Store-buffering litmus test on x86 with no fences

Two hardware threads run raw machine code with no locks, atomics, or memory barriers of any kind, just plain loads and stores. X and Y both start at 0. Thread 0 stores X=1, then loads r1=Y. Thread 1 stores Y=1, then loads r2=X. Is the outcome r1==0 && r2==0 (each thread's load reads the other thread's variable before that thread's store is globally visible) actually observable on real x86 hardware, given that x86 uses the comparatively strong TSO (total store order) memory model?

12 of 47 in Computer Architecture