Re: Pentium M config option for 2.6

Re: Pentium M config option for 2.6

[Mikael Pettersson]

On Sun, 4 Jan 2004 03:28:48 +0100, Tomas Szepe wrote:
>Since the Pentium M has 64 byte cache lines and is not a K7 or K8...  ;)
...
>--- a/arch/i386/Makefile	2003-09-28 11:38:05.000000000 +0200
>+++ b/arch/i386/Makefile	2004-01-04 03:02:52.000000000 +0100
>@@ -35,6 +35,7 @@
> cflags-$(CONFIG_MPENTIUMII)	+= $(call check_gcc,-march=pentium2,-march=i686)
> cflags-$(CONFIG_MPENTIUMIII)	+= $(call check_gcc,-march=pentium3,-march=i686)
> cflags-$(CONFIG_MPENTIUM4)	+= $(call check_gcc,-march=pentium4,-march=i686)
>+cflags-$(CONFIG_MPENTIUMM)	+= $(call check_gcc,-march=pentium4,-march=i686)
> cflags-$(CONFIG_MK6)		+= $(call check_gcc,-march=k6,-march=i586)
> # Please note, that patches that add -march=athlon-xp and friends are pointless.
> # They make zero difference whatsosever to performance at this time.

P-M is not a P4 core, it's an enhanced PIII core.
SSE2 was added, but compiler support for SSE2 f.p.
math shouldn't matter for the kernel.

Using P4 optimisations on a P-M may actually reduce
performance, due to the different micro-architectures.
(P4 made shifts and some leas more expensive, and
simple add/and/sub/etc less expensive.)

IOW, don't lie to the compiler and pretend P-M == P4
with that -march=pentium4.

And since P-M doesn't do SMP, does cache line size even
matter? There are no locks to protect from ping-ponging.

/Mikael

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 13:27 +0100
Mikael Pettersson <mikpe@csd.uu.se> wrote:

> IOW, don't lie to the compiler and pretend P-M == P4
> with that -march=pentium4.

What do you recommend to use as march then?  There is
no pentiumm subarch support in gcc yet;  I was convinced
p4 was the closest match.

-- 
Tomas Szepe <szepe@pinerecords.com>

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 07:27, Mikael Pettersson wrote:


> And since P-M doesn't do SMP, does cache line size even
> matter? There are no locks to protect from ping-ponging.

Cache line size does still come into the picture on UP, albeit not as
much as with SMP - but e.g. it still matters to things like device
drivers doing DMA.

	Rob Love

Re: Pentium M config option for 2.6

[Dave Jones]

On Sun, Jan 04, 2004 at 10:03:28AM -0500, Rob Love wrote:
 > On Sun, 2004-01-04 at 07:27, Mikael Pettersson wrote:
 > > And since P-M doesn't do SMP, does cache line size even
 > > matter? There are no locks to protect from ping-ponging.
 > 
 > Cache line size does still come into the picture on UP, albeit not as
 > much as with SMP - but e.g. it still matters to things like device
 > drivers doing DMA.

Regardless, Tomas's patch changed CONFIG_X86_L1_CACHE_SHIFT for
that CPU, and CONFIG_X86_L1_CACHE_SHIFT shouldn't affect this.
The cacheline size is determined at boottime using the code in
pcibios_init() and set using pci_generic_prep_mwi().

The config option is the default that pci_cache_line_size starts at,
but this gets overridden when the CPU type is determined.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 11:25, Dave Jones wrote:

> Regardless, Tomas's patch changed CONFIG_X86_L1_CACHE_SHIFT for
> that CPU, and CONFIG_X86_L1_CACHE_SHIFT shouldn't affect this.
> The cacheline size is determined at boottime using the code in
> pcibios_init() and set using pci_generic_prep_mwi().
> 
> The config option is the default that pci_cache_line_size starts at,
> but this gets overridden when the CPU type is determined.

Yah.  I was just answering in the abstract to the "does cache line
matter on non-SMP" question.

I actually like this patch (perhaps since I have a P-M :) and think it
ought to go in, although I agree with others that the P-M is more of a
super-P3 than a scaled down P4.

	Rob Love

Re: Pentium M config option for 2.6

[Dave Jones]

On Sun, Jan 04, 2004 at 11:33:08AM -0500, Rob Love wrote:

 > Yah.  I was just answering in the abstract to the "does cache line
 > matter on non-SMP" question.
 > 
 > I actually like this patch (perhaps since I have a P-M :) and think it
 > ought to go in, although I agree with others that the P-M is more of a
 > super-P3 than a scaled down P4.

