topic Re: questions about floating point in Intel® Integrated Performance Primitives

questions about floating point

enriparm — Fri, 27 Jan 2006 00:15:03 GMT

Hi.
I'd like to have some clarifications about floating point.
Now we're using library IPP 5.0 for IXP but our target is the Intel Xscale 255 ( the version of the IPP is only for 250 ).

We have problems linking our app with the library IPP caused by the VFP flag on the library. Now we're looking for a toolchain that supports these flag.
So i have some questions:

-Does IXP 4xx CPU implement the VFP or FPA floating point instructions? And the Xscale 255?
-Is the code of the lib IPP optimized for the FPA or VFP instructions? We opened the code but we haven't seen any FP instructions.
-Is it possible that in future Intel will release other versions of IPP that optimizes the code with FPA or VFP instructions?
Or other versions optimized for the PXA Xscale ( but without WMMX instructions)?

Re: questions about floating point

Vladimir_Dudnik — Sat, 04 Feb 2006 01:28:05 GMT

Hi,

We do not use VFP instrutions, usually the reason of that message is not appropriate tool chain. Did you look at IPP for IXP release notes file?

According its software requirements section you should use MontaVista gcc compiler:

Software Requirements to Develop Applications for the IXP4XX product line

Linux system with glibc 2.2.4, 2.2.5, 2.2.93, 2.3.2 or 2.3.3 and the 2.4.X or 2.6.X Linux kernel as represented by the following distributions. Note: Not all distributions listed are validated and not all distributions are listed.
- Red Hat* Linux 7.3, 8, 9
- Red Hat Enterprise Linux* 2.1, 3
- SUSE* Linux 8.2, 9.1
- SUSE Linux Enterprise Server* 8 or 9
A supported C compiler (Intel IPP has been tested with the following):
- iwmmxt_le-gcc (MontaVista*) for little endian applications
- xscale_be-gcc (MontaVista*) for big endian applications

Regards,
Vladimir