[PLUG] Floating Point arithmetic in the Kernel
guru.is.on.vi at gmail.com
Tue Oct 23 04:24:38 PDT 2007
hope this does not repeat what somebody else may have already said or
is already known.. in that case feel free to ignore
This also may need amendment but here is what i has heard long time back..
logic required to implement a floating point arithmetic mechanism is
complex than that required for processing integers (probably has got
something to do with fitting transistors on thin wafer...) so in order
to avoid exceeding the boundaries permitted with each time technology
upgrade people try to do same arithmetic using integers.
This affects the manufacturing costs of the chip.. Costs in terms of
space on wafer and in terms of few other things so originally the
trade off was to use integer arithmetic to achieve the same result.
It also required CPU more clock cycles to perform floating point
arithmetic than that required for integers.
People have done research (i truely believe) about at what point of
time floating point arithmetic becomes economical from CPU utilization
point of view...
But again as i would like to point out ***I Am Not A Texpert***
technically speaking :-)
At the assembly code (just like finally it boils down to 0.0001 second
in 100 mtr hurdles) it boils down to saving cpu cycles or what can be
making your code executing fast by going to assembly level etc. etc.
etc... (that enough of tech talk for me to get tired for next 10
billion nanosenconds :-) That's when these things matter..
More information about the plug-mail