FWIW, I agree with it too on the grounds that its non obvious the optimal
setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
helptext to read...

 "Pentium II"
 "Pentium III / Pentium 4M"
 "Pentium 4"

My other mail may have sounded like I objected to the patch per se, I don't.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 11:50, Dave Jones wrote:

> FWIW, I agree with it too on the grounds that its non obvious the optimal
> setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
> helptext to read...
> 
>  "Pentium II"
>  "Pentium III / Pentium 4M"
>  "Pentium 4"

Oh, very much agreed.  Giving it a separate configure option also opens
the door for easily adding an march=pentiumm whenever the gcc folks get
around to adding that.

> My other mail may have sounded like I objected to the patch per se, I don't.

I did not think that, I thought perhaps that you thought that I objected
:)

	Rob Love

Re: Pentium M config option for 2.6

[Troels Walsted Hansen]

Dave Jones wrote:

>FWIW, I agree with it too on the grounds that its non obvious the optimal
>setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
>helptext to read...
>
> "Pentium II"
> "Pentium III / Pentium 4M"
> "Pentium 4"
>  
>
Especially since Pentium M != Pentium 4M :-)

Troels

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 12:01 -0500
Rob Love <rml@ximian.com> wrote:

> On Sun, 2004-01-04 at 11:50, Dave Jones wrote:
> 
> > FWIW, I agree with it too on the grounds that its non obvious the optimal
> > setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
> > helptext to read...
> > 
> >  "Pentium II"
> >  "Pentium III / Pentium 4M"
> >  "Pentium 4"
> 
> Oh, very much agreed.  Giving it a separate configure option also opens
> the door for easily adding an march=pentiumm whenever the gcc folks get
> around to adding that.

Yes.  That was the door I was aiming to open. <g>

-- 
Tomas Szepe <szepe@pinerecords.com>

Re: Pentium M config option for 2.6

[Dave Jones]

On Sun, Jan 04, 2004 at 06:03:47PM +0100, Troels Walsted Hansen wrote:

 > >FWIW, I agree with it too on the grounds that its non obvious the optimal
 > >setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
 > >helptext to read...
 > >
 > >"Pentium II"
 > >"Pentium III / Pentium 4M"
 > >"Pentium 4"
 > >
 > Especially since Pentium M != Pentium 4M :-)

Indeed. Colour me confused 8-)

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: Pentium M config option for 2.6

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1002 bytes --]

On Sun, 04 Jan 2004 11:33:08 EST, Rob Love said:

> I actually like this patch (perhaps since I have a P-M :) and think it
> ought to go in, although I agree with others that the P-M is more of a
> super-P3 than a scaled down P4.

Same here - /proc/cpuinfo says:

vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 Mobile CPU 1.60GHz

Question for those more knowledgeable: Are there any known Pentium4 features
enabled in the kernel with the PENTIUM4 options that simply Will Not Work on a
4M chipset (similar to a kernel built for a 586 not working on a 486), or are
the differences limited to "sub-optimal performance" (for example, compiling
with -mpentium4 results in code that runs but schedules less optimally)?  If
there are, they must be fairly obscure corner cases, since I haven't knowingly
hit one in several months.. :)

2.7 timeframe - are there any added features of a P4 core we would *like*
to exploit that aren't on a P4M?

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: Pentium M config option for 2.6

[Vojtech Pavlik]

On Sun, Jan 04, 2004 at 04:50:28PM +0000, Dave Jones wrote:
> On Sun, Jan 04, 2004 at 11:33:08AM -0500, Rob Love wrote:
> 
>  > Yah.  I was just answering in the abstract to the "does cache line
>  > matter on non-SMP" question.
>  > 
>  > I actually like this patch (perhaps since I have a P-M :) and think it
>  > ought to go in, although I agree with others that the P-M is more of a
>  > super-P3 than a scaled down P4.
> 
> FWIW, I agree with it too on the grounds that its non obvious the optimal
> setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
> helptext to read...
> 
>  "Pentium II"
>  "Pentium III / Pentium 4M"
>  "Pentium 4"
> 
> My other mail may have sounded like I objected to the patch per se, I don't.

