[PATCH] D33577: CodeGen: BlockPlacement: Use Branching factor to choose between near equals.

[David Li via Phabricator via llvm-commits]

davidxl added a comment.

Looking at the example. Comparing two layout decisions: 1) always pick the next test as layout successor ; and 2) always pick the  (else) if-then block  as the layout successor.  
A) the number of taken branches is the same
B) assuming 50% prob for all branches,   the dynamic count of taken branches with 2)  is actually smaller.

So I am not convinced this is a right heuristic unless more data is provided.

Here is the suggestion. You can write up a microbenchmark  with this CFG shape. Make all the branch's real probability to be 50% (using controlled by random data make branch less predictable).  Using __buitlin_expect to force the layout to be either 1) or 2) described above and compare 
a) runtime performance
b) taken branches with PMU.


Repository:
  rL LLVM

https://reviews.llvm.org/D33577