In Beitrag "Goertzel-Algorithmus und Signed Fractional Format" I presented a cordic implementation of the atan function for integer arguments. Now I expanded the algorithm to 61-Bit arguments and results referring to Beitrag "Näherung für ArcusTangens?" The precision should be better than 51 Bit. My measure for the precision tests is the IEEE 754 double-type implementation of atan, so its built-in precision of ~51 Bit is the limiting factor. The implemantation is mere C language. It is again cordic type, a 64Bit integer complex number is rotated with known angles from a 62 x 64Bit table. It was a bit tricky, I left the debug stuff for inspection, learning and fun. Comments etc. welcome! It was real big fun! math rulez! Cheers Detlef
Gast
#6028627
This looks interesting since CORDIC is easy to implement into FPGA-Systems. I wonder if the built in CORES can reach that recision. To my understanding this is the case, since they can be configured to nearly any precision and depth.