16-bit fixed-point math library for the C6416?

Thanks, but I already have the source for IQmath which I got from TI and which I am currently using on the C6416.  However, it is a 32-bit implementation and uses some intrinsics that are not supported by the C6416 so they need to be simulated.  As a result the fixed-point math is running even slower than the floating-point emulation. 

I am looking for a 16-bit implementation of the same thing which should run much faster on the C6416. 

Thanks,

Susan

Susan, what function are you talking about here? What intrinsic you are mentioning. I can probably try to suggest better way of implementing the operation. Also note, instead of making function call to the C implementation of intrinsic, you can inline the intrinsic C function. Or simply, you can copy the C code for th eintrinsic implementation and include as part of the IQMath function source. That should reduce the overhead.

Please make sure you compile the source with -o3 compiler option as well to get performance

Gagan

Thanks for the reply.  I contacted TI via email for the source code to the IQmath library, which they provided.  However, my C6416 device was missing some of the C64x+ intrinsics used by the source code (in particular _dmv, _mpy32, _mpy32ll, _mpy32su, _mpy32u, and _mpy32us).  I was told to copy the source code for these intrinsics from http://processors.wiki.ti.com/index.php/Run_Intrinsics_Code_Anywhere, which I did.  Unfortunately this implementation runs very slow on the C6416 and is not fast enough for my application.  So I was hoping to find a faster, 16-bit implementation of a fixed-point library that could be used instead.  Do you have a suggestion for another way to speed these up?

Thanks for the advice on compiler optimization.  That certainly helped and although it is running faster than the floating-point emulation version it is still too slow for my application.

Thanks,

Susan