Re: Package with x86 vector instruction support

2011-05-02 Thread Christoph Frieben
On Wed, Apr 20, 2011 at 5:27 PM, Jussi Lehtola wrote: > x86_64: plain and sse2 This does not look good to me, because the x86-64 instruction set only includes sse2. sse3 is not available on every amd64 processor and thus should not be used by default. The maintainer should rather build the base pa

Re: Package with x86 vector instruction support

2011-04-20 Thread Jerry James
On Wed, Apr 20, 2011 at 5:27 PM, Jussi Lehtola wrote: > What package are we talking about? > > There is already at least one package in Fedora which does a similar > thing: ATLAS. It normally automatically tunes itself for the optimal > build flags, so some special tricks have to be done for packa

Re: Package with x86 vector instruction support

2011-04-20 Thread Jussi Lehtola
On Wed, 20 Apr 2011 11:59:18 -0600 Jerry James wrote: > I'm packaging a library that does some heavy mathematical > computations. The build system tries to cleverly determine whether an > Intel CPU with vector instructions is being used. What package are we talking about? There is already at le

Re: Package with x86 vector instruction support

2011-04-20 Thread Adam Jackson
On 4/20/11 5:57 PM, Jerry James wrote: > On Wed, Apr 20, 2011 at 3:32 PM, Adam Jackson wrote: >> The least capable processor we've gone out of our way to fix things for is >> the AMD Geode GX, which has MMX and 3DNow, but is also slightly less than a >> proper Pentium Pro in that it doesn't have n

Re: Package with x86 vector instruction support

2011-04-20 Thread Jerry James
On Wed, Apr 20, 2011 at 3:32 PM, Adam Jackson wrote: > The least capable processor we've gone out of our way to fix things for is > the AMD Geode GX, which has MMX and 3DNow, but is also slightly less than a > proper Pentium Pro in that it doesn't have nopl or cmov. > > However, rpm sets --march=i

Re: Package with x86 vector instruction support

2011-04-20 Thread Adam Jackson
On 4/20/11 1:59 PM, Jerry James wrote: > I'm packaging a library that does some heavy mathematical > computations. The build system tries to cleverly determine whether an > Intel CPU with vector instructions is being used. Until I can > convince upstream to do this at runtime, I'll need to build

Package with x86 vector instruction support

2011-04-20 Thread Jerry James
I've got a question whose answer I expected to find on one of these pages, but no joy: https://fedoraproject.org/wiki/Architectures https://fedoraproject.org/wiki/Architectures/x86 https://fedoraproject.org/wiki/Architectures/x86-64 I'm packaging a library that does some heavy mathematical comput