[PATCH] D52061: [RegisterCoalescer] Only look at main ranges in valuesIdentical/followCopyChain

Mon Sep 17 09:30:50 PDT 2018

qcolombet added a comment.

> But I still believe that trying to spot the identical case in subranges is pointless because we have already proved that the values are identical in main ranges. It is impossible for the values to be not identical in subranges; it's just a bit tricky to fix up the code to demonstrate that.

Essentially what you're saying is that the live-ranges are the same at the main live-range level but we fail to prove it at the subrange level.

What would happen in a situation like this:

  100B %23:sub0 = ...
  110B %23:sub1 = ...
  120B %23:sub2 = ...
  130B %42 = COPY %23
  135B %23:sub2 = ... <-- redefine only a sublane
  140B %51 = COPY %23

In particular, are the main live-range correctly identified as being different?

Basically, I am trying to check that we do prove that the main live-range are indeed identical (i.e., that part "we have already proved that the values are identical in main ranges.").

Cheers,
-Quentin

Repository:
  rL LLVM

https://reviews.llvm.org/D52061