On Tue, Mar 24, 2015 at 05:46:57PM +0000, Alan Lawrence wrote:
> Hmmm. This is not the right fix: the tests Richard fixed, were failing because
> of lack of constant propagation and DCE at compile-time, which then didn't
> eliminate the call to link_error. The AArch64 test is failing because this
> from
> aarch64/constraints.md:
>
> (define_constraint "S"
> "A constraint that matches an absolute symbolic address."
> (and (match_code "const,symbol_ref,label_ref")
> (match_test "aarch64_symbolic_address_p (op)")))
>
> previously was seeing (and being satisfied by):
>
> (const:DI (plus:DI (symbol_ref:DI ("test") [flags 0x3] <function_decl
> 0x7fb7c60300 test>)
> (const_int 4 [0x4])))
>
> but following Richard's patch the constraint is evaluated only on:
>
> (reg/f:DI 73 [ D.2670 ])
I don't think we should get too concerned by this. There are a number
of other constraints which we define which we can only satisfy given
a level of optimisation. Take the I (immediate acceptable for an ADD
instruction) constraint, which will fail for:
int foo (int x)
{
int z = 5;
__asm__ ("xxx %0 %1":"=r"(x) : "I"(z));
return x;
}
at O0 and happily produce:
xxx x0 5
with optimisations.
I think your original patch to add -O is just fine, but Marcus or
Richard will need to approve it.
Cheers,
James
