> I was going to put O3 back to the old way and then the workaround in > inorder. > > This problem all centers around marking the hwrei instruction w/the > isControl flag.
Seems to me that it would make more sense to look for the isControl flag in O3 and ignore it if the instruction is hwrei. My guess is that is where the issue is anyway and since O3 is the problem, it seems that O3 should get the workaround. All that said, if you look at the way that hwrei works, it sets an IPR with the value of the PC. This should not be a speculative operation because IPRs are not renamed. Our palcode is designed for an EV5 which is a nonspeculative machine, so there is no special sauce in the palcode to avoid the speculation. Currently, the instruction is marked IsSerializing and IsSerializeBefore, but shouldn't it really be IsNonSpeculative? Nate _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
