[gentoo-user] common flags for 2 cpu?

[gentoo-user] common flags for 2 cpu?

[Jorge Almeida]

I'm in need of some expert advise about CFLAGS. I'm going to install
Gentoo in two systems, an Atom 330 and a P4 prescott. Since compiling
lots of stuff in an Atom is less than joyful, I intend to do almost
all compilation on the prescott and produce binary packages to use on
the Atom. So, something like
CFLAGS="-march=native ..." is out. I know I could set
CFLAGS="-march=i686 -O2 -pipe -fomit-frame-pointer", but I would like
to extract a bit more juice from the hw, while still keeping stable
systems. So, I thought of setting CFLAGS for both systems as
-march=i686, plus the flags that are pulled by -march=native in both
systems, plus some other flags:

-march=i686
-mno-aes
-mno-pclmul
-mno-popcnt
-mno-abm
-mno-lwp
-mno-fma
-mno-fma4
-mno-xop
-mno-bmi
-mno-bmi2
-mno-tbm
-mno-avx
-mno-avx2
-mno-sse4.2
-mno-sse4.1
-mno-lzcnt
-mno-rdrnd
-mno-f16c
-mno-fsgsbase
--param l1-cache-size=16
--param l1-cache-line-size=64
--param l2-cache-size=1024
-pni
-mtrr

This is what is pulled by -march=native, using gcc -march=native -E -v
- </dev/null 2>&1 | grep cc1

Prescott:
-march=prescott -mno-cx16 -mno-sahf -mno-movbe -mno-aes -mno-pclmul
-mno-popcnt -mno-abm -mno-lwp -mno-fma -mno-fma4 -mno-xop -mno-bmi
-mno-bmi2 -mno-tbm -mno-avx -mno-avx2 -mno-sse4.2 -mno-sse4.1
-mno-lzcnt -mno-rdrnd -mno-f16c -mno-fsgsbase --param l1-cache-size=16
--param l1-cache-line-size=64 --param l2-cache-size=1024
-mtune=prescott

Atom:
-march=atom -mcx16 -msahf -mmovbe -mno-aes -mno-pclmul -mno-popcnt
-mno-abm -mno-lwp -mno-fma -mno-fma4 -mno-xop -mno-bmi -mno-bmi2
-mno-tbm -mno-avx -mno-avx2 -mno-sse4.2 -mno-sse4.1 -mno-lzcnt
-mno-rdrnd -mno-f16c -mno-fsgsbase --param l1-cache-size=24 --param
l1-cache-line-size=64 --param l2-cache-size=512 -mtune=atom

Flags supported (from cat /proc/cpuinfo|grep flags):

Prescott:
fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat
pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe constant_tsc
pebs bts pni dtes64 monitor ds_cpl cid xtpr

Atom:
fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat
clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx lm constant_tsc
arch_perfmon pebs bts aperfmperf pni dtes64 monitor ds_cpl tm2 ssse3
cx16 xtpr pdcm movbe lahf_lm dtherm


Some questions:

1) Is this strategy right? If so, any other flags to add? (or any
flags to remove from the list?)

2) The --param flags are the ones of the computer that will do the
compiling. I'm guessing the produced binaries are compatible with cpu
with different --param flags. Is this right?

TIA

Jorge Almeida

Re: [gentoo-user] common flags for 2 cpu?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 2423 bytes --]

