On 28.02.19 00:46, Richard Henderson wrote:
> On 2/26/19 3:39 AM, David Hildenbrand wrote:
>> +static DisasJumpType op_vst(DisasContext *s, DisasOps *o)
>> +{
>> + /*
>> + * FIXME: On exceptions we must not modify any memory.
>> + */
>> + store_vec_element(s, get_field(s->fields, v1), 0, o->addr1, MO_64);
>> + gen_addi_and_wrap_i64(s, o->addr1, o->addr1, 8);
>> + store_vec_element(s, get_field(s->fields, v1), 1, o->addr1, MO_64);
>> + return DISAS_NEXT;
>
> Should handle alignment though.
Again, as unaligned access must not trigger an exception, I don't think
we can use MO_ALIGN.
>
> FWIW, there is a probe_write function that can be called to make sure a region
> is writable before actually accessing it. But this is common enough that we
> should probably just handle 16-byte quantities as a native tcg type.
That would make most sense. It won't help for the VECTOR_STORE_MULTIPLE
part, though.
I'll keep it as is for now.
Thanks!
>
> Reviewed-by: Richard Henderson <richard.hender...@linaro.org>
>
>
> r~
>
--
Thanks,
David / dhildenb
