[LiveVariables] Mark use as implicit-def if defined at instr

llvm/include/llvm/CodeGen/LiveVariables.h

@@ -164,10 +164,8 @@ class LiveVariables {
   MachineInstr *FindLastRefOrPartRef(Register Reg);
 
   /// FindLastPartialDef - Return the last partial def of the specified
-  /// register. Also returns the sub-registers that're defined by the
-  /// instruction.
-  MachineInstr *FindLastPartialDef(Register Reg,
-                                   SmallSet<Register, 4> &PartDefRegs);
+  /// register.
+  MachineInstr *FindLastPartialDef(Register Reg);
 
   /// analyzePHINodes - Gather information about the PHI nodes in here. In
   /// particular, we want to map the variable information of a virtual

llvm/lib/CodeGen/LiveVariables.cpp

@@ -214,11 +214,7 @@ void LiveVariables::HandleVirtRegDef(Register Reg, MachineInstr &MI) {
 }
 
 /// FindLastPartialDef - Return the last partial def of the specified register.
-/// Also returns the sub-registers that're defined by the instruction.
-MachineInstr *
-LiveVariables::FindLastPartialDef(Register Reg,
-                                  SmallSet<Register, 4> &PartDefRegs) {
-  Register LastDefReg = 0;
+MachineInstr *LiveVariables::FindLastPartialDef(Register Reg) {
   unsigned LastDefDist = 0;
   MachineInstr *LastDef = nullptr;
   for (MCPhysReg SubReg : TRI->subregs(Reg)) {
@@ -227,7 +223,6 @@ LiveVariables::FindLastPartialDef(Register Reg,
       continue;
     unsigned Dist = DistanceMap[Def];
     if (Dist > LastDefDist) {
-      LastDefReg  = SubReg;
       LastDef     = Def;
       LastDefDist = Dist;
     }
@@ -236,14 +231,6 @@ LiveVariables::FindLastPartialDef(Register Reg,
   if (!LastDef)
     return nullptr;
 
-  PartDefRegs.insert(LastDefReg);
-  for (MachineOperand &MO : LastDef->all_defs()) {
-    if (MO.getReg() == 0)
-      continue;
-    Register DefReg = MO.getReg();
-    if (TRI->isSubRegister(Reg, DefReg))
-      PartDefRegs.insert_range(TRI->subregs_inclusive(DefReg));
-  }
   return LastDef;
 }
 
@@ -262,27 +249,11 @@ void LiveVariables::HandlePhysRegUse(Register Reg, MachineInstr &MI) {
     // ...
     //    = EAX
     // All of the sub-registers must have been defined before the use of Reg!
-    SmallSet<Register, 4> PartDefRegs;
-    MachineInstr *LastPartialDef = FindLastPartialDef(Reg, PartDefRegs);
+    MachineInstr *LastPartialDef = FindLastPartialDef(Reg);
     // If LastPartialDef is NULL, it must be using a livein register.
     if (LastPartialDef) {
-      LastPartialDef->addOperand(MachineOperand::CreateReg(Reg, true/*IsDef*/,
-                                                           true/*IsImp*/));
-      PhysRegDef[Reg.id()] = LastPartialDef;
-      SmallSet<MCPhysReg, 8> Processed;
-      for (MCPhysReg SubReg : TRI->subregs(Reg)) {
-        if (Processed.count(SubReg))
-          continue;
-        if (PartDefRegs.count(SubReg))
-          continue;
-        // This part of Reg was defined before the last partial def. It's killed
-        // here.
-        LastPartialDef->addOperand(MachineOperand::CreateReg(SubReg,
-                                                             false/*IsDef*/,
-                                                             true/*IsImp*/));
-        PhysRegDef[SubReg] = LastPartialDef;
-        Processed.insert_range(TRI->subregs(SubReg));
-      }
+      LastPartialDef->addOperand(
+          MachineOperand::CreateReg(Reg, true /*IsDef*/, true /*IsImp*/));
     }
   } else if (LastDef && !PhysRegUse[Reg.id()] &&
              !LastDef->findRegisterDefOperand(Reg, /*TRI=*/nullptr))

llvm/test/CodeGen/AArch64/ldrpre-ldr-merge.mir

@@ -756,7 +756,7 @@ body:             |
     ; CHECK: liveins: $x0, $x1, $x2
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: early-clobber renamable $x1, renamable $x0 = LDRSWpre renamable $x1, 40, implicit $w1, implicit $w1_hi :: (load (s32))
-    ; CHECK-NEXT: renamable $w2 = LDRWui renamable $x1, 1, implicit-def $x2, implicit $w2_hi :: (load (s32))
+    ; CHECK-NEXT: renamable $w2 = LDRWui renamable $x1, 1, implicit-def $x2 :: (load (s32))
     ; CHECK-NEXT: STPXi renamable $x0, renamable $x2, renamable $x1, 0 :: (store (s64))
     ; CHECK-NEXT: RET undef $lr
     early-clobber renamable $x1, renamable $x0 = LDRSWpre killed renamable $x1, 40 :: (load (s32))

