We have a Sharc21489 ezboard and have built several projects written in floating-point codes. Currently, we want to rewrite these codes in fixed-point and want to know whether the fixed-point arithmetic and codes are supported by AD21489. Otherwise we may need a platform based on Blackfin processor.

As we know TI's 6747 processor suppots both fixed- and floating-point arithmetic, how about AD21489?

Many thanks!

All SHARCs are both fixed and floating point processors. In fact, it is a very good fixed point processor.

It is often true that we tend to think of a SHARC as a floating point DSP since floating point emulation is incredibly inefficient. This becomes a major distinction when comparing pure fixed point processors.

The SHARC uses 32 bit words for data with an 80 bit accumulator for multiplication. This makes it's single precision multiplier essentially the same as double precision in a Blackfin and many other processors. The is even a double precision ap note for SHARC using 64 bit data.

The SHARC has special instructions that convert from fixed to float and the converse. I have written applications that routinely take advantage of both since some things work better with numbers represented in one format versus the other. With floating point, you get dynamic range and easy scaling. With fixed point you get precision.

Many years ago, there were AES journal papers about fixed versus float, that I found as thinly veiled marketing for the old Motorola (Freescale) 56K versus SHARC. The 56K was a 24 bit fixed processor. I always found this amusing since the SHARC is also a better fixed point processor than the 56K (8 bits more precision than the 56K).

So jump in, the 21489 will work fine as a fixed point DSP.

Al Clark

www.danvillesignal.com