MKL and IPP on Celeron or Athlon PC

[fiopa]

Hi

I'm developing a musical application, and I'm valuating the opportunity to buy
Math Kernel Library and/or Integrated Performance Primitives for Windows platform.
I'm using Visual Studio .NET on Windows XP Pro.

I've some questions:

1) If I develop a vectorial algorithm optimized for SSE, this work optimized with all SSE supporting CPU,
like last generation Intel Celeron or AMD Athlon, or work optimized only on Pentium IV and above ?
In other words, if program run on a Athlon, can library automatically detect SSE support
and use them efficiently ?

2) There are any problem / limitation to use, for development and test purpose, a PC with
last generation Centrino or AMD Athlon ?

Thanks

[TimP]

If you have a P4-compatible Celeron, it should be detected correctly by the library and the P4-optimized code path used. Athlon CPUs are likely to be treated as generic Pentium compatible and a non-SSE code path used.
Centrino likewise should be detected as capable of executing SSE parallel instructions in P4-compatible library code.
If you are generating code for Athlon with an Intel 8.1 compiler, it is up to you to pick the correct SSE option. Nearly all Athlons should work with -QxK, SSE2 compatible Athlons also work with -QxW. Run-time detection (e.g. -QaxW) is unlikely to recognize the Athlon as SSE compatible. The Microsoft compilers which support SSE2 likewise don't detect which architecture you have, you simply specify what you want.
The specific compiler options for Banias or Dothan, -QxB or -QaxB, recognize those processors as supporting vectorization but preferring x87 for most scalar floating point operations.