tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: MI overrides of bus_dma(9), bus_space(9), pci(9)



I haven't considered David's proposal, however:

On Wed, Mar 10, 2010 at 11:35:42PM +0900, Izumi Tsutsui wrote:

> - How much performance impact is expected for bus_space(9) changes
>   on ports that use macro for access functions?

In the grand scheme of things, converting from macro to function proper
doesn't matter since devices are not regular memory.  Compared to cache
or main memory, access to them is very slow. 

Also, I note that when we converted to functions on x86, the kernels
dropped quite a bit in size.  I think it was 200kB or so.  The compiler
was getting "clever" and generating massive code path duplication to handle
the conditionals in the bus_space macros, with a minimal number of branches.

In that instance, ignoring any percived benefit of inline function "calling"
we had already shot ourselves in the foot by blowing all kinds of internal
caches on the processor.



Home | Main Index | Thread Index | Old Index