This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH], PowerPC support to enable -mlra and/or -mfloat128


On Mon, Jul 11, 2016 at 10:07 PM, Michael Meissner
<meissner@linux.vnet.ibm.com> wrote:
> These configuration switches will allow the PowerPC GCC developers to switch
> defaults in the compiler to debug the code, before making the decision to flip
> the default permanently.  In the future, when the defaults have been changed,
> these configuration options would allow developers to go back to the previous
> versions without modifying the code using the --disable-<switch> form.
>
> The first option is --enable-lra, which changes the compiler so that the
> default is to use the LRA register allocator instead of the older RELOAD
> allocator. The PowerPC team would like to switch the default, but there is a
> critical bug in LRA that must be fixed before we can change the default:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69847
>
> The second option is --enable-float128, which changes the compiler so that the
> default for VSX systems is to enable the __float128 keyword to allow users
> access to the IEEE 128-bit floating point implementation without having to use
> the keyword.
>
> Both of these switches are debug switches, and are not meant to be used by
> non-developers.
>
> The --enable-lra swich causes the following tests to fail:
>
>     * testsuite/gcc.target/powerpc/bool3-p7.c
>     * testsuite/gcc.target/powerpc/bool3-p8.c
>
> See bug 71846 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71846) for more
> details.
>
> The --enable-float128 switch causes libquadmath and libstdc++ to fail because
> we do not yet have enough of the support in the compiler to allow these
> libraries to build.  It is our intention, that we will use the
> --enable-float128 option and work on getting the libraries fixed.  If I build
> just a C compiler and disable building libquadmath, there are no regressions in
> the C tests with __float128 enabled or disabled.
>
> Can I check these options into the trunk as non-default options?

Instead of adding more configury can we please enable LRA on trunk by default
_now_?  Otherwise the amount of testing it recieves won't really increase.

Richard.

>
> --
> Michael Meissner, IBM
> IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
> email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
>


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]