llvm/test/CodeGen/AMDGPU/fncall-implicitdef.ll

@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -O1 %s -o - | FileCheck %s
+
+define amdgpu_ps <4 x float> @caller(ptr %ptr) {
+; CHECK-LABEL: caller:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    flat_load_dword v1, v[0:1]
+; CHECK-NEXT:    s_mov_b32 s0, 0
+; CHECK-NEXT:    s_mov_b32 s1, 0
+; CHECK-NEXT:    s_mov_b32 s2, 0
+; CHECK-NEXT:    s_mov_b32 s5, fn@abs32@hi
+; CHECK-NEXT:    s_mov_b32 s4, fn@abs32@lo
+; CHECK-NEXT:    s_mov_b64 s[8:9], 0
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    s_mov_b32 s3, 0
+; CHECK-NEXT:    v_mov_b32_e32 v2, 0
+; CHECK-NEXT:    s_mov_b32 s32, 0
+; CHECK-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; CHECK-NEXT:    ; return to shader part epilog
+  %L = load i32, ptr %ptr, align 4
+  %R = call <4 x float> @fn(<4 x i32> zeroinitializer, i32 0, i32 %L, i32 0)
+  ret <4 x float> %R
+}
+
+declare hidden <4 x float> @fn(<4 x i32> inreg, i32, i32, i32)

llvm/test/CodeGen/AMDGPU/livevars-implicitdef.mir

@@ -0,0 +1,91 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn --run-pass=livevars -o - %s | FileCheck %s
+---
+# Check that super register is defined for an sgpr copy.
+name:            sgpr_copy
+tracksRegLiveness: true
+body:             |
+  bb.0:
+
+    ; CHECK-LABEL: name: sgpr_copy
+    ; CHECK: %sval:sreg_32 = S_MOV_B32 0
+    ; CHECK-NEXT: $sgpr0 = COPY %sval
+    ; CHECK-NEXT: $sgpr1 = COPY %sval
+    ; CHECK-NEXT: $sgpr2 = COPY %sval
+    ; CHECK-NEXT: $sgpr3 = COPY killed %sval
+    ; CHECK-NEXT: SI_RETURN implicit killed $sgpr0_sgpr1_sgpr2_sgpr3
+    %sval:sreg_32 = S_MOV_B32 0
+
+    $sgpr0 = COPY %sval
+    $sgpr1 = COPY %sval
+    $sgpr2 = COPY %sval
+    $sgpr3 = COPY %sval
+    SI_RETURN implicit $sgpr0_sgpr1_sgpr2_sgpr3
+
+...
+---
+# Check that super register is defined for a vgpr vector copy.
+name:            vgpr_copy
+tracksRegLiveness: true
+body:             |
+  bb.0:
+
+    ; CHECK-LABEL: name: vgpr_copy
+    ; CHECK: %vval:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    ; CHECK-NEXT: $vgpr0 = COPY %vval
+    ; CHECK-NEXT: $vgpr1 = COPY %vval
+    ; CHECK-NEXT: $vgpr2 = COPY %vval
+    ; CHECK-NEXT: $vgpr3 = COPY killed %vval
+    ; CHECK-NEXT: dead [[COPY:%[0-9]+]]:vgpr_32 = COPY killed $vgpr0_vgpr1_vgpr2_vgpr3
+    %vval:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+
+    $vgpr0 = COPY %vval
+    $vgpr1 = COPY %vval
+    $vgpr2 = COPY %vval
+    $vgpr3 = COPY %vval
+    %0:vgpr_32 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+
+...
+---
+# Check that super register is defined when there is a hole.
+name:            sgpr_copy_hole
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: sgpr_copy_hole
+    ; CHECK: %sval:sreg_32 = S_MOV_B32 0
+    ; CHECK-NEXT: $sgpr0 = COPY %sval
+    ; CHECK-NEXT: $sgpr2 = COPY %sval
+    ; CHECK-NEXT: $sgpr3 = COPY killed %sval
+    ; CHECK-NEXT: SI_RETURN implicit killed $sgpr0_sgpr1_sgpr2_sgpr3
+    %sval:sreg_32 = S_MOV_B32 0
+
+    $sgpr0 = COPY %sval
+    $sgpr2 = COPY %sval
+    $sgpr3 = COPY %sval
+    SI_RETURN implicit $sgpr0_sgpr1_sgpr2_sgpr3
+
+...
+---
+# Check that super register is defined when a pair interrupts the sequence.
+name:            vgpr_copy_pair
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: vgpr_copy_pair
+    ; CHECK: %vval:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    ; CHECK-NEXT: $vgpr0 = COPY %vval
+    ; CHECK-NEXT: $vgpr1 = COPY %vval
+    ; CHECK-NEXT: $vgpr2 = COPY %vval
+    ; CHECK-NEXT: $vgpr3 = COPY killed %vval
+    ; CHECK-NEXT: dead [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1_vgpr2
+    ; CHECK-NEXT: dead [[COPY1:%[0-9]+]]:vgpr_32 = COPY killed $vgpr0_vgpr1_vgpr2_vgpr3
+    %vval:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+
+    $vgpr0 = COPY %vval
+    $vgpr1 = COPY %vval
+    $vgpr2 = COPY %vval
+    $vgpr3 = COPY %vval
+    %0:vgpr_32 = COPY $vgpr1_vgpr2
+    %1:vgpr_32 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+...