IR Transformations

Could anyone give me pointers to papers on the subject of IR
transformations to suit code generation?


More specificaly, suppose we want to generate code for x = 2 * a; The
problem is how to know, at the IR level (before code is generated), that
the above expression can be rewritten as x = a + a (assuming + is cheaper
than *). The problem is that the above transformation may be context
dependent and not always desirable. In other words, sometimes 2*a may be
covered cheaper than a+a, e.g. by a shift left, which some processors
accomplish for free, bundled in a previous instruction.


Thanks for any help


Jim Yu
jimy@eckert.ece.wisc.edu
--

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