On 11/16/2013 8:52 AM, Allan Jude wrote: > I see this was in the release notes for 9.0 and 8.3, but other than > that, I don't see how anyone was supposed to find out about this. > Maybe it would make sense to print 'Starting memory test, set > hw.memtest.test=0 to disable' before that starts, so anyone stuck > waiting will have a hint about what to do. It takes less effort to speed it up than to document that it is slow.. For now, in my sources, I've lengthened the testing stride to 64KB. A better fix would leave the cache on (don't fight the cache in a memory test - it is your friend! :-) and to group all writes and reads together in a larger chunk - say 16MB - so as to take advantage of write combining and cache line fetching. And add writes of address-specific values so address space aliasing can be detected.Received on Sat Nov 16 2013 - 17:45:24 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:44 UTC