Signed-off-by: Frans Pop <[email protected]>
---
Change was suggested by Sam Ravnborg; see http://lkml.org/lkml/2009/7/18/15.
Untested.
diff --git a/arch/ia64/Makefile b/arch/ia64/Makefile
index 58a7e46..2bf6107 100644
--- a/arch/ia64/Makefile
+++ b/arch/ia64/Makefile
@@ -41,10 +41,8 @@ $(error Sorry, you need a newer version of the assember, one that is built from
ftp://ftp.hpl.hp.com/pub/linux-ia64/gas-030124.tar.gz)
endif
-ifeq ($(call cc-version),0304)
- cflags-$(CONFIG_ITANIUM) += -mtune=merced
- cflags-$(CONFIG_MCKINLEY) += -mtune=mckinley
-endif
+cflags-$(CONFIG_ITANIUM) += $(call cc-ifversion, -eq, 0304, -mtune=merced)
+cflags-$(CONFIG_MCKINLEY) += $(call cc-ifversion, -eq, 0304, -mtune=mckinley)
KBUILD_CFLAGS += $(cflags-y)
head-y := arch/ia64/kernel/head.o arch/ia64/kernel/init_task.o
On Thu, Jul 23, 2009 at 08:56:50PM +0200, Frans Pop wrote:
> Signed-off-by: Frans Pop <[email protected]>
> ---
>
> Change was suggested by Sam Ravnborg; see http://lkml.org/lkml/2009/7/18/15.
> Untested.
>
> diff --git a/arch/ia64/Makefile b/arch/ia64/Makefile
> index 58a7e46..2bf6107 100644
> --- a/arch/ia64/Makefile
> +++ b/arch/ia64/Makefile
> @@ -41,10 +41,8 @@ $(error Sorry, you need a newer version of the assember, one that is built from
> ftp://ftp.hpl.hp.com/pub/linux-ia64/gas-030124.tar.gz)
> endif
>
> -ifeq ($(call cc-version),0304)
> - cflags-$(CONFIG_ITANIUM) += -mtune=merced
> - cflags-$(CONFIG_MCKINLEY) += -mtune=mckinley
> -endif
> +cflags-$(CONFIG_ITANIUM) += $(call cc-ifversion, -eq, 0304, -mtune=merced)
> +cflags-$(CONFIG_MCKINLEY) += $(call cc-ifversion, -eq, 0304, -mtune=mckinley)
The above change is correct.
But I really wonder if the original code was correct?
Do we really only want to use the -mtune options for this specific gcc version?
If this is indeed the case this deserves a comment explaning this.
So the patch has my:
Acked-by: Sam Ravnborg <[email protected]>
But the code that this patch highligt (but does not chage) looks questionable.
Sam
>>>>> "Sam" == Sam Ravnborg <[email protected]> writes:
Sam> On Thu, Jul 23, 2009 at 08:56:50PM +0200, Frans Pop wrote:
Sam> The above change is correct. But I really wonder if the original
Sam> code was correct? Do we really only want to use the -mtune
Sam> options for this specific gcc version? If this is indeed the
Sam> case this deserves a comment explaning this.
I suspect it should be all compilers after this one. -mtune=mckinley didn't
work very well in the early gcc 3 compilers and didn't exist in
version 2.
--
Dr Peter Chubb http://www.gelato.unsw.edu.au [email protected]
http://www.ertos.nicta.com.au ERTOS within National ICT Australia
On Friday 24 July 2009, Peter Chubb wrote:
> >>>>> "Sam" == Sam Ravnborg <[email protected]> writes:
> Sam> The above change is correct. But I really wonder if the original
> Sam> code was correct? Do we really only want to use the -mtune
> Sam> options for this specific gcc version? If this is indeed the
> Sam> case this deserves a comment explaning this.
>
> I suspect it should be all compilers after this one. -mtune=mckinley
> didn't work very well in the early gcc 3 compilers and didn't exist in
> version 2.
