Cross-level optimization

Algorithm for determinig common-subexpression is well-known, but in case of
presence 20 different common-subexpressions with sophisticated relations we
should decide which to eliminate.
Also depend on target machines some subexpressions should be eliminated and
other shouldn't.
[Example]
x = a + 4 * b;
y = a + 4 * b;


In some machene it would be a good practice to calculate (a+b) only once,
but it produce 3 assignment vs 2 in sample.
The problem is to taking into account target-specific during hi-level
optimization.
This is a common problem but I havn't found any reference for this theme.


I'm working on such algorithm and looking for any literature about this
theme (preferable in e-form) for comparing.




Best regards,
Nikolay Mikhaylov
Astrosoft Research Lab.

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