Re: ringPointerTest

[Michael Davidsaver <mdavidsaver@bnl.gov>]

On 11/18/2014 05:48 PM, Andrew Johnson wrote:
> ...
> If a test needs a specific thread priority, please start a thread that
> meets those requirements and run it in that. The VxWorks shell by
> default runs at the highest priority.

Ok, I'll change this.

> Also I am a little nervous about future compilers noticing that you're
> creating pointers out of random integer offsets from a NULL pointer,

I thought you might be.  My goal is to catch corrupt reads/writes.
Simply reading through a bunch of sequential addresses means that it is
likely that only one byte is different.  This could mask corruption in
the other 3/7 bytes.

So the test does i|i<<16 so that I'm sure at least 2 bytes are changing
in every read/write.  So if I write 0x000a000a followed by 0x000b000b
but find that I read 0x000a000b, then I know there is a problem.

>
http://docs.cray.com/books/004-2179-001/html-004-2179-001/rvc5mrwh.html#ZFIXEDAWYU1A7W
> Could this program not just malloc() a large buffer (1MB say) and make
> up its pointers from offsets into that array?

This could be done.  I would want 16MB (i|i<<12 for i<4000).