call (mem:DI (symbol_ref:DI ("memcpy") [flags 0x41]
<function_decl 0xffffb7acc700 __builtin_memcpy>) [0 __builtin_memcpy S8 A8])
                    (const_int 0 [0])))
            (return)
        ]) "mempcpy.c":3 -1

Also check this case:

void f4(void *p, void *q, int i) { __builtin_mempcpy(p, q, i); }