[AMDGPU][Scheduler] Fix compilation fail in EXPENSIVE_CHECKS

[lucas-rami]

Bug introduced by #182853 (Remat is now a pointer).

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Lucas Ramirez (lucas-rami)

Changes

Introduced by #182853 (Remat is now a pointer).

---

Full diff: https://github.com/llvm/llvm-project/pull/183745.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp (+1-1)

diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
index f7ce2272d1149..77c322eb3178e 100644
--- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
@@ -3003,7 +3003,7 @@ MachineInstr *PreRARematStage::ScoredRemat::rematerialize(
     if (LI.hasSubRanges() && MO.getSubReg())
       LM = DAG.TRI->getSubRegIndexLaneMask(MO.getSubReg());
 
-    LaneBitmask LiveInMask = DAG.LiveIns[Remat.UseRegion].at(UseReg);
+    LaneBitmask LiveInMask = DAG.LiveIns[Remat->UseRegion].at(UseReg);
     LaneBitmask UncoveredLanes = LM & ~(LiveInMask & LM);
     // If this register has lanes not covered by the LiveIns, be sure they
     // do not map to any subrange. ref: