[Open-graphics] Multipliers in oga1hq
urkedal at nbi.dk
Sat Sep 1 18:59:19 EDT 2007
On 2007-09-01, Farhan Mohamed Ali wrote:
> > So, let's consider integrating Farhan's version in the nanocontroller.
> > Given that the VGA code will use 16 bit, would it be better to reduce the
> > multiplier to 16x16->32? Will this be insufficient for the DMA code?
> > (Does DMA require multiply at all, other than powers of 2?) Conversely is
> > 33 cycles multiply to slow for the VGA code, and would 17 cycles be fast
> > enough?
> 16x16 will take 9 cycles. Perhaps it could be fast enough to be clocked
> at 200mhz if it is a dedicated 16x16 part. I'll try this out. I will also
> try adding a special mode to the current 32x32 version that assumes 16
> bit inputs and takes 9 cycles to complete for that mode.
To complete your corrections, the 32x32 radix-4 multiplier is 17 cycles,
not 33 as I said. I think a 16x16 mode is more interesting than
early-termination. Now, we're also getting down to quite manageable
delays, as seen from the programmer trying to fill in the instruction
slots. This also means the cycles used on I/O become more important.
More information about the Open-graphics