[PATCH] D105836: [AMDGPU] Add TTI::isAmbientPhysRegUse() to allow VOP rematerilization

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

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

Any def of EXEC prevents rematerialization of any VOP instruction
because of the physreg use. Create a callback to check for the
ambient physreg use and allow rematerialization.


https://reviews.llvm.org/D105836

Files:
  llvm/include/llvm/CodeGen/TargetInstrInfo.h
  llvm/lib/CodeGen/LiveRangeEdit.cpp
  llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
  llvm/lib/Target/AMDGPU/SIInstrInfo.h
  llvm/test/CodeGen/AMDGPU/remat-vop.mir


Index: llvm/test/CodeGen/AMDGPU/remat-vop.mir
===================================================================
--- llvm/test/CodeGen/AMDGPU/remat-vop.mir
+++ llvm/test/CodeGen/AMDGPU/remat-vop.mir
@@ -52,6 +52,31 @@
     S_ENDPGM 0
 ...
 ---
+name:            test_remat_v_mov_b32_e32_exec_def
+tracksRegLiveness: true
+machineFunctionInfo:
+  stackPtrOffsetReg:  $sgpr32
+body:             |
+  bb.0:
+    ; GCN-LABEL: name: test_remat_v_mov_b32_e32_exec_def
+    ; GCN: renamable $vgpr0 = V_MOV_B32_e32 1, implicit $exec
+    ; GCN: renamable $vgpr1 = V_MOV_B32_e32 2, implicit $exec
+    ; GCN: S_NOP 0, implicit killed renamable $vgpr0
+    ; GCN: S_NOP 0, implicit killed renamable $vgpr1
+    ; GCN: renamable $vgpr0 = V_MOV_B32_e32 3, implicit $exec
+    ; GCN: S_NOP 0, implicit killed renamable $vgpr0
+    ; GCN: $exec = S_ANDN2_B64_term $exec, undef renamable $sgpr0_sgpr1, implicit-def $scc
+    ; GCN: S_ENDPGM 0
+    %0:vgpr_32 = V_MOV_B32_e32 1, implicit $exec
+    %1:vgpr_32 = V_MOV_B32_e32 2, implicit $exec
+    %2:vgpr_32 = V_MOV_B32_e32 3, implicit $exec
+    S_NOP 0, implicit %0
+    S_NOP 0, implicit %1
+    S_NOP 0, implicit %2
+    $exec = S_ANDN2_B64_term $exec, undef %4:sreg_64, implicit-def $scc
+    S_ENDPGM 0
+...
+---
 name:            test_remat_v_mov_b32_e64
 tracksRegLiveness: true
 body:             |
Index: llvm/lib/Target/AMDGPU/SIInstrInfo.h
===================================================================
--- llvm/lib/Target/AMDGPU/SIInstrInfo.h
+++ llvm/lib/Target/AMDGPU/SIInstrInfo.h
@@ -181,6 +181,8 @@
   bool isReallyTriviallyReMaterializable(const MachineInstr &MI,
                                          AAResults *AA) const override;
 
+  bool isAmbientPhysRegUse(const MachineOperand &MO) const override;
+
   bool areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
                                int64_t &Offset1,
                                int64_t &Offset2) const override;
Index: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -126,6 +126,11 @@
   }
 }
 
+bool SIInstrInfo::isAmbientPhysRegUse(const MachineOperand &MO) const {
+  // And implicit use of exec is not a real register read.
+  return MO.getReg() == AMDGPU::EXEC && MO.isImplicit();
+}
+
 bool SIInstrInfo::areLoadsFromSameBasePtr(SDNode *Load0, SDNode *Load1,
                                           int64_t &Offset0,
                                           int64_t &Offset1) const {
Index: llvm/lib/CodeGen/LiveRangeEdit.cpp
===================================================================
--- llvm/lib/CodeGen/LiveRangeEdit.cpp
+++ llvm/lib/CodeGen/LiveRangeEdit.cpp
@@ -115,7 +115,7 @@
 
     // We can't remat physreg uses, unless it is a constant.
     if (Register::isPhysicalRegister(MO.getReg())) {
-      if (MRI.isConstantPhysReg(MO.getReg()))
+      if (MRI.isConstantPhysReg(MO.getReg()) || TII.isAmbientPhysRegUse(MO))
         continue;
       return false;
     }
Index: llvm/include/llvm/CodeGen/TargetInstrInfo.h
===================================================================
--- llvm/include/llvm/CodeGen/TargetInstrInfo.h
+++ llvm/include/llvm/CodeGen/TargetInstrInfo.h
@@ -129,6 +129,14 @@
              isReallyTriviallyReMaterializableGeneric(MI, AA)));
   }
 
+  /// Given \p MO is a PhysReg use return if it is a real or ambient use.
+  /// An ambient use is to model the execution environment not affecting
+  /// an instruction execution itself. Used to allow rematerialization of
+  /// an instruction using a predicate like register.
+  virtual bool isAmbientPhysRegUse(const MachineOperand &MO) const {
+    return false;
+  }
+
 protected:
   /// For instructions with opcodes for which the M_REMATERIALIZABLE flag is
   /// set, this hook lets the target specify whether the instruction is actually


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D105836.358031.patch
Type: text/x-patch
Size: 3934 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210712/4d326b5e/attachment-0001.bin>