c - Is it possible to use memory barriers only on the storing side -


First of all, some references: I am working with the pre-C11, inline-ASM-based nuclear model, But I am happy to ignore the C aspect (and any compiler blocking issues that I can deal with separately) and only consider it as an asm / cpu-architecture question.

Suppose I have a code that looks like:

  different store barrier store flag blocks   

I flag from another CPU core and concluded that different stores were already made and visible, it is possible to do this without any kind of memory barrier instructions loading side < / Em>? Obviously this is possible at least on some CPU architectures, for example x86 where there is no need for a clear memory hurdle either on the core or on the core. But what about general? Is it widely widespread by the CPU arc, is it possible?

"post-text" itemprop = "text">

If there is a CPU load again there will be a load constraint for your code to work correctly, there are lots of architectures Who do such reinvention; See the table for some examples.

In this type of normal code, your code requires load constraints.

The x86 is not very specific, guarantees a very strong memory sequence in it for discussion.

Comments

Popular posts from this blog

java - ImportError: No module named py4j.java_gateway -

python - Receiving "KeyError" after decoding json result from url -

.net - Creating a new Queue Manager and Queue in Websphere MQ (using C#) -