How would you like to handle that? As it is essentially a separate issue,
my suggestion would be: apply my patch as is and then (if needed) commit
a separate patch to fix the incorrect comparisons on top.
Cheers,
FJP
On Fri, Jul 24, 2009 at 11:23:11AM -0700, Frans Pop wrote:
> On Friday 24 July 2009, Peter Chubb wrote:
> > >>>>> "Sam" == Sam Ravnborg <[email protected]> writes:
> > Sam> The above change is correct. But I really wonder if the original
> > Sam> code was correct? Do we really only want to use the -mtune
> > Sam> options for this specific gcc version? If this is indeed the
> > Sam> case this deserves a comment explaning this.
> >
> > I suspect it should be all compilers after this one. -mtune=mckinley
> > didn't work very well in the early gcc 3 compilers and didn't exist in
> > version 2.
>
> How would you like to handle that? As it is essentially a separate issue,
> my suggestion would be: apply my patch as is and then (if needed) commit
> a separate patch to fix the incorrect comparisons on top.
Between GCC version 3.4.0 and 4.3.3 (including 3.4.0 and 4.3.3), -mtune=merced is
implemented in GCC. Starting from 4.4.0, -mtune=merced is deprecated.
Even implemented in versions between 3.4.0 and 4.3.3, the -mtune=merced
feature has been broken in some of the versions. For example, GCC 4.1.2 reports
interanl tuning function errors during kernel building with -mtune=merced. Or GCC
Bugzilla 16130 reports another -mtune=merced issue on GCC 3.4.1.
So I would remove the -mtune=merced from IA64 kernel build. Without this option,
kernel on Merced will remain the same except losing an unstable and out-of-date
performance tunning feature.
Since GCC version 3.4.0, -mtune=mckinley has been implemented. The
-mtune=mckinley option functions the same as mtune=itanium2. And mtune=itanium2
is the default option. So we don't need to add mtune=mckinley either since its
been the default option in any GCC version which implements this option.
Signed-off-by: Fenghua Yu <[email protected]>
---
arch/ia64/Makefile | 5 -----
1 files changed, 5 deletions(-)
diff --git a/arch/ia64/Makefile b/arch/ia64/Makefile
index 58a7e46..e7cbaa0 100644
--- a/arch/ia64/Makefile
+++ b/arch/ia64/Makefile
@@ -41,11 +41,6 @@ $(error Sorry, you need a newer version of the assember, one that is built from
ftp://ftp.hpl.hp.com/pub/linux-ia64/gas-030124.tar.gz)
endif
-ifeq ($(call cc-version),0304)
- cflags-$(CONFIG_ITANIUM) += -mtune=merced
- cflags-$(CONFIG_MCKINLEY) += -mtune=mckinley
-endif
-
KBUILD_CFLAGS += $(cflags-y)
head-y := arch/ia64/kernel/head.o arch/ia64/kernel/init_task.o
On Saturday 25 July 2009, Fenghua Yu wrote:
> Between GCC version 3.4.0 and 4.3.3 (including 3.4.0 and 4.3.3),
> -mtune=merced is implemented in GCC. Starting from 4.4.0, -mtune=merced
> is deprecated.
>
> Even implemented in versions between 3.4.0 and 4.3.3, the -mtune=merced
> feature has been broken in some of the versions. For example, GCC 4.1.2
> reports interanl tuning function errors during kernel building with
> -mtune=merced. Or GCC Bugzilla 16130 reports another -mtune=merced
> issue on GCC 3.4.1.
>
> So I would remove the -mtune=merced from IA64 kernel build. Without
> this option, kernel on Merced will remain the same except losing an
> unstable and out-of-date performance tunning feature.
>
> Since GCC version 3.4.0, -mtune=mckinley has been implemented. The
> -mtune=mckinley option functions the same as mtune=itanium2. And
> mtune=itanium2 is the default option. So we don't need to add
> mtune=mckinley either since its been the default option in any GCC
> version which implements this option.
>
> Signed-off-by: Fenghua Yu <[email protected]>
Fine by me, but you'll probably want to give the commit a different
subject :-)