TI DSP's Q float system

1. It's not floating point, it's fixed.

1. This would be a good question for comp.dsp.

2. It's mostly implemented as a convention, although there's a C library that supports it. In assembly you get to set the radix any darn way you want. The MAC instruction has a built-in shift, but I don't know if it's used for these operations (that's tool chain specific).

1. If you can't find stuff in TI's DSP documentation join the club. It's usually there, but sometimes it seems like it'd be quicker to build your own chip starting from sand than to find it in their documentation.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

...

Sand? Nah, let somebody else do the sand part:

?sGlobalNavPick=PRODUCTS&sSecondaryNavPick=Intellectual+Property&_sortSpec=Vendor%20desc&searchJSP=%2Fxebiz%2Fsearch%2Fipsrch.jsp&_iPageNumber=1&resultNumber=25&_Vendor_1=Xilinx&_Vendor_2=%3Cnot%3EXilinx&_ResultsView=Standard&_IPCategory=Digital+Signal+Processing&_IPProducts_1129=Core (mind the wrap)

;-)

Cheers! Rich

I think you are referring to TI's fixed point math library

is sort of a bridge between writing pure fixed point code with all the headaches and using a pure software floating point library (slow but easy to write).

Its about 4-5 times slower then pure fixed point (if I remember correctly), but much faster then floating point emulation (20-100 slower then fixed point depending on the algorithm your implementing). It still requires to figure out what the dynamic range is for your signals and choose an appropriate "Q" factor.

Just my opinion but for me, its the worse of both worlds, slow and still have to worry about dynamic ranges, but others think is the best of both worlds, fast and little scaling worries.

Hi Steve,

I just want to use the image lib provide by TI C55x. And it use "Q", as you said it is a math lib, I see.

Thanks! Davy