[AMDGPU] Avoid put implicit_def into bundle that break reg's liveness

llvm/lib/CodeGen/MachineInstrBundle.cpp

@@ -94,6 +94,22 @@ static DebugLoc getDebugLoc(MachineBasicBlock::instr_iterator FirstMI,
   return DebugLoc();
 }
 
+/// Check if target reg is contained in given lists, which are:
+/// LocalDefsV as given list for virtual regs
+/// LocalDefsP as given list for physical regs, in BitVector[RegUnit] form
+static bool containsReg(SmallSetVector<Register, 32> LocalDefsV,
+                        const BitVector &LocalDefsP, Register Reg,
+                        const TargetRegisterInfo *TRI) {
+  if (Reg.isPhysical()) {
+    for (MCRegUnit Unit : TRI->regunits(Reg.asMCReg()))
+      if (!LocalDefsP[Unit])
+        return false;
+
+    return true;
+  }
+  return LocalDefsV.contains(Reg);
+}
+
 /// finalizeBundle - Finalize a machine instruction bundle which includes
 /// a sequence of instructions starting from FirstMI to LastMI (exclusive).
 /// This routine adds a BUNDLE instruction to represent the bundle, it adds
@@ -115,6 +131,7 @@ void llvm::finalizeBundle(MachineBasicBlock &MBB,
   Bundle.prepend(MIB);
 
   SmallSetVector<Register, 32> LocalDefs;
+  BitVector LocalDefsP(TRI->getNumRegUnits());
   SmallSet<Register, 8> DeadDefSet;
   SmallSet<Register, 16> KilledDefSet;
   SmallSetVector<Register, 8> ExternUses;
@@ -130,7 +147,7 @@ void llvm::finalizeBundle(MachineBasicBlock &MBB,
       if (!Reg)
         continue;
 
-      if (LocalDefs.contains(Reg)) {
+      if (containsReg(LocalDefs, LocalDefsP, Reg, TRI)) {
         MO.setIsInternalRead();
         if (MO.isKill()) {
           // Internal def is now killed.
@@ -165,8 +182,10 @@ void llvm::finalizeBundle(MachineBasicBlock &MBB,
         }
       }
 
-      if (!MO.isDead() && Reg.isPhysical())
-        LocalDefs.insert_range(TRI->subregs(Reg));
+      if (!MO.isDead() && Reg.isPhysical()) {
+        for (MCRegUnit Unit : TRI->regunits(Reg.asMCReg()))
+          LocalDefsP.set(Unit);
+      }
     }
 
     // Set FrameSetup/FrameDestroy for the bundle. If any of the instructions

llvm/test/CodeGen/AArch64/blr-bti-preserves-operands.mir

@@ -8,7 +8,7 @@
 # The arguments to the call must become implicit arguments, because the branch
 # only expects to get 1 explicit operand which is the branch target.
 
-# CHECK:    BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit $sp, implicit $x0, implicit $w1 {
+# CHECK:    BUNDLE implicit-def $lr, implicit-def $sp, implicit $sp, implicit $x0, implicit $w1 {
 # CHECK:      BL @_setjmp, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $x0, implicit $w1, implicit-def dead $lr, implicit $sp, implicit-def $sp
 # CHECK:      HINT 36
 # CHECK:    }