Am 17.08.2012 10:58, schrieb Jorge Almeida:
> I'm in need of some expert advise about CFLAGS. I'm going to install
> Gentoo in two systems, an Atom 330 and a P4 prescott. Since compiling
> lots of stuff in an Atom is less than joyful, I intend to do almost
> all compilation on the prescott and produce binary packages to use on
> the Atom. So, something like
> CFLAGS="-march=native ..." is out. I know I could set
> CFLAGS="-march=i686 -O2 -pipe -fomit-frame-pointer", but I would like
> to extract a bit more juice from the hw, while still keeping stable
> systems. So, I thought of setting CFLAGS for both systems as
> -march=i686, plus the flags that are pulled by -march=native in both
> systems, plus some other flags:
> 
[...]
>
> This is what is pulled by -march=native, using gcc -march=native -E -v
> - </dev/null 2>&1 | grep cc1
> 
> Prescott:
> -march=prescott -mno-cx16 -mno-sahf -mno-movbe -mno-aes -mno-pclmul
> -mno-popcnt -mno-abm -mno-lwp -mno-fma -mno-fma4 -mno-xop -mno-bmi
> -mno-bmi2 -mno-tbm -mno-avx -mno-avx2 -mno-sse4.2 -mno-sse4.1
> -mno-lzcnt -mno-rdrnd -mno-f16c -mno-fsgsbase --param l1-cache-size=16
> --param l1-cache-line-size=64 --param l2-cache-size=1024
> -mtune=prescott
> 
> Atom:
> -march=atom -mcx16 -msahf -mmovbe -mno-aes -mno-pclmul -mno-popcnt
> -mno-abm -mno-lwp -mno-fma -mno-fma4 -mno-xop -mno-bmi -mno-bmi2
> -mno-tbm -mno-avx -mno-avx2 -mno-sse4.2 -mno-sse4.1 -mno-lzcnt
> -mno-rdrnd -mno-f16c -mno-fsgsbase --param l1-cache-size=24 --param
> l1-cache-line-size=64 --param l2-cache-size=512 -mtune=atom
> 
[...]
>
> Some questions:
> 
> 1) Is this strategy right? If so, any other flags to add? (or any
> flags to remove from the list?)
> 
> 2) The --param flags are the ones of the computer that will do the
> compiling. I'm guessing the produced binaries are compatible with cpu
> with different --param flags. Is this right?
> 
> TIA
> 
> Jorge Almeida
> 

1) Yes. But as you can see, -march=prescott is basically a subset of
atom. In fact, before there was a -march=atom option, prescott was the
best flag for atoms. I think you can avoid some hassle by simply
enabling "-march=prescott --param l1-cache-line-size=64 --param
l2-cache-size=512".

2) Yes, the param flags do not affect compatibility. Using the lower
value will probably be better but this is just an educated guess.

Regards,
Florian Philipp


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [gentoo-user] common flags for 2 cpu?

[Jorge Almeida]

On Fri, Aug 17, 2012 at 2:53 PM, Florian Philipp <lists@binarywings.net> wrote:
> Am 17.08.2012 10:58, schrieb Jorge Almeida:
>>
>> 1) Is this strategy right? If so, any other flags to add? (or any
>> flags to remove from the list?)
>>
>> 2) The --param flags are the ones of the computer that will do the
>> compiling. I'm guessing the produced binaries are compatible with cpu
>> with different --param flags. Is this right?
>>
>> TIA
>>
>> Jorge Almeida
>>
>
> 1) Yes. But as you can see, -march=prescott is basically a subset of
> atom. In fact, before there was a -march=atom option, prescott was the
> best flag for atoms. I think you can avoid some hassle by simply
> enabling "-march=prescott --param l1-cache-line-size=64 --param
> l2-cache-size=512".
>
> 2) Yes, the param flags do not affect compatibility. Using the lower
> value will probably be better but this is just an educated guess.
>

What about:

CFLAGS="-O2 -pipe -fomit-frame-pointer -march=prescott -mtune=atom --param
l1-cache-size=16 --param l1-cache-line-size=64 --param l2-cache-size=1024"

If prescott were exactly a subset of atom, this would yield the best of both
worlds. Can it still be safe?

I read in http://forums.gentoo.org/viewtopic-t-895104.html :
atom
Intel Atom CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3
instruction set support.

Does this mean that these flags are pulled by -mtune=atom, or do we need to
ask for them explicitly? The WiKi shows how to find which flags are pulled by
-march=native, but not the other cases.

