[PATCH] D88397: [SplitKit] Cope with no live subranges in defFromParent

[Quentin Colombet via Phabricator via llvm-commits]

qcolombet accepted this revision.
qcolombet added a comment.

Hi @foad,

LGTM.
The test case could be more descriptive though, see my inline comment.

Cheers,
-Quentin



================
Comment at: llvm/test/CodeGen/AMDGPU/splitkit-nolivesubranges.mir:3
+# RUN: llc -march=amdgcn -run-pass=greedy,virtregrewriter -verify-regalloc %s -o - | FileCheck %s
+
+---
----------------
Could you add a comment on the characteristic of the test?
For instance, "This test aims at triggering a live-range splitting in a place when %0 subranges are all dead, but the main live-range of %0 is still alive. %0 main range is kept alive simply by not using undef. Then, the splitting is triggered by creating two points of high register pressure:
1. One where %0 main range in the only live live-range: This will force the insertion of a split for %0 main range
2. One where %0.subrange is live, so that the interference check in regalloc trigger the splitting heuristic"



Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D88397/new/

https://reviews.llvm.org/D88397