[llvm-commits] [PATCH] Optimize memory usage of MC bundling

[Jim Grosbach]

  Why is the memory usage for the fragments of enough concern to motivate this? Is there really nothing simpler we can do to achieve at least most of the benefits?

  For example, have we run numbers over a significant sample set of code to find out how many fixups there typically are for an MCDataFragment? We currently use "SmallVector<MCFixup, 4> Fixups;". Is 4 the right number to default to here? Likewise, we default to 32 bytes of storage for the contents of the fragment. Is that the most appropriate value for a tradeoff between size and performance here?

  I'm getting more and more concerned about how intrusive this bundling support is becoming into the rest of the MC layer.

http://llvm-reviews.chandlerc.com/D270