Date: Thu, 16 Oct 2008 15:10:12 -0600 (MDT) From: "M. Warner Losh" <imp@bsdimp.com> To: scottl@samsco.org Cc: svn-src-head@FreeBSD.org, marius@FreeBSD.org, src-committers@FreeBSD.org, svn-src-all@FreeBSD.org Subject: Re: svn commit: r183896 - head/sys/dev/bge Message-ID: <20081016.151012.1933498518.imp@bsdimp.com> In-Reply-To: <48F5053D.7070705@samsco.org> References: <200810142028.m9EKShoL015514@svn.freebsd.org> <48F5053D.7070705@samsco.org>
next in thread | previous in thread | raw e-mail | index | archive | help
In message: <48F5053D.7070705@samsco.org>
Scott Long <scottl@samsco.org> writes:
: Marius Strobl wrote:
: > Author: marius
: > Date: Tue Oct 14 20:28:42 2008
: > New Revision: 183896
: > URL: http://svn.freebsd.org/changeset/base/183896
: >
: > Log:
: > Use bus_{read,write}_4(9) instead of bus_space_{read,write}_4(9)
: > in order to get rid of the bus space handle and tag in the softc.
: >
:
: Has anyone looked at the generated code from this interface switch and
: compared it what was getting generated previously? Way back when,
: including <machine/bus_memio.h> made bus_space_read|write_4() compile
: into a direct memory access on machines that supported it. The dubious
: removal of bus_memio.h and bus_pio.h took away that benefit, and I'm
: afraid that it's only getting worse now. Bus writes to card memory are
: still very important to high-performance devices and shouldn't be
: pessimized in the name of simpler-looking C code.
I've looked a little. With changes similar to this change, the
generated code does look better because there's less stack traffic,
but it still isn't a direct write to memory without a function call...
Warner
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20081016.151012.1933498518.imp>