On Thursday 25 June 2009 18:41:43 Christian Schladetsch wrote:
In fact, fixed point is not faster on many architectures than floating point. One assumes that it is mainly PSP, DS and Gamecube guys that want fixed-point just for speed these days (I'm looking at you, Kenny ;).
My work requires fixed-point types for modeling cheap embedded microprocessors and other digital signal processing circuits. After using home-made code, code based on Williams' DDJ article and SystemC fixed-point types, I have finally settled on the AlgorithmicC datatypes from Mentor Graphics. A boost implementation would be extremely useful for DSP modeling. For DSP modeling, N-bit x N-bit = N-bit will not suffice. Typically, hardware implementation in critical paths of DSP chips have code equivalent to the following: s[3:-3] x u[3:-4] = s[4:-3] where s[4:-3] indicates an 9-bit signed fixed-point number with bits representing: sign, 2^4, 2^2, ..., 2^--3. In this case, the general-purpose fixed-point library does the following: 8-bit signed * 8-bit unsigned = 16-bit signed -> 9-bit signed where the 7 bits to be removed are found from the specification above. Regards, Ravi