- Ubuntu 12.04 (use AWS free tier search AMIs)
- llvmpy (install from source)
- clang-3.3 (use apt)
(Useful) combinations of rendering options:
--control (default):
Renders both explicit control flow present in LLVM IR (sequential between statements in basic blocks, jumps between basic blocks), and dataflow dependencies. Control flow has higher weight (that means that control flow edges tend to be more straight).
--dag-control:
Ignore explicit flow control present in LLVM IR and instead compute order of evaluation of independent (i.e. disconnected) dataflow DAGs within basic block. Root node of a DAG consider to be an instruction of type void. (The idea is that void instruction is executed solely for side effect, and then it must be last instruction in evaluation of some DAG, i.e. its root. This is clearly a heuristic, which needs to be tested on various inputs yet.)
--block
For both options above, you can add --block to clusterize instructions of the same basic block together within a rectangle block. This seems like natural way to do it, but leaves questions open where to put leaves of dataflow graphs (i.e. variables, constants, etc.) So far, these are rendered as DAG structure also, which means they are not part of any basic block cluster. But rendering them in such way leads to edges going from different basic blocks to the same leaf nodes, leading to a mess in the graph. Possible other options: duplicate leaf nodes; don't render at all (can be kinda assumed).
--block-edges
This makes control edges between basic blocks actually go between basic blocks, not specific instructions in them. This may be useful for some kinds of presentations. This also removes extra nodes to represent labels. Results of the latter changes are mixed though, it leads to not ideal placing of leaf non-cluster nodes and thus deformed graphs.