RenameIndependentSubregs: try to only implicit def used subregs

llvm/lib/CodeGen/RenameIndependentSubregs.cpp

@@ -306,6 +306,7 @@ void RenameIndependentSubregs::computeMainRangesFixFlags(
     const IntEqClasses &Classes,
     const SmallVectorImpl<SubRangeInfo> &SubRangeInfos,
     const SmallVectorImpl<LiveInterval*> &Intervals) const {
+  const TargetRegisterInfo &TRI = *MRI->getTargetRegisterInfo();
   BumpPtrAllocator &Allocator = LIS->getVNInfoAllocator();
   const SlotIndexes &Indexes = *LIS->getSlotIndexes();
   for (size_t I = 0, E = Intervals.size(); I < E; ++I) {
@@ -314,6 +315,25 @@ void RenameIndependentSubregs::computeMainRangesFixFlags(
 
     LI.removeEmptySubRanges();
 
+    // Try to establish a single subregister which covers all uses.
+    // Note: this is assuming the selected subregister will only be
+    // used for fixing up live intervals issues created by this pass.
+    LaneBitmask RegMask = MRI->getMaxLaneMaskForVReg(Reg);
+    LaneBitmask UsedMask = LaneBitmask::getNone();
+    for (LiveInterval::SubRange &SR : LI.subranges())
+      UsedMask |= SR.LaneMask;
+    SmallVector<unsigned> SubRegIdxs;
+    unsigned Flags = 0;
+    unsigned SubReg = 0;
+    if (TRI.getCoveringSubRegIndexes(MRI->getRegClass(Reg), UsedMask,
+                                     SubRegIdxs) &&
+        SubRegIdxs.size() == 1) {
+      SubReg = SubRegIdxs.front();
+      RegMask = UsedMask;
+      Flags = RegState::Undef;
+    }
+    LaneBitmask UnusedMask = RegMask & ~UsedMask;
+
     // There must be a def (or live-in) before every use. Splitting vregs may
     // violate this principle as the splitted vreg may not have a definition on
     // every path. Fix this by creating IMPLICIT_DEF instruction as necessary.
@@ -336,19 +356,18 @@ void RenameIndependentSubregs::computeMainRangesFixFlags(
           MachineBasicBlock::iterator InsertPos =
             llvm::findPHICopyInsertPoint(PredMBB, &MBB, Reg);
           const MCInstrDesc &MCDesc = TII->get(TargetOpcode::IMPLICIT_DEF);
-          MachineInstrBuilder ImpDef = BuildMI(*PredMBB, InsertPos,
-                                               DebugLoc(), MCDesc, Reg);
+          MachineInstrBuilder ImpDef =
+              BuildMI(*PredMBB, InsertPos, DebugLoc(), MCDesc)
+                  .addDef(Reg, Flags, SubReg);
           SlotIndex DefIdx = LIS->InsertMachineInstrInMaps(*ImpDef);
           SlotIndex RegDefIdx = DefIdx.getRegSlot();
-          LaneBitmask Mask = MRI->getMaxLaneMaskForVReg(Reg);
           for (LiveInterval::SubRange &SR : LI.subranges()) {
-            Mask = Mask & ~SR.LaneMask;
             VNInfo *SRVNI = SR.getNextValue(RegDefIdx, Allocator);
             SR.addSegment(LiveRange::Segment(RegDefIdx, PredEnd, SRVNI));
           }
-
-          if (!Mask.none()) {
-            LiveInterval::SubRange *SR = LI.createSubRange(Allocator, Mask);
+          if (!UnusedMask.none()) {
+            LiveInterval::SubRange *SR =
+                LI.createSubRange(Allocator, UnusedMask);
             SR->createDeadDef(RegDefIdx, Allocator);
           }
         }

llvm/test/CodeGen/AMDGPU/collapse-endcf.ll

@@ -981,7 +981,7 @@ define void @scc_liveness(i32 %arg) local_unnamed_addr #0 {
 ; GCN-NEXT:    v_cmp_gt_i32_e32 vcc, s4, v0
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
 ; GCN-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-NEXT:    v_mov_b32_e32 v0, 0
 ; GCN-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-NEXT:    s_branch .LBB5_3
 ; GCN-NEXT:  .LBB5_1: ; %Flow
@@ -1004,36 +1004,36 @@ define void @scc_liveness(i32 %arg) local_unnamed_addr #0 {
 ; GCN-NEXT:  ; %bb.4: ; %bb2
 ; GCN-NEXT:    ; in Loop: Header=BB5_3 Depth=1
 ; GCN-NEXT:    s_or_b64 exec, exec, s[6:7]
-; GCN-NEXT:    v_mov_b32_e32 v8, v7
-; GCN-NEXT:    v_mov_b32_e32 v2, v7
-; GCN-NEXT:    v_mov_b32_e32 v6, v7
+; GCN-NEXT:    v_mov_b32_e32 v1, v0
+; GCN-NEXT:    v_mov_b32_e32 v2, v0
+; GCN-NEXT:    v_mov_b32_e32 v3, v0
 ; GCN-NEXT:    s_and_saveexec_b64 s[10:11], s[4:5]
 ; GCN-NEXT:    s_cbranch_execz .LBB5_2
 ; GCN-NEXT:  ; %bb.5: ; %bb4
 ; GCN-NEXT:    ; in Loop: Header=BB5_3 Depth=1
-; GCN-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
-; GCN-NEXT:    v_mov_b32_e32 v8, v7
-; GCN-NEXT:    v_mov_b32_e32 v2, v7
-; GCN-NEXT:    v_mov_b32_e32 v6, v7
+; GCN-NEXT:    buffer_load_dword v3, v0, s[0:3], 0 offen
+; GCN-NEXT:    v_mov_b32_e32 v1, v0
+; GCN-NEXT:    v_mov_b32_e32 v2, v0
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_cmp_gt_f32_e64 s[6:7], 0, v0
+; GCN-NEXT:    v_cmp_gt_f32_e64 s[6:7], 0, v3
+; GCN-NEXT:    v_mov_b32_e32 v3, v0
 ; GCN-NEXT:    s_and_saveexec_b64 s[12:13], s[6:7]
 ; GCN-NEXT:    s_cbranch_execz .LBB5_1
 ; GCN-NEXT:  ; %bb.6: ; %bb8
 ; GCN-NEXT:    ; in Loop: Header=BB5_3 Depth=1
-; GCN-NEXT:    v_mov_b32_e32 v8, v7
-; GCN-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
-; GCN-NEXT:    ; implicit-def: $vgpr3_vgpr4_vgpr5_vgpr6
+; GCN-NEXT:    v_mov_b32_e32 v1, v0
+; GCN-NEXT:    ; implicit-def: $vgpr2
+; GCN-NEXT:    ; implicit-def: $vgpr3
 ; GCN-NEXT:    s_branch .LBB5_1
 ; GCN-NEXT:  .LBB5_7: ; %bb12
 ; GCN-NEXT:    s_or_b64 exec, exec, s[8:9]
-; GCN-NEXT:    buffer_store_dword v6, v0, s[0:3], 0 offen
+; GCN-NEXT:    buffer_store_dword v3, v0, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    buffer_store_dword v2, v0, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    buffer_store_dword v8, v0, s[0:3], 0 offen
+; GCN-NEXT:    buffer_store_dword v1, v0, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    buffer_store_dword v7, v0, s[0:3], 0 offen
+; GCN-NEXT:    buffer_store_dword v0, v0, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;