llvm/test/CodeGen/AArch64/expand-blr-rvmarker-pseudo.mir

@@ -43,7 +43,7 @@
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $x0, implicit $sp, implicit $xzr, implicit $fp {
+# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit-def $fp, implicit $x0, implicit $sp, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BLR $x0, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
@@ -66,7 +66,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $sp, implicit $x0, implicit $xzr, implicit $fp {
+# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit-def $fp, implicit $sp, implicit $x0, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BL @foo, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $x0, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      $fp = ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
@@ -86,7 +86,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def $x0, implicit-def $w0, implicit-def $w0_hi, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $sp, implicit $x0, implicit $x1, implicit $x2, implicit $xzr, implicit $fp {
+# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $sp, implicit-def $x0, implicit-def $fp, implicit $sp, implicit $x0, implicit $x1, implicit $x2, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BL @foo, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $x0, implicit $x1, implicit $x2, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def $x0
 # CHECK-NEXT:      $fp = ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
@@ -106,7 +106,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $sp, implicit $w0, implicit $w1, implicit $xzr, implicit $fp {
+# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit-def $fp, implicit $sp, implicit $w0, implicit $w1, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BL @foo, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $w0, implicit $w1, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      $fp = ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
@@ -127,7 +127,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $x8, implicit $sp, implicit $w0, implicit $w1, implicit $xzr, implicit $fp {
+# CHECK-NEXT:     BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit-def $fp, implicit $x8, implicit $sp, implicit $w0, implicit $w1, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BLR $x8, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $w0, implicit $w1, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      $fp = ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
@@ -149,7 +149,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit-def $fp, implicit-def $w29, implicit-def $w29_hi, implicit $sp, implicit undef $x0, implicit $xzr, implicit $fp {
+# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit-def $fp, implicit $sp, implicit undef $x0, implicit $xzr, implicit $fp {
 # CHECK-NEXT:      BL @foo, csr_darwin_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit undef $x0, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      $fp = ORRXrs $xzr, $fp, 0
 # CHECK-NEXT:      BL @objc_retainAutoreleasedReturnValue, implicit-def $lr, implicit internal $sp
@@ -170,7 +170,7 @@ body:             |
 # CHECK:       bb.0:
 # CHECK-NEXT: liveins:
 # CHECK-NEXT: {{  $}}
-# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $w30, implicit-def $w30_hi, implicit-def $sp, implicit-def $wsp, implicit-def $wsp_hi, implicit-def dead $x0, implicit $x0, implicit $sp {
+# CHECK-NEXT:    BUNDLE implicit-def $lr, implicit-def $sp, implicit-def dead $x0, implicit $x0, implicit $sp {
 # CHECK-NEXT:      BLR $x0, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit-def dead $lr, implicit $sp, implicit-def $sp, implicit-def dead $x0
 # CHECK-NEXT:      BL @attachedcall, implicit-def $lr, implicit internal $sp
 # CHECK-NEXT:   }

llvm/test/CodeGen/AArch64/sve-intrinsics-int-arith-merging.mir

@@ -28,7 +28,7 @@ body:             |
     ; CHECK-LABEL: name: bic_i16_zero
     ; CHECK: liveins: $p0, $z0
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit-def $q0, implicit-def $d0, implicit-def $s0, implicit-def $h0, implicit-def $b0, implicit-def $b0_hi, implicit-def $h0_hi, implicit-def $s0_hi, implicit-def $d0_hi, implicit-def $q0_hi, implicit killed $p0, implicit $z0 {
+    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit killed $p0, implicit $z0 {
     ; CHECK-NEXT:   $z0 = MOVPRFX_ZPzZ_H $p0, $z0
     ; CHECK-NEXT:   $z0 = LSL_ZPmI_H killed renamable $p0, internal $z0, 0
     ; CHECK-NEXT:   $z0 = BIC_ZPmZ_H killed renamable $p0, internal killed $z0, internal killed renamable $z0

llvm/test/CodeGen/AArch64/sve-intrinsics-int-binaryComm-merging.mir

@@ -28,7 +28,7 @@ body:             |
     ; CHECK-LABEL: name: fmul_float_zero
     ; CHECK: liveins: $p0, $z0
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit-def $q0, implicit-def $d0, implicit-def $s0, implicit-def $h0, implicit-def $b0, implicit-def $b0_hi, implicit-def $h0_hi, implicit-def $s0_hi, implicit-def $d0_hi, implicit-def $q0_hi, implicit $p0, implicit $z0 {
+    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit $p0, implicit $z0 {
     ; CHECK-NEXT:   $z0 = MOVPRFX_ZPzZ_S $p0, $z0
     ; CHECK-NEXT:   $z0 = LSL_ZPmI_S renamable $p0, internal $z0, 0
     ; CHECK-NEXT:   $z0 = FMUL_ZPmZ_S renamable $p0, internal killed $z0, internal killed renamable $z0

llvm/test/CodeGen/AArch64/sve-intrinsics-int-binaryCommWithRev-merging.mir

@@ -30,7 +30,7 @@ body:             |
     ; CHECK-LABEL: name: fsub_s_zero
     ; CHECK: liveins: $p0, $z0
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit-def $q0, implicit-def $d0, implicit-def $s0, implicit-def $h0, implicit-def $b0, implicit-def $b0_hi, implicit-def $h0_hi, implicit-def $s0_hi, implicit-def $d0_hi, implicit-def $q0_hi, implicit $p0, implicit $z0 {
+    ; CHECK-NEXT: BUNDLE implicit-def $z0, implicit $p0, implicit $z0 {
     ; CHECK-NEXT:   $z0 = MOVPRFX_ZPzZ_S $p0, $z0
     ; CHECK-NEXT:   $z0 = LSL_ZPmI_S renamable $p0, internal $z0, 0
     ; CHECK-NEXT:   $z0 = FSUBR_ZPmZ_S renamable $p0, internal killed $z0, internal killed renamable $z0

