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

Bug#789612: MIPS O32 default to FPXX



Matthias Klose <doko@debian.org> writes:
> On 06/29/2015 07:57 PM, YunQiang Su wrote:
> > On Mon, 22 Jun 2015 19:02:29 +0200 Geert Stappers <stappers@stappers.nl> wrote:
> >> Control: retitle -1 gcc-5-mips-32-fpxx
> >> Control: tag -1 patch
> >>
> >>
> >
> > Any consider of this patch?
> 
> Please could somebody clarify, if this bumps hardware requirements to newer CPU
> versions, or makes some hardware obsolete?  I know we still keep some longsoon2
> work arounds in binutils/gcc.  Is there anything more required, like rebuilding
> packages?

The transitional FP ABI extensions (FPXX, FP64A and finally FP64) are designed
around compatibility so that staged migration is possible. Moving to FPXX by
default is step1 to get packages away from FP32 which is the historic/current
default.

Moving just from FP32 to FPXX itself and keeping MIPS II as the ISA has
no negative impact on users, there would be a slight change in code generation
but the code would work on all systems it already works on and be compatible
with all pre-built packages.

As far as I can see the patch does not change the ISA but instead just modifies
the scheduling to use a more recent CPU (it will equate to a 4KC instead of an
R6000). There will be no MIPS32 instructions generated and the code will be
100% compatible with other pre-built packages.

With this change in place then I would hope you can move to binutils 2.25,
, glibc 2.21 and kernel headers 4.1 if this has not already happened. It is
important that glibc is rebuilt after all the other changes so that
all the new forwards compatibility features are in place. Other packages
can progress more leisurely but the more rebuilds that happen the more
compatible debian will be with new FPU features in modern MIPS processors. In
particular this paves the way for MSA SIMD to be usable in debian in the
future.

Hope I didn't ramble too much, it's a very big carefully designed plan.

Thanks,
Matthew

Reply to: