Hi, Having tried out the newssa approach for a bit, and after working on another backend that doesn't use SSA phi nodes, I would like to share some of my personal thoughts with regards to the changes: 1. Firstly I think it is important to retain the existing "pre-phi" linear code. This is fine for backends such as LLVM which do a good job of transforming code anyway. It is also usually less buggy and serves as a baseline for correctness. So after careful consideration I would vote against adding phi instructions in the initial phase. 2. It also appears to me that the original phi instructions, while not textbook correct in terms of placement, are internally consistent, in the sense that the only issue I have found so far is the inability to handle variables that are initialized in a branch of the code. Perhaps this is just a bug? The other issue with the original implementation appears to be related to the patches that attempted to kill things - but I get the feeling that the original design did not envision removing unused things physically, but just marked them logically dead? 3. It also appears that the SSA construction approach by Matthias Braun et al is inadequate in the sense that it is not a full solution and requires additional hacks. This is unfortunate as I understood the aim was to avoid hacks and ad-hoc changes and go with a more published algorithm. In view of this, perhaps the current experiment has failed and the effort should be directed to building the industry standard approach? 4. Final thought is that although I was previously suggesting changes should go into master, I think for such a large piece of fundamental work maybe the changes should be done on a dedicated branch and only merged when completely tested and validated. I hesitated putting forth my thoughts here as I still have the freedom in my project to accept / reject changes happening in Sparse, and so I am able to take a different approach. Also, I am not really qualified to judge the full implications of the new SSA construction approach - hence I am going by my impressions after testing and using the approach, and observing the issues that have been discussed so far on the list. Regards Dibyendu -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
