Allow machine dce to remove uses in the same instruction

[rampitec]

Summary

Machine DCE cannot remove a dead definition if there are non-dbg uses.
A use however can be in the same instruction:

dead %0 = INST %0

Such instructions sometimes created by Detect dead lanes pass.

Allow this instruction to be deleted despite the use if the only use
belongs to the same instruction.

[arsenm]

lib/CodeGen/DeadMachineInstructionElim.cpp

84–85 ↗

(On Diff #191401)

This needs a search limit in case of huge numbers of uses?
This can also use use_nodbg_instructions

[rampitec]

Use use_nodbg_instructions().

lib/CodeGen/DeadMachineInstructionElim.cpp

84–85 ↗

(On Diff #191401)

It does need a limit. The only way it can run into a very long loop if all huge amount of uses will belong to the same instruction. Otherwise it will immediately return false at the first use from any other place.

[rampitec]

lib/CodeGen/DeadMachineInstructionElim.cpp

84–85 ↗

(On Diff #191401)

I mean it does not need a limit.

[arsenm]

LGTM

llvm/trunk/lib/CodeGen/DeadMachineInstructionElim.cpp

for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
     if (MO.isReg() && MO.isDef()) {
       unsigned Reg = MO.getReg();
       if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
         // Don't delete live physreg defs, or any reserved register defs.
         if (LivePhysRegs.test(Reg) || MRI->isReserved(Reg))
           return false;
       } else {
-        if (!MRI->use_nodbg_empty(Reg))
+        for (const MachineInstr &Use : MRI->use_nodbg_instructions(Reg)) {
+          if (&Use != MI)
             // This def has a non-debug use. Don't delete the instruction!
             return false;
         }
       }
     }
+  }
 
   // If there are no defs with uses, the instruction is dead.
   return true;
 }
 
 bool DeadMachineInstructionElim::runOnMachineFunction(MachineFunction &MF) {
   if (skipFunction(MF.getFunction()))
     return false;

llvm/trunk/test/CodeGen/AMDGPU/dead-mi-use-same-intr.mir

+# RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -run-pass dead-mi-elimination -o - %s | FileCheck -check-prefix=GCN %s
+
+# GCN-LABEL: name: dead_undef
+# GCN:      bb.0:
+# GCN-NEXT: S_ENDPGM 0
+---
+name:            dead_undef
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vgpr_32 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body: |
+  bb.0:
+  %0 = IMPLICIT_DEF
+  %1 = IMPLICIT_DEF
+  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, undef %2:vgpr_32, implicit $exec
+  S_ENDPGM 0
+
+# GCN-LABEL: name: dead_defined
+# GCN:      bb.0:
+# GCN-NEXT: S_ENDPGM 0
+---
+name:            dead_defined
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vgpr_32 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body: |
+  bb.0:
+  %0 = IMPLICIT_DEF
+  %1 = IMPLICIT_DEF
+  %2 = IMPLICIT_DEF
+  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, %2:vgpr_32, implicit $exec
+  S_ENDPGM 0
+
+# Probably this dead mac can be removed anyway.
+# GCN-LABEL: name: dead_def_live_use
+# GCN:      dead %2:vgpr_32 = V_MAC_F32_e32 %0, %1, %2, implicit $exec
+---
+name:            dead_def_live_use
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vgpr_32 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+body: |
+  bb.0:
+  %0 = IMPLICIT_DEF
+  %1 = IMPLICIT_DEF
+  %2 = IMPLICIT_DEF
+  GLOBAL_STORE_DWORD undef $vgpr0_vgpr1, %2, 0, 0, 0, implicit $exec
+  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, %2:vgpr_32, implicit $exec
+  S_ENDPGM 0