[PATCH] arm64: fpsimd: avoid restoring fpcr if the contents haven't changed
Will Deacon
will.deacon at arm.com
Fri Jun 27 01:58:45 PDT 2014
On Thu, Jun 26, 2014 at 10:10:00PM +0100, Ard Biesheuvel wrote:
> Hey Will,
Hi Ard,
> On 26 June 2014 19:34, Will Deacon <will.deacon at arm.com> wrote:
> > Writing to the FPCR is commonly implemented as a self-synchronising
> > operation in the CPU, so avoid writing to the register when the saved
> > value matches that in the hardware already.
> >
> > Reviewed-by: Lorenzo Pieralisi <lorenzo.pieralisi at arm.com>
> > Signed-off-by: Will Deacon <will.deacon at arm.com>
> > ---
> > arch/arm64/include/asm/fpsimdmacros.h | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/arch/arm64/include/asm/fpsimdmacros.h b/arch/arm64/include/asm/fpsimdmacros.h
> > index 768414d55e64..137c0f45b0a8 100644
> > --- a/arch/arm64/include/asm/fpsimdmacros.h
> > +++ b/arch/arm64/include/asm/fpsimdmacros.h
> > @@ -40,6 +40,7 @@
> > str w\tmpnr, [\state, #16 * 2 + 4]
> > .endm
> >
> > +/* Clobbers \state */
> > .macro fpsimd_restore state, tmpnr
> > ldp q0, q1, [\state, #16 * 0]
> > ldp q2, q3, [\state, #16 * 2]
> > @@ -60,7 +61,15 @@
> > ldr w\tmpnr, [\state, #16 * 2]
> > msr fpsr, x\tmpnr
> > ldr w\tmpnr, [\state, #16 * 2 + 4]
> > + /*
> > + * Writes to fpcr may be self-synchronising, so avoid restoring
> > + * the register if it hasn't changed.
> > + */
> > + mrs \state, fpcr
> > + cmp x\tmpnr, \state
> > + b.eq 9999f
> > msr fpcr, x\tmpnr
> > +9999:
> > .endm
> >
>
> I would assume the same would apply to the partial restore that was
> merged during the last window?
Good point; I'll take a look.
Cheers,
Will
More information about the linux-arm-kernel
mailing list