On a system with at least two NUMA nodes - will an accessing cache-lines from another NUMA node (through QPI) abort a RTM/HLE transaction?
I can't find a hint in the Intel documentation.
AFAIK, TSX uses the cc-snoop protocols so I would assume that the snoop protocol messages are not passed through the QPI.
With which error is the transaction aborted? With capacity error or memory confilct ...?
For more complete information about compiler optimizations, see our Optimization Notice.