[PATCH] D111313: [AMDGPU] Preserve MachineDominatorTree in SILowerControlFlow

[Jay Foad via Phabricator via llvm-commits]

foad created this revision.
Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl, arsenm.
foad requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

Updating the MachineDominatorTree is easy since SILowerControlFlow only
ever removes basic blocks. This should save a bit of compile time
because previously we would recompute the dominator tree from scratch
after this pass.

Another reason for doing this is that SILowerControlFlow preserves
LiveIntervals which transitively requires MachineDominatorTree. I think
that means that SILowerControlFlow is obliged to preserve
MachineDominatorTree too as explained here:
https://lists.llvm.org/pipermail/llvm-dev/2020-November/146923.html
although it does not seem to have caused any problems in practice yet.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D111313

Files:
  llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
  llvm/test/CodeGen/AMDGPU/llc-pipeline.ll


Index: llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
+++ llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
@@ -316,7 +316,6 @@
 ; GCN-O1-NEXT:        Eliminate PHI nodes for register allocation
 ; GCN-O1-NEXT:        SI Lower control flow pseudo instructions
 ; GCN-O1-NEXT:        Two-Address instruction pass
-; GCN-O1-NEXT:        MachineDominator Tree Construction
 ; GCN-O1-NEXT:        Slot index numbering
 ; GCN-O1-NEXT:        Live Interval Analysis
 ; GCN-O1-NEXT:        Machine Natural Loop Construction
@@ -600,7 +599,6 @@
 ; GCN-O1-OPTS-NEXT:        Eliminate PHI nodes for register allocation
 ; GCN-O1-OPTS-NEXT:        SI Lower control flow pseudo instructions
 ; GCN-O1-OPTS-NEXT:        Two-Address instruction pass
-; GCN-O1-OPTS-NEXT:        MachineDominator Tree Construction
 ; GCN-O1-OPTS-NEXT:        Slot index numbering
 ; GCN-O1-OPTS-NEXT:        Live Interval Analysis
 ; GCN-O1-OPTS-NEXT:        Machine Natural Loop Construction
@@ -886,7 +884,6 @@
 ; GCN-O2-NEXT:        Eliminate PHI nodes for register allocation
 ; GCN-O2-NEXT:        SI Lower control flow pseudo instructions
 ; GCN-O2-NEXT:        Two-Address instruction pass
-; GCN-O2-NEXT:        MachineDominator Tree Construction
 ; GCN-O2-NEXT:        Slot index numbering
 ; GCN-O2-NEXT:        Live Interval Analysis
 ; GCN-O2-NEXT:        Machine Natural Loop Construction
@@ -1186,7 +1183,6 @@
 ; GCN-O3-NEXT:        Eliminate PHI nodes for register allocation
 ; GCN-O3-NEXT:        SI Lower control flow pseudo instructions
 ; GCN-O3-NEXT:        Two-Address instruction pass
-; GCN-O3-NEXT:        MachineDominator Tree Construction
 ; GCN-O3-NEXT:        Slot index numbering
 ; GCN-O3-NEXT:        Live Interval Analysis
 ; GCN-O3-NEXT:        Machine Natural Loop Construction
Index: llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
+++ llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
@@ -52,6 +52,7 @@
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/CodeGen/LiveIntervals.h"
+#include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 
 using namespace llvm;
@@ -69,6 +70,7 @@
   const SIRegisterInfo *TRI = nullptr;
   const SIInstrInfo *TII = nullptr;
   LiveIntervals *LIS = nullptr;
+  MachineDominatorTree *MDT = nullptr;
   MachineRegisterInfo *MRI = nullptr;
   SetVector<MachineInstr*> LoweredEndCf;
   DenseSet<Register> LoweredIf;
@@ -143,6 +145,7 @@
     // Should preserve the same set that TwoAddressInstructions does.
     AU.addPreserved<SlotIndexes>();
     AU.addPreserved<LiveIntervals>();
+    AU.addPreserved<MachineDominatorTree>();
     AU.addPreservedID(LiveVariablesID);
     MachineFunctionPass::getAnalysisUsage(AU);
   }
@@ -757,6 +760,8 @@
     for (auto &I : MBB.instrs())
       LIS->RemoveMachineInstrFromMaps(I);
   }
+  if (MDT)
+    MDT->eraseNode(&MBB);
   MBB.clear();
   MBB.eraseFromParent();
   if (FallThrough && !FallThrough->isLayoutSuccessor(Succ)) {
@@ -780,6 +785,7 @@
 
   // This doesn't actually need LiveIntervals, but we can preserve them.
   LIS = getAnalysisIfAvailable<LiveIntervals>();
+  MDT = getAnalysisIfAvailable<MachineDominatorTree>();
   MRI = &MF.getRegInfo();
   BoolRC = TRI->getBoolRC();
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D111313.377841.patch
Type: text/x-patch
Size: 3451 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211007/af26b676/attachment.bin>