On 8/27/19 1:01 PM, Richard Henderson wrote:
> Other constraints, such as rd != 15 or imod != 0, should continue to return
> false so that a (potential) grouped insn can match.
Eh. This is not the answer that the TT example suggests.
So far we are able to order the grouped insns such that
decoding directives like
if t == 15 then SEE "TT";
are respected. Since we do not generally do a very good
job of diagnosing all of the UNPREDICTABLE behavior, we
should not rely on getting all of it, e.g. by requiring
that if TT diagnoses some UNPRED that STREX also diagnoses
similar UNPRED.
I'm going to walk through the patch set and fix these.
r~
