Control Dependence Analysis Pass

[AyaElAkhras]

Added a ControlDependenceAnalysis pass and instantiated it in a Transform pass called FuncSSAToGSA

[AyaElAkhras]

Hi Jiahui, thanks again for your feedback. I accounted for most of your suggestions and marked them as resolved :) I left a comment for the suggestions of replacing vectors with sets above. I'm missing a few more comments that I left unresolved but will try to come back to later...

[lucas-rami]

Is this ready for review? If so then conflicts with main need to be resolved. If a smaller contribution that does not conflict can be extracted from the current state of the branch (e.g., just the new analysis) then feel free to remove the other stuff from this branch.

[lucas-rami]

Conflicts with main need to be resolved before I review this.

[AyaElAkhras]

I deleted conflicting files and limited the contribution of this PR to one analysis pass that adds a source file (ControlDependenceAnalysis.cpp), a header file (ControlDependenceAnalysis.h) and the updated CMakeLists.txt

[lucas-rami]

Left a bunch of comments, mostly cosmetics/optimizations except a few on data representation. Nothing that should be too long to address though.

As @Jiahui17 pointed out initially, the code is not clang-tidyed, please install the necessary tool and run it on your code (VSCode extension).

[AyaElAkhras]

Accounted for the cosmetic stuff and left one comment about deleting the getter functions. Will get to the rest of the comments later.

[lucas-rami]

I realize that Jiahui pointed out the vector to set thing as well and that you were worried that doing so would prevent you from easily modifying elements in your data-structure. It's in fact easier to modify in a set than a vector as well as being O(1) instead of O(n) in general, see example below.

mlir::DenseSet<Block*> blocks;
// ... do something to the set ...

// Removing
Block* toRemove = ...;
if (blocks.erase(toRemove)) {
  // The removed block was part of the set and was removed
} else {
  // The removed block was *not* part of the set, nothing happened
}

// Inserting
Block* toInsert = ...;
if (auto [_, newElem] = blocks.insert(toInsert); newElem) {
  // The inserted block was not part of the set and was inserted
} else {
  // The inserted block was *already* part of the set, nothing happened
}

[AyaElAkhras]

Accounted for all comments and retested the implementation. There are some remaining warnings from clang-tidy, but I do not have time to resolve them now. It seems to me that they have to do with variable names--will check later.

[lucas-rami]

Where does this stand w.r.t. #177? Is this stale and should be closed because it will reimplemented differently? Or should it just be pushed to the finish line?

[pcineverdies]

You are right, when dealing with #177 we did not think about this analysis pass. Since a few modifications have been done in the meanwhile, I am going to open a new PR accordingly.