> Am 09.11.2018 um 18:30 schrieb Ulrich Weigand :
>
> Ilya Leoshkevich wrote:
>
>> gcc_assert (GET_CODE (x) != SYMBOL_REF
>> - || !CONSTANT_POOL_ADDRESS_P (x));
>> + || !CONSTANT_POOL_ADDRESS_P (x)
>> + || s390_symbol_relative_long_p (x));
>
> Hmm, it's a bit weird t
Ilya Leoshkevich wrote:
>gcc_assert (GET_CODE (x) != SYMBOL_REF
> - || !CONSTANT_POOL_ADDRESS_P (x));
> + || !CONSTANT_POOL_ADDRESS_P (x)
> + || s390_symbol_relative_long_p (x));
Hmm, it's a bit weird that this routine now uses a different check
than the other tw
Bootstrapped and regtested on s390x-redhat-linux.
r265490 allowed the compiler to choose in a more flexible way whether to
use load or load-address-relative-long (LARL) instruction. When it
chose LARL for literal pool references, the latter ones were rewritten
by pass_s390_early_mach to use UNSPE
