We present the operational semantics of WMM, a Weak Memory Model. Our only
criteria for defining WMM is that single-threaded executions should be correct
without the use of any fence. Every possible reorderings are allowed except
that stores cannot overtake loads. We define the operational semantics of WMM
using 4 types of components: a) instantaneous execution core, b) instantaneous
memory, c) store buffer, and d) invalidation buffer. This provides a far
simpler operational model for weak memory models than the descriptions
involving various types of memory and program reorderings. We prove that these
operational semantics are exactly equivalent to program reordering semantics in
the presence of both memory dependency speculation and load-value speculation.
We thus show that WMM does not break even when the most aggressive
microarchitectural techniques involving out-of-order and speculative execution
are used.  Our preliminary performance evaluation using the SPLASH benchmarks
shows that WMM performs up to 18% and 33% better than the  most aggressive
implementations of TSO and SC, respectively. WMM holds the promise to be a
vendor-independent stable memory model which will not stifle microarchitectural
innovations, as it is a super-set of behaviors permitted by most memory model
specifications that are currently in use.