Oleg Kalnichevski wrote:
sebb wrote:
For example, I used to work on Alpha systems, and writes to the main
memory from one CPU could be seen in a different order on another CPU.
[I understand this was done for performance reasons.]
So a thread on one CPU only sees the correct data if both CPUs issue
the appropriate memory barrier instructions in the correct order.
I only have experience programming in x86 assembly, so I am not in a
position to judge. However, this is indeed the case, we should drop any
pretense those classes could be used by multiple threads and simply
declare them thread-unsafe.
Are instances of these classes limited to single threads in the current
codebase? Does that need to be reviewed to be confirmed?
Is there any foreseeable need to have instances of these classes shared
between threads?
Mike
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
