Intel® ISA Extensions
Use hardware-based isolation and memory encryption to provide more code protection in your solutions.
Announcements
Welcome to the Intel Community. If you get an answer you like, please mark it as an Accepted Solution to help others. Thank you!
For the latest information on Intel’s response to the Log4j/Log4Shell vulnerability, please see Intel-SA-00646

TILESTORED to overlapping addresses

Beulich__Jan
Beginner
423 Views

Could it be clarified please whether TILESTORED behaves similarly to V{,P}SCATTER* in this regard, and if not, what the designated behavior is in such a case? Thanks.

0 Kudos
2 Replies
bval
Employee
301 Views

The accesses in TILESTORED are as defined in the EAS : the rows will not go out of order and the dwords in the row will not go out of order. I think this makes the definition of overlap relatively obvious and same as a scatter : the oldest store will merge into the previous store. But if you have an example that helps me understand the concern?

Beulich__Jan
Beginner
291 Views

Quoting from the SDM for scatters: "If two or more destination indices completely overlap, the “earlier” write(s) may be skipped." To me, you saying "will be merged" is not the same as "will be skipped". I understand that generally overlapping stores (from TILESTORED or any other insn) will be merged. Fully skipping (part of) a store is a different thing, though, and then also raises questions towards exception behavior.

Reply