Re: [openib-general] [PATCH 39 of 39] IB/ipath - use streaming copy in RDMA interrupt handler to reduce packet loss

[Ralph Campbell <ralphc@pathscale.com>]

This is intended to be an architecture specific function
so if the CPU does support HW dma to the CPU's L2 cache, the
architecture specific version of memcpy_nc() would not replace
the default definition which maps memcpy_nc() to memcpy().

For CPUs like the vast majority currently available, there
is a performance benefit by not reading data into the cache
that won't be read a second time.

On Thu, 2006-06-29 at 16:46 -0700, David Miller wrote:
> From: Bryan O'Sullivan <bos@pathscale.com>
> Date: Thu, 29 Jun 2006 16:34:23 -0700
> 
> > I'm not quite following you, though I assume you're referring to Niagara
> > or Rock :-)  Are you saying a memcpy_nc would do worse than plain
> > memcpy, or worse than some other memcpy-like routine?
> 
> It would do worse than memcpy.
> 
> If you bypass the L2 cache, it's pointless because the next
> agent (PCI controller, CPU thread, etc.) is going to need the
> data in the L2 cache.
> 
> It's better in that kind of setup to eat the L2 cache miss overhead in
> memcpy since memcpy can usually prefetch and store buffer in order to
> absorb some of the L2 miss costs.
> 
> _______________________________________________
> openib-general mailing list
> openib-general@openib.org
> http://openib.org/mailman/listinfo/openib-general
> 
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
>