Re: Justifying Optimization

Did you heard about a compiler (even a software or a processor) that
is bug-free ?


Very often, the "bugs" introduced by the optimizing compiler are
caused by bad programming styles. A lot of programmers (especially in
C and fortran) make assumptions about how the compiler would generate
the code, and do not follow the semantics of the language. This is
because some programmers use high level languages and think in a low
level way to hand tune their application.


Example :
float A[10], B[10];


for (i=0; i<20;i++){
A[i]=.... /* the programmer assumes A and B contiguous in memory : use A
as an offset to access B */
}
for (i=0; i<10;i++){
...=B[i]
}
This simple code woul certainly run well if optimization is turned off.
When switching to on, expect to have uncorrect results. Now, who did a
bad job ? the programmer or the compiler ?


BTW, there are numerous formal high level languages that promise you
zero bugs, even with optimizations.


Sid

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