Pentium M and Pentium 4M are two different beasts, by the way.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: Pentium M config option for 2.6

[Vojtech Pavlik]

On Sun, Jan 04, 2004 at 03:34:47PM -0500, Valdis.Kletnieks@vt.edu wrote:
> On Sun, 04 Jan 2004 11:33:08 EST, Rob Love said:
> 
> > I actually like this patch (perhaps since I have a P-M :) and think it
> > ought to go in, although I agree with others that the P-M is more of a
> > super-P3 than a scaled down P4.
> 
> Same here - /proc/cpuinfo says:
> 
> vendor_id       : GenuineIntel
> cpu family      : 15
> model           : 2
> model name      : Intel(R) Pentium(R) 4 Mobile CPU 1.60GHz
> 
> Question for those more knowledgeable: Are there any known Pentium4 features
> enabled in the kernel with the PENTIUM4 options that simply Will Not Work on a
> 4M chipset (similar to a kernel built for a 586 not working on a 486), or are
> the differences limited to "sub-optimal performance" (for example, compiling
> with -mpentium4 results in code that runs but schedules less optimally)?  If
> there are, they must be fairly obscure corner cases, since I haven't knowingly
> hit one in several months.. :)
> 
> 2.7 timeframe - are there any added features of a P4 core we would *like*
> to exploit that aren't on a P4M?

Pentium 4M is a real Pentium 4 core, but with mobile features.

Pentium M is a beefed up Pentium III core, with mobile features and all
		Pentium 4 extra features (instructions, etc).

I think that answers your question.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: Pentium M config option for 2.6

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 387 bytes --]

On Sun, 04 Jan 2004 22:55:41 +0100, Vojtech Pavlik said:

> Pentium 4M is a real Pentium 4 core, but with mobile features.
> 
> Pentium M is a beefed up Pentium III core, with mobile features and all
> 		Pentium 4 extra features (instructions, etc).
> 
> I think that answers your question.

Obviously, I'm confused by more than just the re-used Xeon name, thanks for
the cluestick. :)


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: Pentium M config option for 2.6

[Jakob Oestergaard]

On Sun, Jan 04, 2004 at 01:33:58PM +0100, Tomas Szepe wrote:
> On Jan-04 2004, Sun, 13:27 +0100
> Mikael Pettersson <mikpe@csd.uu.se> wrote:
> 
> > IOW, don't lie to the compiler and pretend P-M == P4
> > with that -march=pentium4.
> 
> What do you recommend to use as march then?  There is
> no pentiumm subarch support in gcc yet;  I was convinced
> p4 was the closest match.

Use the same as for P-III.

The P-M has the same instruction decoder (and execution unit) setup as
the P-III, which is *very* different from P-IV (which has one decoder
only, and then a trace cache for the decoded uops).  This is an
important difference from a code generator point of view.

>From reading Intel's optimization guides, it seems to me like the P-M is
pretty much just a slightly enhanced P-III (more cache AFAIR) which
happens to get shipped with a good mobile chipset - and that package
together is called Centrino.

That would also explain why Centrino leaves the P-IV based laptops in
the dust ;)

Cheers,

 / jakob

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 13:27 +0100
Mikael Pettersson <mikpe@csd.uu.se> wrote:

> IOW, don't lie to the compiler and pretend P-M == P4
> with that -march=pentium4.

What do you recommend to use as march then?  There is
no pentiumm subarch support in gcc yet;  I was convinced
p4 was the closest match.

-- 
Tomas Szepe <szepe@pinerecords.com>

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 07:27, Mikael Pettersson wrote:


> And since P-M doesn't do SMP, does cache line size even
> matter? There are no locks to protect from ping-ponging.

Cache line size does still come into the picture on UP, albeit not as
much as with SMP - but e.g. it still matters to things like device
drivers doing DMA.

	Rob Love

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 11:25, Dave Jones wrote:

> Regardless, Tomas's patch changed CONFIG_X86_L1_CACHE_SHIFT for
> that CPU, and CONFIG_X86_L1_CACHE_SHIFT shouldn't affect this.
> The cacheline size is determined at boottime using the code in
> pcibios_init() and set using pci_generic_prep_mwi().
> 
> The config option is the default that pci_cache_line_size starts at,
> but this gets overridden when the CPU type is determined.

Yah.  I was just answering in the abstract to the "does cache line
matter on non-SMP" question.

I actually like this patch (perhaps since I have a P-M :) and think it
ought to go in, although I agree with others that the P-M is more of a
super-P3 than a scaled down P4.

