importing code in IR (GIMPLE and/or RTL)

[Andrew Haley <aph at redhat dot com>]

Luchezar Belev writes:

 > I want to toss an idea that someone might find usefull.
 > It looks to me that it would be quite interesting if we could
 > do the inverse of the debugging dumps. The dumps of the GIMPLE
 > code are very usefull when one is writting a new GIMPLE optimizing
 > pass because thus he is able to skip all the following passes and
 > ovserve the direct result of the concrete pass. The inverse would
 > be to directly feed the given pass with code in GIMPLE repr., skipping
 > anything before the pass. This way he could much easely analyse
 > the exact behaviour of the pass without the noise added by any
 > prior or following passes.
 > 
 > The same would probably be usefull for the RTL-based passes as well -
 > not only to be able to dump the RTL at given point, but also to be
 > able to import at given point. In such case all prior passes can
 > just be skipped.

This idea comes up every once in a while, often enough to be a FAQ.

The problem is that what you are suggesting might allow someone to
write an unfree front end and feed its output into gcc's back end.
Although anyone is free to make this modification to their own copy of
gcc, we won't do it.

Another problem is that the gimple dumps don't contain everything.

Andrew.