[RFC/PATCH] powerpc: Always use -mno-string & -mno-multiple

[RFC/PATCH] powerpc: Always use -mno-string & -mno-multiple

[Benjamin Herrenschmidt]

powerpc: Always use -mno-string and -mno-multiple

The string and multiple load/store instructions are causing more
damage than help on a lot of processors where they are microcoded
and are generally no good for the kernel to use. Let's make sure
we don't emit them on any powerpc variant.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---

Is that actually true ? What about 601 ? 4xx ? Do we want to keep
generating those for space on embedded processors where they actually
work (if they do) ?

Index: linux-cell/arch/powerpc/Makefile
===================================================================
--- linux-cell.orig/arch/powerpc/Makefile	2007-03-22 17:02:42.000000000 +1100
+++ linux-cell/arch/powerpc/Makefile	2007-03-22 17:05:23.000000000 +1100
@@ -102,9 +102,9 @@ CFLAGS += $(call cc-option,-mno-altivec)
 # kernel considerably.
 CFLAGS += $(call cc-option,-funit-at-a-time)
 
-ifndef CONFIG_FSL_BOOKE
-CFLAGS		+= -mstring
-endif
+# Never use string and multiple load/store instructions as they are
+# often slow when they are implemented at all
+CFLAGS		+= -mno-string -mno-multiple
 
 ifeq ($(CONFIG_6xx),y)
 CFLAGS		+= -mcpu=powerpc

Re: [RFC/PATCH] powerpc: Always use -mno-string & -mno-multiple

[Segher Boessenkool]

> powerpc: Always use -mno-string and -mno-multiple
>
> The string and multiple load/store instructions are causing more
> damage than help on a lot of processors where they are microcoded
> and are generally no good for the kernel to use. Let's make sure
> we don't emit them on any powerpc variant.

And they help a lot on many 32-bit CPUs.  NAK.

If you tell your GCC to compile for a certain CPU where
these insns would be very slow, GCC will automagically
not generate them, already.


Segher

Re: [RFC/PATCH] powerpc: Always use -mno-string & -mno-multiple

[Benjamin Herrenschmidt]

On Thu, 2007-03-22 at 13:02 +0100, Segher Boessenkool wrote:
> > powerpc: Always use -mno-string and -mno-multiple
> >
> > The string and multiple load/store instructions are causing more
> > damage than help on a lot of processors where they are microcoded
> > and are generally no good for the kernel to use. Let's make sure
> > we don't emit them on any powerpc variant.
> 
> And they help a lot on many 32-bit CPUs.  NAK.
> 
> If you tell your GCC to compile for a certain CPU where
> these insns would be very slow, GCC will automagically
> not generate them, already.

multiple do, strings really don't much except maybe on 601.... we don't
quite yet have the options to compile for specific CPUs on arch/powerpc
so -mno-string is a good idea generally, though I agree -mno-multiple
isn't, thus my next patch :-)

Ben.