	Rob Love

Re: Pentium M config option for 2.6

[Dave Jones]

On Sun, Jan 04, 2004 at 11:33:08AM -0500, Rob Love wrote:

 > Yah.  I was just answering in the abstract to the "does cache line
 > matter on non-SMP" question.
 > 
 > I actually like this patch (perhaps since I have a P-M :) and think it
 > ought to go in, although I agree with others that the P-M is more of a
 > super-P3 than a scaled down P4.

FWIW, I agree with it too on the grounds that its non obvious the optimal
setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
helptext to read...

 "Pentium II"
 "Pentium III / Pentium 4M"
 "Pentium 4"

My other mail may have sounded like I objected to the patch per se, I don't.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: Pentium M config option for 2.6

[Rob Love]

On Sun, 2004-01-04 at 11:50, Dave Jones wrote:

> FWIW, I agree with it too on the grounds that its non obvious the optimal
> setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
> helptext to read...
> 
>  "Pentium II"
>  "Pentium III / Pentium 4M"
>  "Pentium 4"

Oh, very much agreed.  Giving it a separate configure option also opens
the door for easily adding an march=pentiumm whenever the gcc folks get
around to adding that.

> My other mail may have sounded like I objected to the patch per se, I don't.

I did not think that, I thought perhaps that you thought that I objected
:)

	Rob Love

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 12:01 -0500
Rob Love <rml@ximian.com> wrote:

> On Sun, 2004-01-04 at 11:50, Dave Jones wrote:
> 
> > FWIW, I agree with it too on the grounds that its non obvious the optimal
> > setting is CONFIG_MPENTIUMIII. This seems cleaner IMO than changing the
> > helptext to read...
> > 
> >  "Pentium II"
> >  "Pentium III / Pentium 4M"
> >  "Pentium 4"
> 
> Oh, very much agreed.  Giving it a separate configure option also opens
> the door for easily adding an march=pentiumm whenever the gcc folks get
> around to adding that.

Yes.  That was the door I was aiming to open. <g>

-- 
Tomas Szepe <szepe@pinerecords.com>

Re: Pentium M config option for 2.6

[Dave Jones]

On Sun, Jan 04, 2004 at 10:03:28AM -0500, Rob Love wrote:
 > On Sun, 2004-01-04 at 07:27, Mikael Pettersson wrote:
 > > And since P-M doesn't do SMP, does cache line size even
 > > matter? There are no locks to protect from ping-ponging.
 > 
 > Cache line size does still come into the picture on UP, albeit not as
 > much as with SMP - but e.g. it still matters to things like device
 > drivers doing DMA.

Regardless, Tomas's patch changed CONFIG_X86_L1_CACHE_SHIFT for
that CPU, and CONFIG_X86_L1_CACHE_SHIFT shouldn't affect this.
The cacheline size is determined at boottime using the code in
pcibios_init() and set using pci_generic_prep_mwi().

The config option is the default that pci_cache_line_size starts at,
but this gets overridden when the CPU type is determined.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: Pentium M config option for 2.6

[Jakob Oestergaard]

On Sun, Jan 04, 2004 at 01:33:58PM +0100, Tomas Szepe wrote:
> On Jan-04 2004, Sun, 13:27 +0100
> Mikael Pettersson <mikpe@csd.uu.se> wrote:
> 
> > IOW, don't lie to the compiler and pretend P-M == P4
> > with that -march=pentium4.
> 
> What do you recommend to use as march then?  There is
> no pentiumm subarch support in gcc yet;  I was convinced
> p4 was the closest match.

Use the same as for P-III.

The P-M has the same instruction decoder (and execution unit) setup as
the P-III, which is *very* different from P-IV (which has one decoder
only, and then a trace cache for the decoded uops).  This is an
important difference from a code generator point of view.

>From reading Intel's optimization guides, it seems to me like the P-M is
pretty much just a slightly enhanced P-III (more cache AFAIR) which
happens to get shipped with a good mobile chipset - and that package
together is called Centrino.

That would also explain why Centrino leaves the P-IV based laptops in
the dust ;)

Cheers,

 / jakob

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 15:10 +0100
Mikael Pettersson <mikpe@csd.uu.se> wrote:

> On Date: Sun, 4 Jan 2004 13:33:58 +0100, Tomas Szepe wrote:
> > > IOW, don't lie to the compiler and pretend P-M == P4
> > > with that -march=pentium4.
> > 
> > What do you recommend to use as march then?  There is
> > no pentiumm subarch support in gcc yet;  I was convinced
> > p4 was the closest match.
> 
> march=pentium3 is the closest safe choice, at least
> until gcc implements P-M specific support.

