Reciprical float and rsqrt estimation docs and code / discussion
recip.cc - Andrew's code for generating and testing hard coded n-by-n Look Up Table (LUT)approximations.
revised version accepts arguments for LUT size :
parameters seperated by spaces are
optional index-size(default 7 in bits), estimate size (both in bits
these can be followed by\n --verilog to print out LUT table definition
--test to test minimal range that covers the LUT table (default) -or-
--test-long which tests all single float non-NaN values.
Also added is a --nosum directive that suppresses print of informitive and summary parameter info.
and --detail prints LUT entries with details on float values and errors per entry