[AMDGPU][SIInsertWaitcnts] don't crash when printing messages at end of block

[ssahasra]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Sameer Sahasrabuddhe (ssahasra)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/145694.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp (+13-2)

diff --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index f43831016952a..98e1696056f11 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -1353,7 +1353,14 @@ bool WaitcntGeneratorPreGFX12::applyPreexistingWaitcnt(
   MachineInstr *WaitcntInstr = nullptr;
   MachineInstr *WaitcntVsCntInstr = nullptr;
 
-  LLVM_DEBUG(dbgs() << "PreGFX12::applyPreexistingWaitcnt at: " << *It);
+  // clang-format off
+  LLVM_DEBUG(dbgs() << "PreGFX12::applyPreexistingWaitcnt at: ";
+             if (It == OldWaitcntInstr.getParent()->instr_end())
+               dbgs() << "end of block\n";
+             else
+               dbgs() << *It;
+             );
+  // clang-format on
 
   for (auto &II :
        make_early_inc_range(make_range(OldWaitcntInstr.getIterator(), It))) {
@@ -1377,6 +1384,7 @@ bool WaitcntGeneratorPreGFX12::applyPreexistingWaitcnt(
 
       // Merge consecutive waitcnt of the same type by erasing multiples.
       if (WaitcntInstr || (!Wait.hasWaitExceptStoreCnt() && TrySimplify)) {
+        LLVM_DEBUG(dbgs() << "erased waitcnt: " << II);
         II.eraseFromParent();
         Modified = true;
       } else
@@ -1507,7 +1515,10 @@ bool WaitcntGeneratorGFX12Plus::applyPreexistingWaitcnt(
   MachineInstr *CombinedStoreDsCntInstr = nullptr;
   MachineInstr *WaitInstrs[NUM_EXTENDED_INST_CNTS] = {};
 
-  LLVM_DEBUG(dbgs() << "GFX12Plus::applyPreexistingWaitcnt at: " << *It);
+  LLVM_DEBUG(dbgs() << "GFX12Plus::applyPreexistingWaitcnt at: ";
+             if (It == OldWaitcntInstr.getParent()->instr_end()) dbgs()
+             << "end of block\n";
+             else dbgs() << *It);
 
   for (auto &II :
        make_early_inc_range(make_range(OldWaitcntInstr.getIterator(), It))) {

[arsenm]

Use LLVM_DEBUG({}) instead

[ssahasra]

Thanks! That looks so much nicer!