Thanks, here's the updated patch.

-- 
Tomas Szepe <szepe@pinerecords.com>


diff -urN a/arch/i386/Kconfig b/arch/i386/Kconfig
--- a/arch/i386/Kconfig	2004-01-04 03:10:01.000000000 +0100
+++ b/arch/i386/Kconfig	2004-01-04 03:06:09.000000000 +0100
@@ -231,6 +231,13 @@
 	  correct cache shift, and applies any applicable Pentium III
 	  optimizations.
 
+config MPENTIUMM
+	bool "Pentium M (Banias/Centrino)"
+	help
+	  Select this for Intel Pentium M chips.  This option enables
+	  compile flags optimized for the chip, uses the correct cache
+	  shift, and applies any applicable Pentium III/IV optimizations.
+
 config MK6
 	bool "K6/K6-II/K6-III"
 	help
@@ -330,7 +337,7 @@
 	default "7" if MPENTIUM4 || X86_GENERIC
 	default "4" if MELAN || M486 || M386
 	default "5" if MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCRUSOE || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2
-	default "6" if MK7 || MK8
+	default "6" if MPENTIUMM || MK7 || MK8
 
 config RWSEM_GENERIC_SPINLOCK
 	bool
@@ -379,17 +386,17 @@
 
 config X86_GOOD_APIC
 	bool
-	depends on MK7 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
+	depends on MK7 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
 	default y
 
 config X86_INTEL_USERCOPY
 	bool
-	depends on MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
+	depends on MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
 	default y
 
 config X86_USE_PPRO_CHECKSUM
 	bool
-	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
+	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
 	default y
 
 config X86_USE_3DNOW
@@ -512,7 +519,7 @@
 
 config X86_TSC
 	bool
-	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
+	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
 	default y
 
 config X86_MCE
diff -urN a/arch/i386/Makefile b/arch/i386/Makefile
--- a/arch/i386/Makefile	2003-09-28 11:38:05.000000000 +0200
+++ b/arch/i386/Makefile	2004-01-04 03:02:52.000000000 +0100
@@ -35,6 +35,7 @@
 cflags-$(CONFIG_MPENTIUMII)	+= $(call check_gcc,-march=pentium2,-march=i686)
 cflags-$(CONFIG_MPENTIUMIII)	+= $(call check_gcc,-march=pentium3,-march=i686)
 cflags-$(CONFIG_MPENTIUM4)	+= $(call check_gcc,-march=pentium4,-march=i686)
+cflags-$(CONFIG_MPENTIUMM)	+= $(call check_gcc,-march=pentium3,-march=i686)
 cflags-$(CONFIG_MK6)		+= $(call check_gcc,-march=k6,-march=i586)
 # Please note, that patches that add -march=athlon-xp and friends are pointless.
 # They make zero difference whatsosever to performance at this time.
diff -urN a/include/asm-i386/module.h b/include/asm-i386/module.h
--- a/include/asm-i386/module.h	2003-08-23 01:52:22.000000000 +0200
+++ b/include/asm-i386/module.h	2004-01-04 03:08:17.000000000 +0100
@@ -28,6 +28,8 @@
 #define MODULE_PROC_FAMILY "PENTIUMIII "
 #elif defined CONFIG_MPENTIUM4
 #define MODULE_PROC_FAMILY "PENTIUM4 "
+#elif defined CONFIG_MPENTIUMM
+#define MODULE_PROC_FAMILY "PENTIUMM "
 #elif defined CONFIG_MK6
 #define MODULE_PROC_FAMILY "K6 "
 #elif defined CONFIG_MK7

Re: Pentium M config option for 2.6

[Mikael Pettersson]

On Date: Sun, 4 Jan 2004 13:33:58 +0100, Tomas Szepe wrote:
> > IOW, don't lie to the compiler and pretend P-M == P4
> > with that -march=pentium4.
> 
> What do you recommend to use as march then?  There is
> no pentiumm subarch support in gcc yet;  I was convinced
> p4 was the closest match.

march=pentium3 is the closest safe choice, at least
until gcc implements P-M specific support.

/Mikael

Re: Pentium M config option for 2.6

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 470 bytes --]

On Sun, 4 Jan 2004 15:43:50 +0100 Tomas Szepe <szepe@pinerecords.com>
wrote:
| +config MPENTIUMM
| +	bool "Pentium M (Banias/Centrino)"
| +	help
| +	  Select this for Intel Pentium M chips.  This option enables
| +	  compile flags optimized for the chip, uses the correct cache
| +	  shift, and applies any applicable Pentium III/IV

That should probably read "Pentium III/4".