Thanks

Jorge Almeida

Re: [gentoo-user] common flags for 2 cpu?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 2127 bytes --]

Am 17.08.2012 19:57, schrieb Jorge Almeida:
> On Fri, Aug 17, 2012 at 2:53 PM, Florian Philipp <lists@binarywings.net> wrote:
>> Am 17.08.2012 10:58, schrieb Jorge Almeida:
>>>
>>> 1) Is this strategy right? If so, any other flags to add? (or any
>>> flags to remove from the list?)
>>>
>>> 2) The --param flags are the ones of the computer that will do the
>>> compiling. I'm guessing the produced binaries are compatible with cpu
>>> with different --param flags. Is this right?
>>>
>>> TIA
>>>
>>> Jorge Almeida
>>>
>>
>> 1) Yes. But as you can see, -march=prescott is basically a subset of
>> atom. In fact, before there was a -march=atom option, prescott was the
>> best flag for atoms. I think you can avoid some hassle by simply
>> enabling "-march=prescott --param l1-cache-line-size=64 --param
>> l2-cache-size=512".
>>
>> 2) Yes, the param flags do not affect compatibility. Using the lower
>> value will probably be better but this is just an educated guess.
>>
> 
> What about:
> 
> CFLAGS="-O2 -pipe -fomit-frame-pointer -march=prescott -mtune=atom --param
> l1-cache-size=16 --param l1-cache-line-size=64 --param l2-cache-size=1024"
> 
> If prescott were exactly a subset of atom, this would yield the best of both
> worlds. Can it still be safe?
> 
> I read in http://forums.gentoo.org/viewtopic-t-895104.html :
> atom
> Intel Atom CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3
> instruction set support.
> 
> Does this mean that these flags are pulled by -mtune=atom, or do we need to
> ask for them explicitly? The WiKi shows how to find which flags are pulled by
> -march=native, but not the other cases.
> 
> Thanks
> 
> Jorge Almeida
> 

Oh, sorry, I didn't think of the second generation atoms. I guess your
newer atom is a bit more different from prescott than the one I talked
about. Anyway, using -march=prescott is still viable. It just means you
lose a bit more in terms of usable SSE extensions on your atom.

Your CFLAGS look good to me. They won't pull in anything that your
prescott cannot handle.

Regards,
Florian Philipp


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [gentoo-user] common flags for 2 cpu?

[Jorge Almeida]

On Fri, Aug 17, 2012 at 10:05 PM, Florian Philipp <lists@binarywings.net> wrote:
> Am 17.08.2012 19:57, schrieb Jorge Almeida:
>> I read in http://forums.gentoo.org/viewtopic-t-895104.html :
>> atom
>> Intel Atom CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3
>> instruction set support.
>>
>> Does this mean that these flags are pulled by -mtune=atom, or do we need to
>> ask for them explicitly? The WiKi shows how to find which flags are pulled by
>> -march=native, but not the other cases.
>>
>> Thanks
>>
>> Jorge Almeida
>>
>
> Oh, sorry, I didn't think of the second generation atoms. I guess your
> newer atom is a bit more different from prescott than the one I talked
> about. Anyway, using -march=prescott is still viable. It just means you
> lose a bit more in terms of usable SSE extensions on your atom.
>
> Your CFLAGS look good to me. They won't pull in anything that your
> prescott cannot handle.
>
Does -march=prescott not ensure that the prescott will work even if
unsupported flags are added? Of, course such flags wouldn't be of use by the
prescott, but the atom might be able to use them. Can you confirm this?

Thanks,

Jorge Almeida

Re: [gentoo-user] common flags for 2 cpu?

[Volker Armin Hemmann]

Am Freitag, 17. August 2012, 22:34:45 schrieb Jorge Almeida:
> Does -march=prescott not ensure that the prescott will work even if
> unsupported flags are added?

no

every flag on its own. If you set a flag it is your responsibility to check that 
they actually work.

-- 
#163933