[PATCH] D133777: [SelectOpti] Fix lifetime intrinsic bug

[Sotiris Apostolakis via Phabricator via llvm-commits]

apostolakis marked an inline comment as done.
apostolakis added inline comments.


================
Comment at: llvm/lib/CodeGen/SelectOptimize.cpp:432
+    // case any use is sinked).
+    SmallVector<Instruction *, 2> EndLifetimeCalls;
+    for (Instruction &I : *StartBlock) {
----------------
davidxl wrote:
> is it better to track the life end variable and inserted after the target instruction?
Yes as you said, the potential gains are small since the code region is just a single split basic block. 

Also note that, to begin with, currently-emitted lifetimes intrinsics are not very tight to the actual lifetime of objects.
This is probably because other optimizations are also conservative when placing them and moving them . 
For example, for the case that caused the bug, the lifetime ended 4 instructions after the final use of the variable. 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D133777/new/

https://reviews.llvm.org/D133777