-- 
Ciaran McCreesh
Mail:    ciaranm at gentoo.org
Web:     http://dev.gentoo.org/~ciaranm


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Pentium M config option for 2.6

[Mikael Pettersson]

On Sun, 4 Jan 2004 03:28:48 +0100, Tomas Szepe wrote:
>Since the Pentium M has 64 byte cache lines and is not a K7 or K8...  ;)
...
>--- a/arch/i386/Makefile	2003-09-28 11:38:05.000000000 +0200
>+++ b/arch/i386/Makefile	2004-01-04 03:02:52.000000000 +0100
>@@ -35,6 +35,7 @@
> cflags-$(CONFIG_MPENTIUMII)	+= $(call check_gcc,-march=pentium2,-march=i686)
> cflags-$(CONFIG_MPENTIUMIII)	+= $(call check_gcc,-march=pentium3,-march=i686)
> cflags-$(CONFIG_MPENTIUM4)	+= $(call check_gcc,-march=pentium4,-march=i686)
>+cflags-$(CONFIG_MPENTIUMM)	+= $(call check_gcc,-march=pentium4,-march=i686)
> cflags-$(CONFIG_MK6)		+= $(call check_gcc,-march=k6,-march=i586)
> # Please note, that patches that add -march=athlon-xp and friends are pointless.
> # They make zero difference whatsosever to performance at this time.

P-M is not a P4 core, it's an enhanced PIII core.
SSE2 was added, but compiler support for SSE2 f.p.
math shouldn't matter for the kernel.

Using P4 optimisations on a P-M may actually reduce
performance, due to the different micro-architectures.
(P4 made shifts and some leas more expensive, and
simple add/and/sub/etc less expensive.)

IOW, don't lie to the compiler and pretend P-M == P4
with that -march=pentium4.

And since P-M doesn't do SMP, does cache line size even
matter? There are no locks to protect from ping-ponging.

/Mikael

Re: Pentium M config option for 2.6

[Tim Schmielau]

On Mon, 5 Jan 2004, Amit Gurdasani wrote:

> Perhaps -mcpu=pentium3 -march=pentium4 would be a good compromise? From the
> GCC 3.3 info pages:
[...]
>      for CPU-TYPE are the same as for `-mcpu'.  Moreover, specifying
>      `-march=CPU-TYPE' implies `-mcpu=CPU-TYPE'.

... thus leaving the "-mcpu=pentium3" without effect.

Tim

Re: Pentium M config option for 2.6

[Amit Gurdasani]

:On Date: Sun, 4 Jan 2004 13:33:58 +0100, Tomas Szepe wrote:
:> > IOW, don't lie to the compiler and pretend P-M == P4
:> > with that -march=pentium4.
:>
:> What do you recommend to use as march then?  There is
:> no pentiumm subarch support in gcc yet;  I was convinced
:> p4 was the closest match.
:
:march=pentium3 is the closest safe choice, at least
:until gcc implements P-M specific support.
:
:/Mikael

Perhaps -mcpu=pentium3 -march=pentium4 would be a good compromise? From the
GCC 3.3 info pages:

`-mcpu=CPU-TYPE'
     Tune to CPU-TYPE everything applicable about the generated code,
     except for the ABI and the set of available instructions. [...]

     While picking a specific CPU-TYPE will schedule things
     appropriately for that particular chip, the compiler will not
     generate any code that does not run on the i386 without the
     `-march=CPU-TYPE' option being used. [...]

`-march=CPU-TYPE'
     Generate instructions for the machine type CPU-TYPE.  The choices
     for CPU-TYPE are the same as for `-mcpu'.  Moreover, specifying
     `-march=CPU-TYPE' implies `-mcpu=CPU-TYPE'.

Amit

Re: Pentium M config option for 2.6

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 470 bytes --]

On Sun, 4 Jan 2004 15:43:50 +0100 Tomas Szepe <szepe@pinerecords.com>
wrote:
| +config MPENTIUMM
| +	bool "Pentium M (Banias/Centrino)"
| +	help
| +	  Select this for Intel Pentium M chips.  This option enables
| +	  compile flags optimized for the chip, uses the correct cache
| +	  shift, and applies any applicable Pentium III/IV

That should probably read "Pentium III/4".