llvm/test/CodeGen/AMDGPU/bundle-breaks-phy-liveness.mir

@@ -0,0 +1,36 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -run-pass=si-insert-hard-clauses -verify-machineinstrs %s -o - | FileCheck -check-prefix=GCN-CLAUSE %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -run-pass=si-post-ra-bundler -verify-machineinstrs %s -o - | FileCheck -check-prefix=GCN-BUNDLE %s
+
+---
+name: clause_implicit_def
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $sgpr32
+
+    ; GCN-CLAUSE-LABEL: name: clause_implicit_def
+    ; GCN-CLAUSE: liveins: $vgpr0, $sgpr32
+    ; GCN-CLAUSE-NEXT: {{  $}}
+    ; GCN-CLAUSE-NEXT: BUNDLE implicit-def $vgpr2, implicit-def $vgpr3, implicit $vgpr0, implicit $sgpr32, implicit $exec, implicit $flat_scr {
+    ; GCN-CLAUSE-NEXT:   S_CLAUSE 1
+    ; GCN-CLAUSE-NEXT:   SCRATCH_STORE_DWORD_SADDR $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr
+    ; GCN-CLAUSE-NEXT:   $vgpr2 = IMPLICIT_DEF
+    ; GCN-CLAUSE-NEXT:   $vgpr3 = IMPLICIT_DEF
+    ; GCN-CLAUSE-NEXT:   SCRATCH_STORE_DWORDX2_SADDR internal $vgpr2_vgpr3, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr
+    ; GCN-CLAUSE-NEXT: }
+    ;
+    ; GCN-BUNDLE-LABEL: name: clause_implicit_def
+    ; GCN-BUNDLE: liveins: $vgpr0, $sgpr32
+    ; GCN-BUNDLE-NEXT: {{  $}}
+    ; GCN-BUNDLE-NEXT: BUNDLE implicit-def $vgpr2, implicit-def $vgpr3, implicit $vgpr0, implicit $sgpr32, implicit $exec, implicit $flat_scr {
+    ; GCN-BUNDLE-NEXT:   SCRATCH_STORE_DWORD_SADDR $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr
+    ; GCN-BUNDLE-NEXT:   $vgpr2 = IMPLICIT_DEF
+    ; GCN-BUNDLE-NEXT:   $vgpr3 = IMPLICIT_DEF
+    ; GCN-BUNDLE-NEXT:   SCRATCH_STORE_DWORDX2_SADDR internal $vgpr2_vgpr3, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr
+    ; GCN-BUNDLE-NEXT: }
+    SCRATCH_STORE_DWORD_SADDR $vgpr0, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr
+    $vgpr2 = IMPLICIT_DEF
+    $vgpr3 = IMPLICIT_DEF
+    SCRATCH_STORE_DWORDX2_SADDR $vgpr2_vgpr3, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr
+...

llvm/test/CodeGen/AMDGPU/finalizebundle.mir

@@ -9,7 +9,7 @@ body: |
     ; CHECK-LABEL: name: test_overlap
     ; CHECK: liveins: $vgpr0_vgpr1
     ; CHECK-NEXT: {{  $}}
-    ; CHECK-NEXT: BUNDLE implicit-def $vgpr2_vgpr3, implicit-def $vgpr2, implicit-def $vgpr2_lo16, implicit-def $vgpr2_hi16, implicit-def $vgpr3, implicit-def $vgpr3_lo16, implicit-def $vgpr3_hi16, implicit-def $vgpr3_vgpr4, implicit-def $vgpr4, implicit-def $vgpr4_lo16, implicit-def $vgpr4_hi16, implicit $vgpr0_vgpr1, implicit $exec, implicit $vgpr1_vgpr2 {
+    ; CHECK-NEXT: BUNDLE implicit-def $vgpr2_vgpr3, implicit-def $vgpr3_vgpr4, implicit $vgpr0_vgpr1, implicit $exec, implicit $vgpr1_vgpr2 {
     ; CHECK-NEXT:   $vgpr2_vgpr3 = V_LSHLREV_B64_pseudo_e32 1, $vgpr0_vgpr1, implicit $exec
     ; CHECK-NEXT:   $vgpr3_vgpr4 = V_LSHLREV_B64_pseudo_e32 1, $vgpr1_vgpr2, implicit $exec
     ; CHECK-NEXT: }