How is the arm memory model different from ia64?

I have to deal with a lot of multithreaded programming at work. Specifically, I want to know how the ARM memory model differs from IA64. Both are very weak models, but are there any specific differences th开发者_JAVA百科at I should know about? I'm familiar with ia64, not with arm.

I have the same question, and while I haven't completely found the answer, this paper might be useful: Reasoning about the ARM weakly consistent memory model

In the Further Work section, it seems to imply that no one (even at ARM, since they wrote the paper) has answered the question formally:

... there are several completeness questions we need to answer. How does the axiomatic model fit in with the operational one? ... And what is the precise embedding of the ARM model into Alpha, Intel, JMM?

Update: Also look at Memory Barriers: a Hardware View for Software Hackers by Paul McKenney. It discusses the types of memory access reordering that various processors do and what instructions are available to fence the accesses. In particular, note Table 5 on page 16 that summarizes the allowed reorderings for each architecture. ARMv7, IA64, POWER, and SPARC RMO all share the same characteristics, namely, that without a barrier, anything except dependent loads can be reordered.

http://en.wikipedia.org/wiki/Memory_ordering might answer your question. From the table it shows, it seems ARMv7 has the same as IA64.