[gentoo-science] Problems emerging blas-atlas

[gentoo-science] Problems emerging blas-atlas

[Alan Jackson]

I'm having problems emerging blas-atlas-3.6.0-r1 on my AMD64 X2 system.
The configure drops into an infinite loop :

	I am now going to ask for two C compilers, and their associated flags.
	The first such set (CC & CCFLAGS) are used in compiling the non-generated
	ATLAS code.  This code is written in normal C, and responds well to high
	levels of optimization.  Typically, this is set to your default compiler,
	and your highest levels of optimization.

	The second set of C compilers (MCC & MMFLAGS) is used to compile the generated
	ATLAS code.  Generated codes are written at a very low-level (think of C used
	as a kind of portable assembler).  On many platforms, high levels of
	optimization are detrimental, as the compiler tries to pipeline a perfectly
	pipelined code, and succeeds in reducing performance substantially (this
	occurs on DEC ALPHAs & Sun UltraSparcs, for instance).  If the default does
	not work for you, try a midrange optimization such as -O.  The generated code
	does not alias any output arguments, so aliasing optimizations should be OK.

	   Enter ANSI C compiler(CC) [/usr/bin/gcc]:    Enter C Flags (CCFLAGS) [-fomit-frame-pointer -O -mfpmath=387 -m64]: Compiling with = '/usr/bin/gcc -fomit-frame-pointer -O -mfpmath=387 -m64' doesn't seem to work for me.
	   Enter 1 to try different CC/flags, 0 to continue anyway [1]: 

I've turned in a bugzilla ticket - but I thought someone here might have
a suggestion for my platform. Is there a compiler flag I should set that
would make it happy? Mine are pretty simple : CFLAGS="-O2 -march=athlon-xp"
This is a brand new system, so all the ebuilds from the kernel up are completely
current.



-- 
-----------------------------------------------------------------------
| Alan K. Jackson            | To see a World in a Grain of Sand      |
| alan@ajackson.org          | And a Heaven in a Wild Flower,         |
| www.ajackson.org           | Hold Infinity in the palm of your hand |
| Houston, Texas             | And Eternity in an hour. - Blake       |
-----------------------------------------------------------------------
-- 
gentoo-science@gentoo.org mailing list

Re: [gentoo-science] Problems emerging blas-atlas

[Jason Nielsen]

On Sun, 5 Feb 2006, Alan Jackson wrote:

> I'm having problems emerging blas-atlas-3.6.0-r1 on my AMD64 X2 system.
> The configure drops into an infinite loop :
>
> 	I am now going to ask for two C compilers, and their associated flags.
> 	The first such set (CC & CCFLAGS) are used in compiling the non-generated
> 	ATLAS code.  This code is written in normal C, and responds well to high
> 	levels of optimization.  Typically, this is set to your default compiler,
> 	and your highest levels of optimization.
>
> 	The second set of C compilers (MCC & MMFLAGS) is used to compile the generated
> 	ATLAS code.  Generated codes are written at a very low-level (think of C used
> 	as a kind of portable assembler).  On many platforms, high levels of
> 	optimization are detrimental, as the compiler tries to pipeline a perfectly
> 	pipelined code, and succeeds in reducing performance substantially (this
> 	occurs on DEC ALPHAs & Sun UltraSparcs, for instance).  If the default does
> 	not work for you, try a midrange optimization such as -O.  The generated code
> 	does not alias any output arguments, so aliasing optimizations should be OK.
>
> 	   Enter ANSI C compiler(CC) [/usr/bin/gcc]:    Enter C Flags (CCFLAGS) [-fomit-frame-pointer -O -mfpmath=387 -m64]: Compiling with = '/usr/bin/gcc -fomit-frame-pointer -O -mfpmath=387 -m64' doesn't seem to work for me.
> 	   Enter 1 to try different CC/flags, 0 to continue anyway [1]:
>
> I've turned in a bugzilla ticket - but I thought someone here might have
> a suggestion for my platform. Is there a compiler flag I should set that
> would make it happy? Mine are pretty simple : CFLAGS="-O2 -march=athlon-xp"
> This is a brand new system, so all the ebuilds from the kernel up are completely
> current.
>

Version 3.6.0 doesn't support AMD64 chips.  You need to install the 3.7.11 
devel version, it is portage but masked.  If you are looking for a high 
performance Lapack/BLAS for AMD64 you can also try the ACML library 
(http://developer.amd.com/acml.aspx).

Cheers,
Jason
-- 
gentoo-science@gentoo.org mailing list

Re: [gentoo-science] Problems emerging blas-atlas

[Alan Jackson]

On Sun, 5 Feb 2006 13:57:46 -0800 (PST)
Jason Nielsen <jdn@cs.sfu.ca> wrote:

> On Sun, 5 Feb 2006, Alan Jackson wrote:
> 
> > I'm having problems emerging blas-atlas-3.6.0-r1 on my AMD64 X2 system.
> > The configure drops into an infinite loop :
> >
> > 	I am now going to ask for two C compilers, and their associated flags.
> > 	The first such set (CC & CCFLAGS) are used in compiling the non-generated
> > 	ATLAS code.  This code is written in normal C, and responds well to high
> > 	levels of optimization.  Typically, this is set to your default compiler,
> > 	and your highest levels of optimization.
> >
> > 	The second set of C compilers (MCC & MMFLAGS) is used to compile the generated
> > 	ATLAS code.  Generated codes are written at a very low-level (think of C used
> > 	as a kind of portable assembler).  On many platforms, high levels of
> > 	optimization are detrimental, as the compiler tries to pipeline a perfectly
> > 	pipelined code, and succeeds in reducing performance substantially (this
> > 	occurs on DEC ALPHAs & Sun UltraSparcs, for instance).  If the default does
> > 	not work for you, try a midrange optimization such as -O.  The generated code
> > 	does not alias any output arguments, so aliasing optimizations should be OK.
> >
> > 	   Enter ANSI C compiler(CC) [/usr/bin/gcc]:    Enter C Flags (CCFLAGS) [-fomit-frame-pointer -O -mfpmath=387 -m64]: Compiling with = '/usr/bin/gcc -fomit-frame-pointer -O -mfpmath=387 -m64' doesn't seem to work for me.
> > 	   Enter 1 to try different CC/flags, 0 to continue anyway [1]:
> >
> > I've turned in a bugzilla ticket - but I thought someone here might have
> > a suggestion for my platform. Is there a compiler flag I should set that
> > would make it happy? Mine are pretty simple : CFLAGS="-O2 -march=athlon-xp"
> > This is a brand new system, so all the ebuilds from the kernel up are completely
> > current.
> >
> 
> Version 3.6.0 doesn't support AMD64 chips.  You need to install the 3.7.11 
> devel version, it is portage but masked.  If you are looking for a high 
> performance Lapack/BLAS for AMD64 you can also try the ACML library 
> (http://developer.amd.com/acml.aspx).
> 

Interesting. According to the package description on the website, 
3.6.0-r1 is stable for amd64...but, whatever. It appears to be working now.

Thanks!

(I'm working towards compiling SciPy and R)

-- 
-----------------------------------------------------------------------
| Alan K. Jackson            | To see a World in a Grain of Sand      |
| alan@ajackson.org          | And a Heaven in a Wild Flower,         |
| www.ajackson.org           | Hold Infinity in the palm of your hand |
| Houston, Texas             | And Eternity in an hour. - Blake       |
-----------------------------------------------------------------------
-- 
gentoo-science@gentoo.org mailing list