-- 
Ciaran McCreesh
Mail:    ciaranm at gentoo.org
Web:     http://dev.gentoo.org/~ciaranm


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: Pentium M config option for 2.6

[Tomas Szepe]

On Jan-04 2004, Sun, 15:10 +0100
Mikael Pettersson <mikpe@csd.uu.se> wrote:

> On Date: Sun, 4 Jan 2004 13:33:58 +0100, Tomas Szepe wrote:
> > > IOW, don't lie to the compiler and pretend P-M == P4
> > > with that -march=pentium4.
> > 
> > What do you recommend to use as march then?  There is
> > no pentiumm subarch support in gcc yet;  I was convinced
> > p4 was the closest match.
> 
> march=pentium3 is the closest safe choice, at least
> until gcc implements P-M specific support.

Thanks, here's the updated patch.

-- 
Tomas Szepe <szepe@pinerecords.com>


diff -urN a/arch/i386/Kconfig b/arch/i386/Kconfig
--- a/arch/i386/Kconfig	2004-01-04 03:10:01.000000000 +0100
+++ b/arch/i386/Kconfig	2004-01-04 03:06:09.000000000 +0100
@@ -231,6 +231,13 @@
 	  correct cache shift, and applies any applicable Pentium III
 	  optimizations.
 
+config MPENTIUMM
+	bool "Pentium M (Banias/Centrino)"
+	help
+	  Select this for Intel Pentium M chips.  This option enables
+	  compile flags optimized for the chip, uses the correct cache
+	  shift, and applies any applicable Pentium III/IV optimizations.
+
 config MK6
 	bool "K6/K6-II/K6-III"
 	help
@@ -330,7 +337,7 @@
 	default "7" if MPENTIUM4 || X86_GENERIC
 	default "4" if MELAN || M486 || M386
 	default "5" if MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCRUSOE || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2
-	default "6" if MK7 || MK8
+	default "6" if MPENTIUMM || MK7 || MK8
 
 config RWSEM_GENERIC_SPINLOCK
 	bool
@@ -379,17 +386,17 @@
 
 config X86_GOOD_APIC
 	bool
-	depends on MK7 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
+	depends on MK7 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
 	default y
 
 config X86_INTEL_USERCOPY
 	bool
-	depends on MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
+	depends on MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
 	default y
 
 config X86_USE_PPRO_CHECKSUM
 	bool
-	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
+	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
 	default y
 
 config X86_USE_3DNOW
@@ -512,7 +519,7 @@
 
 config X86_TSC
 	bool
-	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
+	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
 	default y
 
 config X86_MCE
diff -urN a/arch/i386/Makefile b/arch/i386/Makefile
--- a/arch/i386/Makefile	2003-09-28 11:38:05.000000000 +0200
+++ b/arch/i386/Makefile	2004-01-04 03:02:52.000000000 +0100
@@ -35,6 +35,7 @@
 cflags-$(CONFIG_MPENTIUMII)	+= $(call check_gcc,-march=pentium2,-march=i686)
 cflags-$(CONFIG_MPENTIUMIII)	+= $(call check_gcc,-march=pentium3,-march=i686)
 cflags-$(CONFIG_MPENTIUM4)	+= $(call check_gcc,-march=pentium4,-march=i686)
+cflags-$(CONFIG_MPENTIUMM)	+= $(call check_gcc,-march=pentium3,-march=i686)
 cflags-$(CONFIG_MK6)		+= $(call check_gcc,-march=k6,-march=i586)
 # Please note, that patches that add -march=athlon-xp and friends are pointless.
 # They make zero difference whatsosever to performance at this time.
diff -urN a/include/asm-i386/module.h b/include/asm-i386/module.h
--- a/include/asm-i386/module.h	2003-08-23 01:52:22.000000000 +0200
+++ b/include/asm-i386/module.h	2004-01-04 03:08:17.000000000 +0100
@@ -28,6 +28,8 @@
 #define MODULE_PROC_FAMILY "PENTIUMIII "
 #elif defined CONFIG_MPENTIUM4
 #define MODULE_PROC_FAMILY "PENTIUM4 "
+#elif defined CONFIG_MPENTIUMM
+#define MODULE_PROC_FAMILY "PENTIUMM "
 #elif defined CONFIG_MK6
 #define MODULE_PROC_FAMILY "K6 "
 #elif defined CONFIG_MK7

Re: Pentium M config option for 2.6

[Mikael Pettersson]

On Date: Sun, 4 Jan 2004 13:33:58 +0100, Tomas Szepe wrote:
> > IOW, don't lie to the compiler and pretend P-M == P4
> > with that -march=pentium4.
> 
> What do you recommend to use as march then?  There is
> no pentiumm subarch support in gcc yet;  I was convinced
> p4 was the closest match.

march=pentium3 is the closest safe choice, at least
until gcc implements P-M specific support.

