Following are some general [performance] issues to be … -- Unroll loops. … -- Avoid intrinsically expensive instructions … For example, on a Pentium, a variable rotate/shift … requires four clocks and cannot be paired with … -- Limit the number of variables. … -- Limit table size. … -- Allow parallelism. The general idea is to … [concurrently] … [p. 246] -- Allow setup time for table indexing … . [p. 248] [For example,] use the previous value of … for lookahead computation.
Google-1
Google-2
Copyright clearance needed for quotation.