Re: Run time optimizations

The technique of generating multiple posible implementations of a
particular piece of source, and then choosing the correct one at run time
based on the actual circumstances present at the time is fairly standard
in high performance vectorising compilers. These often generate both
vector and scalar code and then choose which to use based on the loop
length at run time. The technique is also used to allow vectorisation of
operations which may have a data dependancy for some values of a variable,
but not for others e.g.


do i = 2,10
a(i+j) = a(i)+1
end do


which certainly vectorises if j <= 0 or j >= 9 but does not for 1 <= j <= 8


-- Jim
James Cownie
Meiko Limited, 650 Aztec West, Bristol BS12 4SD, England
Meiko Inc., Reservoir Place, 1601 Trapelo Road, Waltham MA 02154
Phone : +44 454 616171 or +1 617 890 7676
FAX : +44 454 618188 or +1 617 890 5042
E-Mail: jim@meiko.co.uk or jim@meiko.com
--

Return to the comp.compilers page.
Search the comp.compilers archives again.