/Mikael

Pentium M config option for 2.6

[Tomas Szepe]

Since the Pentium M has 64 byte cache lines and is not a K7 or K8...  ;)

-- 
Tomas Szepe <szepe@pinerecords.com>


diff -urN a/arch/i386/Kconfig b/arch/i386/Kconfig
--- a/arch/i386/Kconfig	2004-01-04 03:10:01.000000000 +0100
+++ b/arch/i386/Kconfig	2004-01-04 03:06:09.000000000 +0100
@@ -231,6 +231,13 @@
 	  correct cache shift, and applies any applicable Pentium III
 	  optimizations.
 
+config MPENTIUMM
+	bool "Pentium M (Banias/Centrino)"
+	help
+	  Select this for Intel Pentium M chips.  This option enables
+	  compile flags optimized for the chip, uses the correct cache
+	  shift, and applies any applicable Pentium III/IV optimizations.
+
 config MK6
 	bool "K6/K6-II/K6-III"
 	help
@@ -330,7 +337,7 @@
 	default "7" if MPENTIUM4 || X86_GENERIC
 	default "4" if MELAN || M486 || M386
 	default "5" if MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCRUSOE || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2
-	default "6" if MK7 || MK8
+	default "6" if MPENTIUMM || MK7 || MK8
 
 config RWSEM_GENERIC_SPINLOCK
 	bool
@@ -379,17 +386,17 @@
 
 config X86_GOOD_APIC
 	bool
-	depends on MK7 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
+	depends on MK7 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || MK8
 	default y
 
 config X86_INTEL_USERCOPY
 	bool
-	depends on MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
+	depends on MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M586MMX || X86_GENERIC || MK8 || MK7
 	default y
 
 config X86_USE_PPRO_CHECKSUM
 	bool
-	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
+	depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2
 	default y
 
 config X86_USE_3DNOW
@@ -512,7 +519,7 @@
 
 config X86_TSC
 	bool
-	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
+	depends on (MWINCHIP3D || MWINCHIP2 || MCRUSOE || MCYRIXIII || MK7 || MK6 || MPENTIUMM || MPENTIUM4 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2) && !X86_NUMAQ
 	default y
 
 config X86_MCE
diff -urN a/arch/i386/Makefile b/arch/i386/Makefile
--- a/arch/i386/Makefile	2003-09-28 11:38:05.000000000 +0200
+++ b/arch/i386/Makefile	2004-01-04 03:02:52.000000000 +0100
@@ -35,6 +35,7 @@
 cflags-$(CONFIG_MPENTIUMII)	+= $(call check_gcc,-march=pentium2,-march=i686)
 cflags-$(CONFIG_MPENTIUMIII)	+= $(call check_gcc,-march=pentium3,-march=i686)
 cflags-$(CONFIG_MPENTIUM4)	+= $(call check_gcc,-march=pentium4,-march=i686)
+cflags-$(CONFIG_MPENTIUMM)	+= $(call check_gcc,-march=pentium4,-march=i686)
 cflags-$(CONFIG_MK6)		+= $(call check_gcc,-march=k6,-march=i586)
 # Please note, that patches that add -march=athlon-xp and friends are pointless.
 # They make zero difference whatsosever to performance at this time.
diff -urN a/include/asm-i386/module.h b/include/asm-i386/module.h
--- a/include/asm-i386/module.h	2003-08-23 01:52:22.000000000 +0200
+++ b/include/asm-i386/module.h	2004-01-04 03:08:17.000000000 +0100
@@ -28,6 +28,8 @@
 #define MODULE_PROC_FAMILY "PENTIUMIII "
 #elif defined CONFIG_MPENTIUM4
 #define MODULE_PROC_FAMILY "PENTIUM4 "
+#elif defined CONFIG_MPENTIUMM
+#define MODULE_PROC_FAMILY "PENTIUMM "
 #elif defined CONFIG_MK6
 #define MODULE_PROC_FAMILY "K6 "
 #elif defined CONFIG_MK7