benchmarking with and without "register" storage class specifier
Michael S. Zick
mszick at morethan.org
Wed Mar 29 05:15:09 PST 2006
On Wed March 29 2006 06:54, Robert P. J. Day wrote:
RP>
RP> i did several runs of:
RP>
RP> $ make distclean defconfig
RP> $ time make busybox
RP>
RP> both before and after applying my two earlier submitted patches that
RP> removed a majority of the "register" storage class specifiers, and the
RP> results are indistinguishable in terms of build time.
RP>
Robert,
Those register storage class specifiers may be left-overs from the days
of gcc-2.9x Those compilers had problems dealing with register pressure.
In that compiler, they had a much greater influence than they do in the
series 3 or series 4 gcc compilers.
Of course, it also depends on cpu. If testing on x86, register pressure
will make the register storage class specification a waste.
Testing on machines with a larger set of registers should show a difference,
at least in the generated code.
May also make a difference with some vendor (non-gcc) compilers.
Mike
More information about the busybox
mailing list