On Fri, Aug 11, 2017 at 12:46 PM, Richard Henderson
> On 08/11/2017 11:19 AM, Alistair Francis wrote:
>> The exclusive store operation should return 0 if the operation updates
>> memory and 1 if it doesn't. This means that storing tmp in the rd
>> register is incorrect.
> I'm confused as to what you believe is wrong.
>> tcg_gen_atomic_cmpxchg_i64(tmp, addr, val, tmp,
>> - size | MO_ALIGN | s->be_data);
>> - tcg_gen_setcond_i64(TCG_COND_NE, tmp, tmp, val);
> Yes, we load the result of the cmpxchg into tmp, but then we immediately
> overwrite tmp with 0/1 depending on whether the operation succeeded.
Hmmm... When I looked at the values in tmp I was seeing non 0/1 values in there.
>> This patch updates the succesful opertion to store 0 into the rd
>> register instead of tmp. It also adds a branch to fail if the memory
>> isn't updated.
> Since we use setcond, a branch is not required.
>> + tcg_gen_ext_i64(val, val, memop);
> What is this addition intended to accomplish? Because of the position within
> the code, you know that memop contains MO_64, so that this is a no-op.
This is when size == 2 so it's a 32bit operation so memop contains MO_32.