I do not think Peter was totally clear, if I understand him he is saying: "if you need a routine with optimum/exact timing write it in "C", then work the generated assembler till the timing is right". I have, indeed, used that method on occasion, it has the advantage that all C/assembler interface options are taken care of when you start working the assembler. I once saw someone attempt to "tune" "C" code to match timings, this worked just fine till the next version of the compiler came out. ANSI "C" does not in any way refer to timing, you can be compliant and 100 times slower than another compiler that is also compliant. In other words: if you need to know timing you must use assembler.

Erik