[OpenMP] Use a memory fence before incrementing the dispatch buffer index

[xingxue-ibm]

This patch uses a memory fence in function __kmp_dispatch_next() to flush pending memory write invalidates before incrementing the volatile variable buffer_index to fix intermittent time-outs of OpenMP runtime LIT test cases env/kmp_set_dispatch_buf.c and worksharing/for/kmp_set_dispatch_buf.c, noting that the same is needed for incrementing buffer_index in function __kmpc_next_section() (line 2600 of kmp_dispatch.cpp).

[kkwli]

LGTM

[jprotze]

It is not clear to me what effect the memory barrier is supposed to have at this specific place with the oder two barriers a few lines above and below.

From my perspective the code shouldn't use volatile at all, but atomics where necessary.

[xingxue-ibm]

It is not clear to me what effect the memory barrier is supposed to have at this specific place with the oder two barriers a few lines above and below.

From my perspective the code shouldn't use volatile at all, but atomics where necessary.

Yeah, atomics probably works better and it is a good candidate for the future improvement.