Bug 515307

Summary:	ATLAS doesn't build sse packages under Rawhide
Product:	[Fedora] Fedora	Reporter:	Mark Chappell <fedora>
Component:	atlas	Assignee:	Deji Akingunola <dakingun>
Status:	CLOSED NEXTRELEASE	QA Contact:	Fedora Extras Quality Assurance <extras-qa>
Severity:	medium	Docs Contact:
Priority:	low
Version:	rawhide	CC:	dakingun
Target Milestone:	---
Target Release:	---
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2009-08-04 13:44:59 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description Mark Chappell 2009-08-03 16:43:53 UTC

Building ATLAS on an EL5 derivative and noticed a few oddities with the package:

FC12 has moved to i686, meaning that the hardcoded i586 for SSE packages doesn't get built.  http://koji.fedoraproject.org/koji/buildinfo?buildID=116298

Is hard coding x86 build arches really the best way to go? What's wrong with the ix86 macro?

Also, not building SSE for x86_64 feels "wrong"

Could alternatives be a nice way to select just one blas implementation (including the atlas-sse variations)?  

The include files should probably also be "touched" to set the datestamp to a known value and allow RPM to ignore the "conflict" on multiarch platforms.  


Mark

Comment 1 Deji Akingunola 2009-08-04 13:44:59 UTC

(In reply to comment #0)
> Building ATLAS on an EL5 derivative and noticed a few oddities with the
> package:
> 
> FC12 has moved to i686, meaning that the hardcoded i586 for SSE packages
> doesn't get built.  http://koji.fedoraproject.org/koji/buildinfo?buildID=116298
>
Known issue, fixed build will appear in rawhide shortly.
 
> Is hard coding x86 build arches really the best way to go? What's wrong with
> the ix86 macro?
> 
> Also, not building SSE for x86_64 feels "wrong"
>
For now, we build atlas (on Fedora) for x86_64 arch with the least common denominator of sse instruction set (sse3) on all x86_64 cpus.

> Could alternatives be a nice way to select just one blas implementation
> (including the atlas-sse variations)?  
> 
What problem would it be solving?

Comment 2 Mark Chappell 2009-08-04 14:14:39 UTC

(In reply to comment #1)
> (In reply to comment #0)
> > Also, not building SSE for x86_64 feels "wrong"
> >
> For now, we build atlas (on Fedora) for x86_64 arch with the least common
> denominator of sse instruction set (sse3) on all x86_64 cpus.

Must have missed that bit, sorry.

> > Could alternatives be a nice way to select just one blas implementation
> > (including the atlas-sse variations)?  
> > 
> What problem would it be solving?  

Just wondering out loud if there's a nice way to set atlas/goto/netlib as the "default" blas/lapack for a machine, for things like Macaulay2 and numpy.  Since atlas implements a standard (blas) API, it would however mean repacking the atlas libs to create a libblas.so


Mark